THE TOPOGRAPHY OF HEMISPATIAL NEGLECT: BRAIN-BEHAVIOUR CORRELATIONS WTH CT AND SPECT IMAGING M STROKE by Farrell S. Leibovitch A thesis submiaed in confomity with the requirements for the degree of Master of Science Graduate Department of Institute of Medical Science University of Toronto @ Copyright by Farrell Stuart Leibovitch 1996
185
Embed
BRAIN-BEHAVIOUR CORRELATIONS WTH CT AND SPECT STROKE€¦ · BRAIN-BEHAVIOUR CORRELATIONS WTH CT AND SPECT IMAGING M STROKE by ... Mary Pat McAndrews, my interna1 appraiser, for her
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
THE TOPOGRAPHY OF HEMISPATIAL NEGLECT:
BRAIN-BEHAVIOUR CORRELATIONS
WTH CT AND SPECT IMAGING M STROKE
by
Farrell S. Leibovitch
A thesis submiaed in confomity with the requirements for the degree of Master of Science
Graduate Department of Institute of Medical Science University of Toronto
@ Copyright by Farrell Stuart Leibovitch 1996
National Library 191 of Canada Bibfiotheque nationaie du Canada
Acquisitions and Acquisitions et Bibliographie Services services bibliographiques
The author has granted a non- exclusive licence allowing the National Library of Canada to reproduce, loan, distribute or seii copies of this thesis in microform, paper or electronic formats.
L'auteur a accordé une Licence non exclusive permettant à la Bibliothèque nationale du Canada de reproduire, prêter, distribuer ou vendre des copies de cette thèse sous la forme de microfiche/film, de reproduction sur papier ou sur format électronique.
The author retains ownership of the L'auteur conserve la propriété du copyright in this thesis. Neither the droit d'auteur qui protège cette thèse. thesis nor substantial extracts fkom it Ni la thèse ni des extraits substantiels may be printed or otherwise de celle-ci ne doivent être imprimés reproduced without the author's ou autrement reproduits saas son permission. autorisation.
Topography of Hcmispatid Ncglca
ABSTRACT
THE TOPOGRAPHY OF EIEMISPATIAL NEGLECT:
BRAIN-BEHAVIOUR CORRELATIONS WTH CT AND SPECT IMACLNG IN STROKE
Farrell S. Leibovitch Master of Science
1996 Institute of Medical Science
University of Toronto
Hemispatial neglect, characterized as failure to attend to contralesional space, is
hypothesized to result tiom damage to a network for duected attention which involves the
fiontal, parietal, and cingulate cortices, and the basal ganglia and thalamus. This study identified
the neural correlates of hemispatial neglect in 75 LHD and 120 RHD acute stroke patients using
structural (CT) and fimctional (SPECT) imaging. Multiple Linear Regression and Partial Least
Squares were used to identify the brain regions that predicted performance on the Sunnybrook
Neglect Banery. In LHD patients, the significant regions were the cingulate, frontal, parietal,
laterat occipital, and temporal regions. In RHD patients, the significant regions were the
cingulate, parietal, lateral occipital, and parietotemporal regions. Overall, the parietal region
emerged as the powerful predictor of neglect behaviour. A qualitative difference emerged
between the hemispheres on fùrther inspection of negiect subtest and brain region correlations.
The study shows the value of complementary structural and fùnctional imaging techniques and
neuropsychological tests of behaviour in elucidating brain-behaviour relationships.
Keywords: Hemispatial Neglect, Lefi-Sided Neglect, Right-Sided Neglecq Siioke, Left Hemisphere- Damaged, Right Hemisphere-Damaged, Theoretical Network for Directed Attention, Single Photon Emission Computed Tomography, Computed Tomography, Multiple Linear Regression, Partial Least Squares
Topography of Hanispatial Neglect
1 am very gratefül to many individuals whose help and fnendship has made this thesis possible.
1 want to thank my thesis cornmittee for their suggestions, support, and devoted efforts, specifically: Dr. Sandra Black. my supervisor, has been a great source of motivation both academically and personally. Her fkiendship and professional support has guided me throughout this project. 1 will be forever grateful to Sandy for encouraging, challenging and always having confidence in me. It has been a privilege and pleasure to know her and work for her. Many thanks to Dr. Curtis Caldwell, my other supervisor, for al1 his assistance during this project, especially regarding SPECT. 1 always benefited fiom his suggestions, challenges, and encouragement. My thesis has been greatly improved fiom the advice of Dr. A. Randy McIntosh, especially regarding PLS. To Dr. John Szalai who provided ongoing statistical guidance and Dr. John Wherrett who gave a clinical perspective and gave many helpfûl suggestions.
1 want to acknowledge my examination cornmittee for their careful reading and detailed recommendations, specifically: Dr. Frank Prato, my external appraiser, for al1 of his comments, both in the written appraisal and during the defense. The final version of this thesis has benefited greatly fiom his suggestions following his in-depth examination. Dr. Mary Pat McAndrews, my interna1 appraiser, for her helpful suggestions and recommendations during the defense and in her written report. Drs. Mary Lou Smith and Gordon Winocur for their insightfiil suggestions.
1 am also grateful to Cognitive Neurology Unit staff, who shared their experiences and expertise and provided constant support, guidance, and camaraderie. Special thanks to Patricia Ebert and Kira Barbour for helping with the CT data, Karen Ma for helping with the SPECT data, Joanne Lawrence, Nancy Blair, Jay Bondar, and Doug Martin for collecting the Behavioural data. 1 owe much gratitude to Dr. Lisa Ehrlich and the technicians of Nuclear Medicine for their willing cooperation.
This work was fùnded by the Ontario Mental Health Foundation and the Heart and Stroke Foundation.
Finally, 1 want to thank my wife, Kem Leibovitch, for her constant encouragement, patience and confidence in me and to the rest of my family, especially my parents and my extended family, for their ongoing support and words of wisdom.
TABLE OF CONTENTS
ABSTRACT - . * . ~ w . * - - - - ~ . m m n . m r t . u . - . * ~ - r . r - u . - . - . . - u u * - * * * . * = . * m . m U
ACKNOWLEDGEMENTS --.- ~ ~ . ~ , . ~ . ~ o . . . . ~ . m H I . . H . H . . W m ~ ~ . t ~ ~ ~ ~ . . . ~ ~ . o ~ . ~ ~ . ~ ~ H w n iÜ
LIST OF TABLES .-W.---.-.UI.UIU.nm.n-n... ""UI..tM..C..n-.um*.*..........----*.*-..*..*.." )Iji:
LIST OF FIGURES .,.,.,......~o.........oH..~~ ........ ~..*....*~**..~*.HHH...o.*..-..**...*~.~............. . .......... VU
LIST OF ABBREVIATIONS ~.........ltnuw. .... w..*o..H1...*.w.*.. ~.**..*.-.*~1,.*.~.-.-.*.*.*......-.-~.......*...-....... i~
LIST OF BRAIN REGION ABBREVIATIONS~lt~~~~nw~~~u~~*mw~~u-*~~m~~~**u~*u*~~~~~~~~***~u~o*u=*w*~~
1.1. NEUROP~YCHOLOGICAL MODELS OF NEGLECT ......................................................................................... 2
4.2.2. RHD Group ..................................................................................................................................... 52
TABLE 6: SPECT PERFUSION RATIO SUMMARY FOR 20 REGIONS OF INIEREST ........................................ 5 1 a
TABLE 7: PERCEWAGE OF PATIENTS W ~ T H CT DAMAGE IN THE THEORETICAL NEfWORK FOR DIRECTED
A ~ E N T I O N .................................................................................................................................................. 5 1 b
TABLE 8: AVERAGE SPECT RATIOS OFTHE REGIONS IN TWE THEORETICAL NETWORK FOR D I R E ~ D
AITEXTION .................................................................................................................................................. 5 1 b
TABLE 9: PERCENTAGE OF PATIENTS WlTH DAMAGE TO THE FIVE REGIONS IN THE THEORETICAL NETWORK
FOR DIRECTED A'ITEMION .......................................................................................................................... 53a
TABLE 1 O: POWER CALCULATION SUMMARY FOR MLR ANALYSES ......................................................... 60a
TABLE 1 1 : RESULTS SUMMARY FROM PLS ANALYSES .............................................................................. 6 1 a
TABLE 12: SUMMARY OF THE RESULTS OBTAINED FROM THE MLR AND PLS ANALYSES ........................ 69a
vii
LIST OF FIGURES
........................................................................................ FIGURE 1 : DIAGRAM OF PARTIAL LEAST SQUARES 17a
FIGURE 2: COMPLETE NEGLECT BAT~ERY SCORE VS. COMPOSITE SCORE USING DRAWING ESTIMATE ....-..4la
FIGURE 3: COMPLETE NEGLECT BA-ITERY SCORE VS. COMPOSITE SCORE USCNG LINE BISECTION
ESTIMATE ............................. ........................................................................................................ -...--..----4 la
FIGURE 4: COMPLETE NEGLECT BA'ITERY SCORE VS. COMPOSITE SCORE USiNG LINE CANCELLAT~ON
..................................................................................................................................................... EST~MATE 4 1 a
FIGURE 5: COMPLETE NEGLECT B A ~ E R Y SCORE VS. COMPOS~TE SCORE USCNG SHAPE CANCELLATION
................................. FIGURE 6: SINGULAR IMAGE FOR THE FIRST LATENT VARIABLE IN THE LHD GROUP 62a
........................... FIGURE 7: SINGULAR IMAGE FOR THE SECOND LATENT VARIABLE m THE LHD GROUP -.63a
FIGURE 8: SINGULAR IMAGE FOR THE T HIRD LATENT VARIABLE IN THE LHD GROUP ................................ 63b
FIGURE 9: MAGE VS. SUBTEST SCORES FOR LV 1 iN THE LHD GROUP ........................................................ 64a
FIGURE 10: IMAGE VS. SUBTEST SCORES FOR LV2 IN THE L H D GROUP ...................................................... Wb
FIGGRE 1 1: IMAGE VS. SUBTEST S c O R f 3 FOR LV3 iN THE LHD GROUP ...................................................... 65a
.............................. FIGURE 12: SINGULAR IMAGE FOR THE FIRST LATENT VARIABLE IN THE RHD GROUP -67a
..................................... F I G ~ ~ R E 13: IMAGE VS. SUBTEST SCORES FOR LV1 IN THE RHD GROUP
LIST OF ABBREVIATIONS
mT~-HMPAO AC-PC ANOVA CI CT fMRI FWHM GE LHD LR LV MBq MCA MLR MR MRI OMHF PET PLS RHD ROI SI SLR SM3 SPECT SSCBC SVD TPO VIF VSB
9%-Tec hnetiurn Hexamethyl propyleneiunineoxUne Anterior Commissure - Posterior Commissure Analysis of Variance Confidence Interval Computed Tomography Functionai Magnetic Resonance Imaging Full Widîb Half Maximum General Electric LeA Hemisphere-Damaged Lin- Regression Latent Variable Mega-Becquerel Middle Cerebral Axtery Mu1 tiple Lioear Regression Magnetic Resonance Magnetic Reso~mce Imaging Ontario Mental Health Foundation Positron Ernission Tomography Partial Least Squares Right Hemisphere-Damaged Region of Interest Singular Image Simple Linear Regression Sunnybrook Neglect Battery Single Photon Emission Computed Tomography Summed Squared Cross-Block Correlation Singular Value Decomposition Temporal-Parietai-Occipital Variance Inflation Factor Visual Search Board
LIST OF BRAIN REGION ABBREVIATIONS
ACing AntWM BG CentWM Deep-TPO F F-Inf FLi FLS-Ant FLS-POS~ F-Mid FOF-Aat FOF-POS~ F-Sup IC-Ant IC-Post Lat0 MedO Motor O P P-Inf PostWM P-Sup PT Sensory SM T TH
Anterior Cingulate Anterior White Matter Basal Ganglia Centrai White Matter White Matter deep beneath the Temporal-Parietal Occipital Junction Frontal Cortex Merior Frontal Cortex Inferior Longitudinal Fasciculus Antenor Superïor Longitudinal Fasciculus Posterior Superior Longitudinal Fasciculus Middle Frontal Cortex Antenor Frontal-Occipital Fasciculus Posterior Frontal-Occipital Fasciculus Superior Frontal Cortex Anterior Intemal Capsule Posterior Interna1 Capsule Lateral Occipital Cortex Medial Occipital Cortex Pnmary Motor Strip or Pre-Central Gyms Occipital Cortex Parietal Cortex inferior Parietal Cortex Posterior White Matter Superior Parietal Cortex Parietal-Temporal Cortex Primary Sensory Strip or Post-Central Gyms Sensorimotor Cortex Temporal Cortex Thalamus or Thalamic Nuclei
INTRODUCTION
Hemispatial neglect is a cognitive disorder characterized by a failure to attend to
stimuli in one's personal or extrapersonal space contralateral to the side of brain damage,
when this failure cannot be attributed to either sensory or motor defects (Heilman &
Valenstein, 1979). Nthough it has k e n observed following damage to the left
hemisphere, hemispatial neglect is encountered most fiequently in association with a
lesion in the right hemisphere as a failure of patients to attend to stimuli in the lefi side of
space (Weintraub & Mesulam, 1987). In severe cases, patients may fail to dress the left
side of their body or rnay eat food only from the right side of their food tray. Male
patients may shave only the right side of their face and female patients might fail to put
makeup on the left side of their face. On clinical or experimental tests of neglect (Black,
analyses were performed on a group of patients ( ~ 2 3 2 ) with SNB testing and Visual
Search Board (VSB) (Kimura, l986), within 1 day of each other. VSB testing requires
special apparatus that is not readily available for bedside testing, which was the intent of
this study. The group consisted of both right and left hemisphere-darnaged stroke patients
with and without neglect. Al1 subtests were significantly correlated with the total neglect
score (r-0.8, p<0.001) and with each other ( ~ 4 . 6 , p<0.001), thus demonstrating intemal
consistency within the battery for each individual subtest. Factor analysis was used to
assess redundancy of the subtests withh the neglect battery. .Ml four subtests conmbuted
to a single factor (eigenvalue = 2.78), accounting for 69.4% of the variance. Each subtest
was positively correlated with that factor, indicating that al1 four subtests were needed to
capture the visuoconstnictive neglect phenomenon. To assess external content validity,
the neglect battery, based on weighted scores, was compareci to performance on the VSB.
Chi-square test of independence showed that subject group identity, according to the
SN%, was in agreement with that on the VSB @<0.001). Shape cancellation was the most
sensitive subtest (76%) and drawkopy was the most specific (99%) (see Appendix A.6
for a complete listing). Logistic regression of the 4 subtests against the VSB was highly
significant (pc0.00 1 ) providing M e r evidence of the vaiidity of the neglect battesr and
i ts subtests against an external standard.
Stroke patients underwent CT scanning of their head usually within 48 hours post-
stroke. Since it is known that a lesion may not appear on a CT scan if it is done too early
post-stroke, patients with an initial negative scan (approximately 30% of patients) had a
repeat scan perfomed at a later date. Whichever scan maximally represented the lesion
was used as the CT scan for al1 data purposes (n=211, mean 7.6k16.4 days pst-stroke,
95% CI [5.4,9.9], range 0- 154). Scans were perfomed parallel to the orbitomeatal line, a
commonly used reference line which extends h m the canthus of the eye to the extemal
auditory rneatus. Unfortunately, it was not possible throughout the five year study, due to
financial restrictions as well as disk storage limitations, to store and quanti@ the original
digitized CT scans. Thus, for each CT scan, one centimetre thick slices in the traasaxial
plane, approximately 12 in total, were yrinted on X-ray film for fixther analysis and
interpretation.
The films were used to obtain both the structural lesion volumes and anatomical
localizations. To obtain the lesion volumes, the lesions were traced fiom the x-ray film
ont0 paper. With the aid of a digitizing scanner (Sigma ScanM Version 3.0, Jandel
Scientific, Sausalito, California), the are;? corresponding to the lesion for each slice was
digitized and then summed to arrive at a lesion volume for that scan.
Anatomical localization was performed by reference to a stereotactic atlas
(Talairach & Townoux, 1988). The lesion seen on each of the transaxial slices was drawn
on the best matched template from the Talairach-Toumoux atlas. Since each slice fiom
the atlas has an underlying grid, a detailed lesion localization was possible (refer to
Appendix B for examples). From the grid of the atlas, each region that was affected was
detailed on a checklist by region, Brodmann's area and x-y coordinate. The checklist was
entered into a database (Filemaker Prom, Claris Corp., California) on a Macintosh
Cornputer (Apple IISi). This technique allowed for dichotomous categorization, Le., the
particular region was denoted as damaged or not. A problern with this method is that if
the same area, region EZ, is damageci in two different patients (A & B), a '1'
(corresponding to a 'yes') wouid be entered into the &tabase corresponding to the
presence of damage Le., irrespective of its size or depth. Thus, patient A could have 75%
of the entire region R damaged (i.e., of its entire volume damaged) and patient B 25%
damage and yet both would show only a 'yes' response regarding the site of damage.
To allow an estimation of the degree of damage in each region, a quantification
procedure was devised, based on the number of transaxial slices in which that region
appeared in the Talairach-Toumoux atlas (1988). Specifically, the number of slices on
which the region of interest appears was first determined (maximum of 12 slices). For
example, the inferior parietal lobe is found on four slices (slice 3, 4, 5, 6; Appendix B.3).
A ratio of structural damage was calcutated to be the number of slices with damage to a
region divided by the number of slices that the region of interest appeared. In the case of
the inferior parietal lobe, for example, the ratio was taken over the 4 slices. Patient A
fiom above, for example, would show 3 of 4 slices damageci whereas patient B had only 1
of 4 slices damaged. The resultant variables would be coded as 0.75 for patient A and
0.25 for patient B, instead of 1 .O for both. In this way, the degree of region involvement
could be quantifieci on a =aie h m O to 1 for each region of interest. This quantification
approach allowed for a measure of the vertical depth of damage in a region.
Thkteen CT regions were used in subsequent analyses (please refer to Appendix
B.3 for stereotactic breakdown of regions), al1 of which came h m the ipsilesional
hemisphere (single lesions). There were 7 cortical regions, including the anterior
cinguiate cortex (Acing), the fiontai cortex 0, the parietal cortex (P), the temp~ral
cortex (T), the lateral occipital cortex (LatO), the primary motor strip (Motor), and the
primary sensory strip (Sensory). Since the frontal and parietal regions were of specific
interest in this study, each of these regions was M e r subdivided into smaller more
spec i fic anatomical subdivisions which were us ed in speci fic pst-hoc analyses. The
fiontal region was subdivided into the inferior frontal gyrus (F-id), the middle frontal
gyms @-Mid), the superior frontal gyrus (F-Sup), and the parktal region was subdivided
into the inferior parietal cortex (P-Inf) and the superior parietal cortex (P-Sup).
There were 2 subcortical nuclei, specifically the basal ganglia (BG) and the
thalarnic nuclei (TH). White matter regions comprised the remaining 4 CT regions
inc luding anterior white matter (AntWM), central white matter (CentWM), posterior
white matter (PostWM), and a subdivision in the PostWM deep beneath the parietal-
temporal-occipital junction (Deep-TPO). Each region was a composite of the white
matter tracts enclosed in the corresponding area defined by the Talairach-Tournoux
(1988) atlas. Damage to the AntWM region was calculated by averaging the amount of
damage in the following regïons: the antenor superior longitudinal fasciculus (FLS-Ant),
the anterior frontal-occipital fasciculus (FOF-Ant), the anterior intemal capsule (IC-Am),
and the anterior centmm semiovale (CS-Ant). Similady, the PostWM was a composite of
the posterior portion of the above regions (FLS-Post, FOF-Post, IC-Post, and CS-Post),
and also included the idenor longitudinal fasciculus (FLi). For certain pst-hoc analyses,
each of the above white matter subdivisions were used to refine matornical localization.
2.3. SPECT Scans
At the time this study was conducted stroke patients undenvent SPECT scanning
of the head as part of their clinical assessment. Two-hundred and twenty-one patients
were Maged on a GE single head gamma camera following injection of 740 MBq of
9 9 M ~ c - ~ ~ ~ ~ ~ in the Medical irnaging Department at Sunnybrook Health Science
Centre (see Appendix C.1 for an example of a SPECT scan). Although there was no
specific patient preparation prior to or during injection, patients were generally seated,
had their eyes open, in a quiet environment with normal iighting. Scans were acquired in
the first two weeks (n454, mean 7.Sf 15.2 days pst-stroke, 95% CI [S. 1, 9.91, range O-
187), and repeated when appropriate as part of the standard clinical protocol at two to
three weeks (n=84, mean l7 .8S5.5 days pst-stroke, 95% CI [12.3, 23.31, range 8-167).
For research purposes, scans were repeated at thirteen months (n=54, 406.W79.3 days
post-stroke, 95% CI [384.4,427.7], range 156-766) in a subset of consenthg survivors. In
addition, nineteen SPECT scans were performed on normal volunteers.
SPECT scans were acquired on a GE Mode1 400 AT single head gamma camera
with patients in the supine position (imaging tirne approximately 30 minutes). Using step
and shoot mode, 64 planar views were acquired over 360 degrees, with a 64 pixel x 64
pixel acquisition frame per view (25 seconds per view) and a magnification factor of
1.33. Following acquisition, each SPECT scan was reconstnrctd to correct for any head
tilt and to align each brain so it was parallel to the orôitomeatal (OM) line in the
transaxial plane, as delheated on the mid-sagittal slice. The reconstnic tion procedure
(ramp and Buttenvorth filter with a power factor of 15 and a cut-off fkquency of 0.4cm-',
attenuation correction p=û. 12cnf1 (Sorenson, 1974)) took approximately twenty minutes
per SPECT scan. A correction for non-linearity was applied (Lassen, Anderson, Friberg,
& Paulson, 1988). Recumtnicted image spatial resolution was approximately 1.2 cm full
width at half maximum (FWHM). Each brain was realigned in the coronal, sagittal, and
transaxial planes during the reconstruction procedure in order to correct for head tilt and
standardize individual brains to a set of standardized axes (Appendix C.2). Thus when
viewing the brain from the transaxial plane, ail slices were aligneci in the sarne
orientation.
Since each 0.96 cm thick transaxial slice (pixel size = 0.48cm x 0.48cm) may not
contain the sarne brain regions across subjects, depending on brain size, a linear scaling
technique was applied. In this technique, the length of each of the three major axes was
determined for each brain (lefi side-right side "x-axis", anterior-posterior "y-axis", dorsal-
ventral "2-ariis"). Prior to cdculating the axes lengths and in order to decrease noise in
the image, the image was tfvesholded and converted to binary format. The first step
involved padding the image with zems on al1 sides, for use with the erosion procedure
described later. Next, the median value of d l pixels in the image that were greater than
7% of the maximum value was calculated. A threshold was applied such that d l pixels
with a value p a t e r than 55% of the median were given a value of 1 while al1 pixels less
than or equal to 55% of the median were assigneci a value of O, thus converting the image
to binary. Finally, in order to make the edges of the brain more continuous and fil1 in any
holes that may have been caused by a stroke, which would lead to incorrect size estimates
of the axes, the image was dilated (24~24x3 pixel kernel) and eroded (25x25~3 pixel
kernel) (Russ, 1992). The dilation procedure set any outer background pixels to 1, that
touched an inner brain pixel with a value of 1. Conversely, the erosion procedure set any
inner brain pixels to O, that bordered on a background pixel with a value of 1. The
technique of using dilation then erosion is referred to as a closing morphological
operation (Russ, 1992). The height of each brain (dorsal-ventrai) was calculated on the
rnid-sagittal plane and the maximum length (anterior-posterior) and width (side-to-side)
were found using the transaxial slices. This procedure, including the threshold
percentages and morphological filter kemel sizes, was developed using a subset of fifty
SPECT images.
in addition, the image midline was found using a Stochastic Sign Change
algorithm (Minoshima, Berger, Lee, & Mintun, 1992). In cases where large asymrnetries
were found between the widths on opposite sides of the midline, probably corresponding
to a large hypoperfused ara, the width of the larger side was mirrored and a width was
subsequently recalculated. Based on the known width, height, and length of the brain,
each of the axes was rescaled to preset lengths ( x 4 8 y=54 pixels 2=12 slices), so that
each brain was compresseci or expanded into a predetermined volume with known axes
(therefore pixel size was altered and propodonal to each of the axes). Following scaling,
the resultant image was centered in the rniddle of the m y (image, or array, size = 64
pixels wide by 64 pixels long by 12 siices high). Once the brain had been rescded,
transaxial slices were able to be matched across subjects with improved accuracy. Each
slice could now also be more accurately rnatched to a stereotactic atlas for anatomical
localization.
To analyze the SPECT scans two different methods were employed: a cortical rim
procedure and an autornated region-of-interest (ROI) analysis (Appendices C.3 & C.4).
For the cortical rim analysis, the procedure, adapted from Hellman et al., (1989) and
written in Interactive Data Language (RSI, hc., Boulder Colorado) on a Sunm
workstation (Sun View Mountain, California), involved finding the b e r and outer brain
edges, and then dividing the corresponding rim into equal annular segments for
cornputation of counts. Prior to hding the edge of the brain, counts in the cerebellum
were found using an automatic algorithm, described in more detail in the automated ROI
procedure below. The cortical rim was performed on each of 6 transaxial slices separateiy
(corresponding to slices 2-7 of the 1inearIy scaled brain). To find the outer edge of the
brain for slice 2 the program searched for the first outside pixel above 22% of the counts
in the cerebellum and then zeroed aii pixels outside of that pixel. The thresholds for slices
2-7 were as follows: slice 3(26%), 4(30%), 5(33%), 6(36%), 7(42.2%). The thresholds for
each slice were found by optimizing the resultant image followiag thresholding on a set
of 50 brains. For five patients (out of 221) the threshold values required adjustment to
optimize edge detection in two slices per patient, for a total of 10 slices with adjusted
thresholds. The resultant images on each slice containecl only brain, with the first pixel
above O corresponding to the first pixel of brain.
Using a dilation and erosion procedure similar to the one described above, an 11
pixel thick rim was obtained (using an 1 lx1 lx1 pixel kemel), which corresponded
approximately to 3.5 cm of cortex. An automatic evaluation of the rim was made to see if
any large lefi-right asymmetries existed. The difference in distance behveen any two
pixels on each side (hemisphere) of the cortical rim should not differ by more than a few
pixels. If a large difference existed, possibly as a result of a stroke, the side with the
smaller difference was automaticdly mirrored across the midline on to other side.
Finally, the rim was divided into 24 equal annular segments, 12 per hemisphere.
This technique was used on 6 slices, autornaticaliy selected from the rescaled brain to be
slices 2-7 (out of 12) for a total of 72 segments per hemisphere. In addition, a similar
technique was used on slice 1, the most dorsal slice reliably identifiable as brain.
Segments on this slice were obtained in a different way since it generally contained bain
that was too small to use the previous methoâ. Here, the same outer rim that was found
for slice 2 was overlaid on slice 1 but instead of eroding the rirn, al1 of the brain intenor
to the edge was used, divided into 8 segments. Since this dorsal slice was smaller than the
slices beneath, the number of pixels in each segment remained similar to the lower
segments. In total then there were 152 segments calculated fiom the cortical rim
proc edure.
In order to measure counts in subcortical regions, an additional ROI analysis was
performed. On each of ten slices, preset regions of interest were placed, proportional to
slice size, over anatomical regions of interest including the basal ganglia, thalamus,
antenor cingulate gps, and cerebellum. There were 18 ROIS placed on each hemisphere
for a total of 36 ROI segments. Measurement of counts in the cerebellum was slightly
different than for the other ROIS, in that, initially the cerebellar ROI was placed over the
cerebellar region on slices 1 1 and 12, proportional to the length and width of each slice.
Final placement, however, involved a local search for the maximum pixel value and the
cerebellar ROI was then centered over it.
From both automated procedures described above, there were 188 segments, 94
per hemisphere, in which mean counts, standard deviation of mean cowts and the
number of pixels in each ROI were obtained and used in analyses. Al1 segments
corresponded to matched anatomical regions fiom the reference atlas previously
described. To reduce the data for certain analyses, segments from similar regions (e-g.,
parietal) were grouped and averaged. in total, there were 10 SPECT scan averaged
regions per side including 8 cortical regions as follows: the fiontal cortex (F), the anterior
cinguIate gyms (ACing), the parietal cortex (P) , the parietal-temporal region (PT), the
temporal cortex (Temp), the sensorhotor cortex (SM), the medial occipital region (Med
O), and the occipital cortex (O), and 2 subcortical regions which were the basal ganglia
(BG) and the thalamic nuclei (TH). Since the parietal and frontal cortices were of specific
interest in this study, each of these regions was subdivided into smaller regions (3 fiontal
and 2 parietal subregions per side) for more specific anatomical localization. The fiontal
cortical region was subdivided into the idenor frontal gyrus (F-Inf), the rniddle h n t a l
gyms (F-Mid), the superior fiontal gym (F-Sup) and the parietal region was subdivided
into the inferior parietal cortex (P-Inf) and the superior parietal cortex (P-Sup).
Two patients, who also had Magnetic Resonance Imaging for cli&al
reasons, were used ui an SPECT-MR superposition experiment to ver@ localization of
SPECT regions (Appendix CS). The two patients were imageû on a 1.5 Tesla MR
magnet (Signa, Version 4.7; General Electric Medical Systems, Milwaukee, USA). A
volumetric 3-D sequence was performed in the sagittal plane, covering the entire brain,
that resulted in 1 24 contiguous 1 -3 mm slices in thickness. The scans were acquued in
14.4 minutes using a Tl-weighted sequence, 192 phase-encoding steps, with a T M E of
3Y5 ms, flip angle of 35", and a field of view of 20 cm. m e superposition technique,
developed by Woods et al., (1993) which nins on a S u P workstation (Sun View,
California), compares voxels in the MR image to voxels in the SPECT image. For a
SPECT-MR superposition, where the SPECT scan is superimposed on the MR scan, the
algorithm first divides the MR brain into 256 separate components (nonbrain structures
aiready removed), which differ based on voxel intensity 0). This technique makes the
assumption that voxels with similar inteasity correspond to sirnilar brain tissues. For each
of the 256 voxels intensities in the MR image, the algorithm finds a weighted average of
the normalized standard deviations (a'') for the value of the corresponding voxels in the
SPECT image in the following way. First, the number of voxels (n,) corresponding to
each voxel intensity is tabulated (i.e., nj conesponds to the number of voxels with an MR
voxel intensity ofjl. Next, the mean (a;) and standard deviation (aj') of al1 SPECT voxels
in the same locations corresponding to the M . image are caiculated (prior to this step, the
SPECT image had already been linearly Uiterpolated to be identical in size to the MR
image). Finally, a weighted average of the normalized standard deviations is calculated
Topography of Hanispatial Ncglcct
using the following formula a" = (qïai)*(nh. At the start of the algorithm, it is assumed
that the reslice parameters, correspondhg to the x-, y-and z-axis rotations and translations
needed to register the two images, are set to zero. Following calculation of the weighted
average for each voxeI intensity, the program changes the reslice parameters and
recalculates new weighted averages. The assumption is that smaller weighted averages
correspond to more accurate registration of the two images. The algorithm minimizes a"
by adjusting the reslice parameters and recalculating d' iteratively.
These two superpositioned brains were used to aid anatomical localization. Since
the MR brain was aliped parallel to the anterior commissure-posterior commissure (AC-
PC) line, this ensured that the superimposed SPECT was also aligned with the same
angulation. Two stereotactic atlases (Talairach & Tournoux, 1988; Damasio, 1995) were
used to optirnize anatomical interpretation. By looking at the MR anatomy and the
SPECT overlay, it was possible to identiQ lobar anatomy on the SPECT scan more
precisely. Siace al1 SPECT scans were reconstructed, parallel to the OM line, and linearly
scaled, each scan should be standardized such that when viewing transaxial slices, slice 5
should correspond to very similar brain anatomy across al1 subjects (Appendix C.6). It
was then possible to use the two superimposed SPECT-MR scans as reference atlases for
anatomical matching with the rest of the SPECT scans in our population.
To ven@ that the OM line SPECT reconstruction was beïng estimated correctly,
the rotation between the MR and SPECT in the sagittai-plane was calculated. Using
output coordinates fiom the superposition program, the difference between the SPECT
rotated to the MR (rotated to the AC-PC line) and the SPECT prior to superposition
(estimated parallel to the OM line h m the reconstruction procedure) was found to be
-0.97" (Range -3.1,0.23 degrees) based on five SPECT-MR reconstructions.
ANALY SIS
3.1. Population Inclusion Criteria
In order to characterize the patient populations, prior to any subsequent analyses,
univariate statistics were performed on al1 datasets, including the calculation of means,
standard deviations, and simple correlations. The patients selected for analysis in this
study came fiom a combined population of patients (n=561) from three consecutive,
partly overlapping, prospective studies, the Stroke SPECT study ( ~ 4 6 9 ; funded by the
Heart and Stroke Foundation), the Spatial Attention Study ( ~ 1 7 0 ; fimded by the Ontario
Mental Health Foundation), and the Neglect Study (n=64; also fùnded by the OMHF),
over the course of six years (1988-94). Although some patients were in more than one
study at the same time, they were included only once in this analysis. Each population of
patients to be analyzed, e.g. lefi hemispheredamaged (Lm) patients in a CT analysis,
was compnsed of patients who conformed to a nwnber of inclusion criteria. For al1
anaIyses, al1 patients were right-handed, had at least 20/40 vision generally with
corrective glasses, were able to undergo SNE3 testing, had a single CT-confirmed lesion,
for a total of 297 patients from which to choose. Eighty percent of those patients also had
a SPECT scan. Exclusion criteria included patients who were too il1 or disabled to
undergo testing (loss of 156), bilateral acute stroke (loss of 10), previous brain i n j q
such as an earlier stroke (loss of 67), negative CT scans and no SPECT scan [and
therefore no useable imaging data] (loss of 13), no scans at al1 due to misplacement (loss
of 9, and lefi handedness (loss of 1 3).
ui addition, for al1 analyses, each patient must have had a wglect battery
administered to them within a specified period of time. Al1 patients were tested as soon as
they were able to sit up and undergo the testing procedure. Because neglect tends to
diminish over time rather than increase and because it can dîsappear quickly (Stone,
Patel, Greenwood, & Halligan, 1992), different temporal inclusion criteria were allotted
to patients with and without neglect. The requirement for patients without neglect was
that a complete neglect battery had to be administered to them within 14 days pst-stroke.
Fifty-four patients out of a possible 157 were excluded for these reasons (n=23 for
incomplete battery and n=3 1 for time restriction). For patients with neglect, the timeline
for testing was extended to 120 days pst-stroke; 3/140 patients were excluded because of
this restriction. This was doue in order to maximize inclusion of the patients exhibiting
neglect behaviour. Patients with severe neglect were frequently tw il1 from their stroke to
undergo neglect battery testing in a shorter tirne period. Despite this later testing,
however, they still showed neglect; thus, theoretically, even if their neglect had improved,
compared to performance which rnight have been anticipated had they been testable
earlier, they still demonstrated presence of neglect on the battery. A total of 5 1 additional
patients could thus be included by expanding this time window for inclusion.
To M e r increase the population of patients with neglect, those with partial
scores were also included if the score exceeded the neglect cut off (score L 6) despite the
fact that the battery could not be completed due to the severity of the patient's deficits.
For example, severe constructional apraxia excluded scoring of drawings in 14 patients; a
further 18 patients had difficulty with the visual discrimination of targets from non-
targets required for the shape cancellation task. In order to be able to use the neglect score
fiom patients with a partial score, such as 40/70, a composite score was extrapolated
based on the patient population who completed al1 subtests, as discussed below. For any
rnissing subtest, a score was calculated based on the mean neglect score in the population
of patients with neglect (n=112, mean neglect score = 38) multiplied by the average
proporîional ratio, with the calculated score not to exceed the maximum possible score
for the subtest. This mean ratio was calculated by finding the proportion of a subtest to
the total neglect score for an individual and then finding the mean proportion for the
population. For example, the score on drawings accounted for 17% of the total neglect
score (Le., 38) in the population and thus a missing value of 7 (0.17 x 38) was allotted for
patients for their drawing/copying subtest (n=14). In this way, composite scores for al1
subtests were calculated for those patients with incomplete batteries. Sirnilady patients
with a missing score on the line cancellation task (n=l), line bisection task (n=3), and
shape cancellation task (n=18) were given calculated missing scores of 4 (0.10 x 38), 10
(0.28 x 38), 15 (0.40 x 38), respectively. For al1 patients with incomplete batteries, the
corresponding calculated population-averaged subtest scores (7 for drawings, 4 for line
canceliation, 10 for line bisection, and 15 for shape canceliation) were substituted for
missing subtest scores, in order to calculate an overall neglect battery score. This may
have underestimated their performance, but it allowed a total score to be derived for each
patient (total n=25), for entry into the data analysis. To calculate the error associated with
each of the composite subtest scores, four linear regressions were used to predict the tnie
SNI3 score from the composite SNB score for those with completed SNBs (i.e., the
composite SNB score was tabulated using a composite subtest score (i.e., drawings)
summed with the true scores for the other three SNE! subtests). The correlations for al1
four composite scores were high (r4.94 or higher; Figures 2-5), indicating that the
composite scores were providing a reasonable estimate of neglect performance.
3.2. CT Inclusion CriteRa
From the above inclusion and exclusion criteria, there was a population of 240
potential patients on which CT analysis could be performed. From this popdation, al1
patients with negative CT scans were removed (n=27), since lesion localization could not
be performed. Demographic and descriptive (i-e., visual analysis) data was calculated on
the remaining population of 2 1 3 patients, 83 le A hemisphere-damaged ( L m ) patients (45
patients without neglect and 38 with neglect) and 130 right hemisphere-damaged (RHD)
patients (41 without neglect and 89 with neglect). Due to the fact that lesion volume was
found to be a confounding variable, al1 patients included in the CT linear regression
analyses required a CT Iesion volume measurement, which was not attainable in 18
patients because the original CT scans had been lost. In total, 195 patients were available
for CT analysis, 75 LHD patients (42 patients without neglect and 33 with neglect) and
1 20 REID patients (38 without neglect and 82 with neglect).
In order to address the hypothesis that al1 patients with negIect will show damage
to at least one of the predîcted key matornical regions in the directed attention network,
lesion locaiization data fiom the CT scans were investigated (refer to Table 3 for a
TABLE 3: ANALYSIS SUMMARY
Hypot hcsis Addrcsscd
Part A.
Patients with neglect will
have at least one key
rcgion or the
interconnected white
matter fibre bundles
damaged.
Part B.
The regions that will
emerge in MLR andlor
PLS will be the five
predicted t heoretical
regions.
Variables Examincd
--
a) Neglect Category(y1n) - dependent -
24 CT Regions (dainage yln) - indepcndent
b) Transformed Neglect Score - dependent
24 CT Transformed Regions - independent
c) Transformed Neglect Score - dependent
5 SPECT anatomicnl ratios - independcnt
5 CT Transformed Regions - independent
b) Transformed Neglect Score - dependent
24 CT Transformed Regions - indepcndent
c) Transformed Neglect Score - dependent
16 SPECT anatomical ratios - independent
d) 4 Neglect subtest scores - dependent
160 SPECT segment ratios - indepcndcnt
# of Patients
a) 84 LHD
a) 130 RHD
b) 75 LHD
b) 120 RHD
c ) 59 LHD
c) 89 RHD
b) 75 LHD
b) 120 RHD
c) 59 LHD
c) 89 RHD
d) 44 LHD
d) 68 RHD
Stutistic Employcd
a) Visual Data Inspection
Frequency Calculation
b) Linear regression with transformed
neglect scores & CT regions.
c) Linear regression wi th transformed
neglect scores & CT and SPECT
regions.
b) Linear regression with transfonned
ncglect scores & CT regions.
C) Lincar rcgression with transformed
neglect scores & SPECT regions
d) Partial Least Squares with transformed
subtest scores & SPECT segments
summary of al1 analyses). Al1 patients with neglect and a single lesion were exarnined. If
a patient displayed neglect and had a lesion that did not include the hypothesized regions
of interest, they were M e r anaiyzed anatomically in order to see if any other common
regions emerged. in addition, frequency tables were charted for ail regions, within each
subpopulation (i.e., RHD patients with neglect). For cornparison, lesion localization data
fiom the group of patients who did not show negiect were also investigated.
3.4. SPECT Inclusion Criteria
Since lesion volume was also an issue in the SPECT population, one limiting
factor in the selection of patients was that they must have had a CT scan with
measurement of lesion volume. In addition, due to the dynamic, functional nature of
SPECT, it was considered prudent to match the date of acquisition of the SPECT scan as
closely as possible to the date of administration of the neglect battery. However, the
criteria for inclusion in relation to this difference depended on which manoeuvre came
first. As explained above, neglect performance rarely detenorates in a patient yet it can
recover relatively quickly; in this population, 70% of patients, who initially showed
neglect, had normal performance on the SNI3 by three months. Therefore, if the SPECT
scan was performed prior to the neglect battery and the patient displayed neglect, it is
relatively safe to assume that the patient would have displayed neglect if tested on the
sarne day as the SPECT scan. On the other hand, if the patient was assessed prior to the
SPECT, it can not be assumed that the patient would have had neglect at the time of
SPECT scanning, unless a neglect battery was performed at a date some time after the
SPECT scan and the patient displayed neglect at that time. Based on these assurnptions,
patients without neglect had their date for inclusion restricted to -3 to 12 days Corn the
tirne of their SPECT to the time of their neglect battery (SPECT Date - Neglect Battery
Date). Patients with neglect were Iimited to -120 days to 7 days between the t h e of their
SPECT to the time of their neglect battery. In total then, 147 patients remained in the
SPECT population for analysis, 59 LHD patients (3 1 patients without neglect and 28 with
neglect) and 8 8 RHD patients (3 1 without neglect and 57 with neglect).
3 . Statistical Normalization Procedure
3.5.1. Negfect Score m d Subtest Log Tmnsformation
To investigate the importance of predicted key anatomical regions, linear
regression analyses were planned. In order to capitalize on the full range of neglect
behaviour, it seemed prudent to use the whole score from the neglect battery. However,
on visual inspection the distribution of neglect scores was markedly skewed to the right.
Univariate statistics showed that the distribution had a skewness value of 1.2, a kurtosis
of 0.25, and a significant value of 0.2220 (p<0.001) on the Lilliefors test of normality (a
modification of the Kolmogorov-Smirnov test). The transformation that best improved
the univariate statistics of normality was the log transformation. Since scores on the
neglect battery could have a value of zero and the logarittun of zero is not solvable, a
constant was added to the iog transformation equation, specifically loglo(neglect score +
2.1), which improved normality. The transformation improved skewness to 0.06, and
kurtosis changed to -1 -4. Although the Lilliefors test was still found to be significant
@<0.00 1 ), the value calculateci improved by a factor of two (value = 0.1 OS). Therefore
al1 regression (and other) analyses involving the neglect score were perfoxmed on the log
transformed scores. In addition, the same log transformation was applied to the individual
subtest scores, to be used in the PLS aoalysis. This improved normality in a similar
fashion for the subtest scores.
3.5.2. CT Regional Arc Sine Transformation
The CT data used in the regression analyses was the quantitative structural data,
as described in the methods section. Each anatomical variable was detennined first by
calculating the ratio of slices showing damage in a particular anatornical region to the
number of slices on which the region appeared. For example, a lesion in the inferior
parietal lobe appearing on 2 slices occupied 2 out of 4 possible slices for that region.
Since the measurements calculated are proportions based on different denominators,
regions with a small denominator, such as 2, have linle room to vary compared to a larger
region with a denominator of 10. To decrease this difference in variance potential
between regions, it was necessary to transform the ratios. A commonly applied
transformation to proportional data is the arc sine transformation of the ratio multiplied
by two (2 sin-' ( p ) ) (Winer, 1971). This was calculated for each CT variable and used in
the CT analyses. Transformation is also important since the power associated with finding
a significant difference between proportions based on different denominators decreases
and thereby increases the Type II error in analyses (Stevens, 1986).
3.6. Mufticollineatig in CT and SPECT Data
Prior to regression analyses, a Pearson product-moment correlation matrk was
produced to explore the issue of multicollinearity beniveen anatomical variables. In
addition, a test of rnulticollinearity of the variables enter4 into the regression equation
was aerformed. This was done for both the CT and SPECT regional variables.
For each of the individual regression analyses, a multicolhxuity test was
performed which cornputes both the tolerance of each variable, defineci as (1-~i') where
Ri is the multiple comlation coefficient when the ith independent variable is predicted
fkom the other independent variables, plus a variance inflation factor (VIF) score (which
is the inverse of the tolerance) (Belsley, Kuh, & Welsch, 1980). If the tolerance of a
variable is low (or if the VIF score is high), then its contribution to the regression
equation can probably be explained by a linear combination of the other variables and
thus is highiy collinear with them. Eigenvalues and condition indexes (defined as the
square-root of [the maximum eigenvalue divided by the eigenvalue for the ith
independent variable]) were computed fiom the scaled, uncentered cross-produçts matrix
of the independent variables. Cornparison of these values can also identiQ collinear
variables. For example, eigenvalues with corresponding high condition indexes are
probably the result of collinear variables.
3.7. CT Linear Regression Anaiysis
Two approaches were used in building regression equations. The first type was
exploratory in nature and involved stepwise regression analysis. Thirteen anatomical
regions, detailed earlier, were entered in a stepwise linear regression ushg the log
transfomeci neglect score as the dependent variable, and the arc sine transforrned CT data
as independent variables. Regional significance was set at a conservative F0.004
(Bon ferroni correction for 1 3 regions for a=0.05). The second approac h involved entering
into a mode1 the five regions from the hypothesized network for directed attention. The
analyses were performed on the LHD (n=75) and RHD (n= 120) populations separately
for these analyses. Prior to regression analysis, t-tests were performed and if any variable,
such as age or volume, was found to be significant between the two groups, it was added
as a covariate in the regression analyses. Al1 analyses were perfomed using the statistical
software package SPSS (SPSS inc. 1995).
3.8. SPECT Noma&ztion and Standardization
Prior to SPECT scan analysis, the data for each anatornical segment were
standardized. Mean counts obtained in each segment from the cortical rim and subcorticai
ROI programs were normalized by dividing by the mean counts in the higher of the two
cerebelli, which was the ipsilateral cerebellum in 90% of cases, on an individual basis.
The basis for using the higher cerebellum rather than the ipsilateral cerebellum was that
in a small subset of patients (3% of our population) with occipital damage, the
ipsilesional cerebellum was f o n d to have structural damage. Each segment ratio (X,J
was then standardized by subtracting fkom it a specified mean (x,,) and dividing the
result by the standard deviation of that mean (+p), using the formula xsa - P P .. The
specified mean and standard deviation values were calculated as the mean and standard
deviation of the homologous segment (same hemisphere) from the SPECT scans of the 19
normal volunteers, excluding values that were greater than 2 standard deviations from
that mean (Le., calculahg n o d meam based on the values with 95% codidence).
Stated differently, for each segment a mean and standard deviation were calculated fiom
counts in the normal population corresponding to the segment in the same hemisphere.
From these calculations, standardized values were calculated for each segment. Twenty
larger anatomical regions were obtained by averaging those standardized segments thar
corresponded to the region of interest. To correct for the size of each segment, a weighted
average of al1 segments based on the number of pixels in each segment was used in the
calculation of larger averaged regions.
3.9. CT-SPECT Linear Regression Analysis
Following the CT regression analyses above, two additional regressions were
performed using the both the CT and SPECT regions for the LHD and RHD populations
separately. First, the five hypothesized CT regions were entered into a model, along with
their corresponding covariates, and then stepwise regression was used on the
corresponding SPECT regions to see if any would enter in addition to the structural data.
Second, al1 ten regions were forced into a model to see if the sarne regions that were
significant in the independent analyses also emerged here or if a new combination of
structural and functional regions arose. The reason for this anaiysis was to see the effect
of entering both structural and functional information into the same model, and how both
modalities would predict neglect score.
3 m I U m Power Calculations for Linear Regression Analysis
In order to calculate the probability of failhg to confïrm a significant difference
(Type iI error), pst-hoc power calculations were perfomed for the CT regression
analyses. The power calculations were performed for a medium effect size (Cohen, 1988)
with two sets of independent variables, one for the covariates, R' of 0.25, and the other
for the variables of interest producing an iacremental R~ of . l O.
3. II. SPECT PartZal Least Squares Anaiysis
For the PLS anaiysis, which was performed usuig Matlab software (The
MathWorks hc. 1 994), the four dependent variables which were entered into the neglect
subtest block (matrix) were the log transformed neglect subtest scores. The image data
block (matrix) contained 160 segments fiom the cortical nrn and ROI procedures,
including 152 segments ffom the cortical rim program and 8 fiom the ROI program - 2
for each Iefi and right thalamic nuclei and 2 for each lefi and right basal ganglia. Pnor to
analysis and similar to the above regression analyses, the influence of any variables found
to be significantly different between the groups on univariate tests must be removed.
Removing the contribution of covariates was doue by regressing the volume on the
individual subtest scores and then using the calcuiated residualized scores ~ o m the
regression analyses as the new subtest scores.
Following analysis, the saliences from the PLS output were remapped into image
space. To test the significance of the sinplar values computed by PLS, a permutation test
was used to assess the extent to which the computed imaging latent variables characterize
performance on the neglect subtests. As descnbed earlier, the test involved reordering the
rows of the subtest matrix, breaking the association between the blocks, and cornpuMg a
new cross-correlation matrix. MLR was then used to regress the raw subtest scores on to
the imaging latent variables produceci. This process was repeated 10 000 times in order to
test the signïficance of the output Grom the original dataset in a large enough sample.
Topography of Hanispatial Ncglect
4. RESULTS
4.1. Demographic Data - Resu&s
Popdation demographics were analyzed for each subgroup (Table 4). For the
LHD group, in both the CT and SPECT populations, there were no significant differences
found between the population of neglect patients and non-neglect patients with regard to
age, sex, and education. There was a significant difference found with regard to CT lesion
volume between the two groups; patients with neglect had larger lesions. For the RHD
group, there were no differences for sex or education but both lesion volume and age
differed significantly; patients with neglect were older and had larger lesions (This age
difference was not significant for the smaller SPECT population, but the trend was stitl
present). Based on these results, lesion volume was entered in al1 regression analyses for
the LHD population regression analyses as a covariate, and both age and lesion volume
were entered in a11 regression analyses for the RHD population as covariates. Table 5
contains a suMnary of the mean and standard deviation of percent damage, number of
patients with damage and the percent of patients with damage for al1 CT regions,
separated according to group and presence of neglect. Nine regions were found to have
sustained significantly more damage, assessed by percentage of the number of slices with
lesions, between the RHD groups with and without neglect, afier conecting for multiple
cornparisons with an a=0.05 (i.e., Bonferroni correction for 27 regions). The nine regions
were the parietal cortex, antenor white matter, posterior white matter, supramarginal
gyms, motor strip, sensory strip, FLi, posterior FOF and postenor FLS.
Posterior White Matter (PostWM)
hiean Percent Damage + Standard Drvhak i n 4 ofpople with damage f % ofpropde uith èamegeJ
ANTEFUOR CINGULATE (ACING) 1
BASAL GANGLIA (Be) FRONTAL (F)
PAIUETAL (P) r
Deep Temporal-Parietal-Occipital Junction (Deep-TPO) 9.8 ' I 9 a 0 18.6 + 28.1 10.8+312 20.9k31.8 n=/1/?4%] , n=I5 (38x1 n=8 [ lOO/o l n=35 f39%j
Motor Strip (Motor) 4 2 5 10.1 10.4 2 22.5 3.9 f l 2.4' 17.7 f 253' n=8 ( /8%j n= IO f X % j n-5 /12%/* r J I /46%/*
Sensory S trip (Sensory) 6.3 2 14.4 10.4 r 20.6 4 i 2 6 20.8 f 28.4+ n=8/16%1 n=/O(26%] a d / I 5 % / * r42/47%/*
I
Occipital (O) 5.0 f 18.0 112 + 18.8 2.1 i 8 2 7.4 2 16.9 n=7(/676] n=l3(33%] n=4f1O0%J n=25[18%J
1
Temporal (T) 4.7 i 10.6 8.6 r 12.4 5.0 I 9.6 11.5 k 14.4
n=/f?%j n =4 /1 O%] n=O [PA] n=2 (2%j
hferior Frontal (FM) 3 2 r 9.6 9.5 I 19.8 7.0 i 17.0 14.7 f 23.4 - n=5 11 1%j n=IO [26%j n=8 /2O0Aj n=34 [38%]
N- (0-45) 0.5 123
n=2 f4%J
14.8 + 19.9 n=?O (44x1
1.4252 n=6 fI3%]
2 3 15.1 n=/ l f24%]
15.1 + 25.0
n=l[??G. , n=6 /15%] n=5 /12%] n=34 [38%] 1
Superior Frontal (FSup) 0.9 2 5.8 o s f 2.1 1.6 I 6.8 1.8 + 7.3 . . n=/ [2%j n =4 (1 Ph] n=3 /7%] n=7 [PA]
Anterior Superior Longitudinal Fasciculus (FLS-Ant) 1-4 20-1 20.0 32-8 13.2 i 34.2 34.2 i 47.7 n=l2/27%] n= l2 f3I%J r-8/20%/* ~ r l6 /52%/ *
L
Anterior Frontal-Occipital Fasciculilc mnF-Ant\ 8.2 2 14.8 I 1.7 r 2.0 11.1+.205 21.1f26.5
N+ (1138)
1.96 fiî n =6 f I5%]
14.8 f 20.0 n=19 f49°%]
3.6 f 7 5 n = I I [?PA]
10.9 f 262 n=ll[?8%1
10.4 2 24.5
Anterior Interna1 Capsule (IC-Ant) I
Posterior Superior Longitudinal Fasciculus (FLS-Post) 1
Posterior Frontal-Occipital Fasciculus (FOF-Post)
Posterior Interna1 Capsule (IC-Post)
Inferior Longitudinal Fasciculus (FLi)
N- ( m 4 i)
2.75 11.1 n 4 ( /PA ]
3 8 f 8 n 4 . 2 (54%]
3.5 k 9 2 n=9 f 2 % ]
1.2 f: 5.7' a-3 /7%/'
1 1.8 f 24.7
N+ (i-89) 2.8 2 95
n=13 (IS%] 21 -8 f 23.0 n=57 f64%J
6.0 5 10.5 n=40 /45%j
9.6 i 21.2' II-37[42%/* 22.6 f 34.8
Note: N- refers to patients without ncglcct and N+ rcfcrs to patients with ncglcct Rcgions in bdd correspond to the fivc prcdictcd regions in the thcorctical nework for directcd attention. Numbcrs with an astcrisk w m found to be statistically d i f fmn t h m each other. af ia comcting for multiple cornparisons (p<O.OOZ).
6.5 I 15.3 n=8 (18%j
7.9 I 183 n=8 (18%/
7.0 + 17.7 n = 7 /16%J
7.3 k 193 n=6 /13%]
5.3 + 15.1 n=5 (13%j
n2:c, i 8.9 I 3 5 s
n=/3 f33%]
1 1.6 2 26.8 n=7/18%j
10.0 f 21.6 n=9 (23%] -
5.8 + 15.2 n=6 /I5%]
4.9 I 16.2, a 4 /IO%/*
9.6 I 19.6 n=9 f22Sq
5.7 f IM+ n=7[17%j _
6.9 + 14.7 n=18 /,'PA]
::yzK" I
24.6 *«).IO
a-35 /39%/*
12.7 .C 25.2 n=2/ (24%] 19.4 + XII* n=29 (33x1
Similar to the table above, Table 6 displays the mean count ratios for the SPECT
regions. On visual examination, it can be seen that for the LHD and RHD groups, patients
with neglect generally had lower average ratios, in the lefi and right regions, respectively,
compared to the group of patients without neglect. The only statistically significant
difference, after correcthg for multiple cornparisons with an a4.05 (i.e., Bonferroni
correction for 10 ipsilaterai regions), was in the basal ganglia of the RHD group ( ~ 3 . 2 ,
p<0.001). Table 7 shows the percentage of patients with damage to each of the five
predicted theoretical regions and Table 8 shows the average ratios (over the cerebellum)
for those same regions.
4.2. CT Ksual Analysis - Resufts
4.2.1. LHû Group
Examination of the CT data revealed that in the LHD neglect group, 29 of 38
patients (76%) had damage to at least one of the key theoretically predicted matornical
regions, which included the fiontal, parietal, cingulate cortical regions, as weH as the
basal ganglia and thalamus. By cornparison, however, 32 of 45 patients (71%) without
neglect also had damage which included darnage to at least one predicted region, and the
difference between the two groups was not significant &,2=0.29, n.s.). Cornpanson of
the average extent of damage to the theoretically predicted regions also showed no
statistical difference between the patients with neglect (6.8 f 7.8% average percent
damaged of the five predicted regions) and the patients without neglect (8.4 f 10.7%
average damage; t,+-0.73, m.). The group with neglect had damage to two or more of
Topognphy of Hcmispatial NegIcct
TABLE 6: SPECT PERFUSION RATIO SUMMARY FOR 20 REGIONS OF INTEREST
-
SPECT SPECT SPECT SPECT LHD LHD MD RHD
Mean corresponding Average Ratio f SD
Note: N- rcfcrs to patients without ncglcc an osrcrisk wcrc found to be strrtisticilly diffcrcnt h m cûch ohm. after corrccting for multiple compmisons (p-zO.005).
The purpose of using MLR with CT and SPECT data was to see LY a stronger
equation could be built based on both stmcîural and functional information. 13ie
equations oniy used regions involved in the theoretical neiwork for directed attention to
predict neglect pet$ormance on the SNB, addressing part of the hypothesis @art B). Two
h?,pothesis-driven approaches were used io see zfany SPECT regions added to the rnodel,
following CT regions. In the M D group, no SPECT regions entered following CT
regions. In the RHD group, the right pan-etal count ratios entered, following the CT data.
In the second approach. al1 ren regrgronsI five fiom each irnaging rn~dali@~ were entered to
see whether a srronger model could be built. No regions signifcantly entered an equation
in either the LHD or RHD analyses, although the RHD did produce a significant model
that accounted for approxirnateiy 40% of the variance.
4.8. Power Calcu&àtions for Linear Regression Anulyss - Resulfs
For each of the regression equations in the LHD and RHD group, power
calculations were performed in order to calculate the probability of detecting a significant
difference, in our dataset, if one existed. Power calculations were performed using an
a=0.004 for the exploratory approach (as explained above) and an a=0.05 for the a priori
hypothesis approach. The e f k t size used was 0.15, a medium effect size according to
Cohen (1988), with the covariate(s) accounting for 25% of the variance and the test
variables accounting for an increase of 10%. The results of the power calculations cm be
seen in Table 10 and show that there was more power for the hypothesis-driven
regression analyses. For the exploratory analyses with the CT data in both the LHD
@ower=O. 14) and RHD (power-0.40) groups, there was little power associated with the
regression equations, suggesting that strong conclusions should not be based upon the
results of these analyses. Similarly, the LHD in the hypothesis-driven approach with
SPECT data had small power (powe~0.56) associated with its regression equation and
one should be careh1 not to infer any strong conclusions based on this analysis. For the
remaining analyses, the power associated with each regression equation did not appear to
be a limiting factor.
4.9. SPECT Pclrtrài L e m Squares Anabsis - Results
Multicollinearity between regions is common with fùnctional imaging data, a
finding which poses a problem for MLR as described above. Thus PLS, which is not
adversely affected by multicollinearity, was utilized for cornparison with the results fiom
TABLE 10: POWER CALCULATION SUMMARY FOR MLR ANALYSES
# of
Covariates
Cumulative
R*
# of Test
Variables
Cumulative
R'
Effect Sizc Power
LHD - Explorrtory
LHD - Hypothesis-Driven
RHD - Exploratory
RHD - Hypothesls-Driven
SPECT ANALYSES
LHD - Hypothesis-Driven
RHD - Hypothesls-Driven
variables, covari~tcs ant nc for thc variabl ; of interest, and an
the MLR analyses. Table 1 1 shows the output for both the LHD and RHD patients, which
will be discussed separately.
4.9.1. PLS Findings in the LHD Croup
There were four singular values computed for the LHD patients. The first singular
value was 3.3363 and accounted for 84% of the summed squared cross-bloçk correlation
(SSCBC). As outlined earlier, multiple linear regression of the subtests on the latent
variables was used in conjunction with a permutation test to assess the statisticd
significance of the PLS output. MLR analysis of the subtests on the associated latent
variable for the imaging data was significant @=0.0142) with a model that accounted for
17% of the variance explained by the siibtests. The MLR performed here was testing to
see how much of the variability of the imaging latent variable was explained by the
subtest scores.
The second singular value was 1.209 and accounted for 12% of the SSCBC. MLR
analysis of the subtests on the second latent imaging variable produced a model which
accounted for 23% of the variance (p=0.0816). Similarly, the third singular value was
0.6307 accounting for 3% of the SSCBC. MLR on its related latent variable produced a
mode1 which accounted for 22% of the variance @=O. 1 0 10). The fourth singular value
accounted for less than 1% of the SSCBC and was less reliable given the high p-value
(p=0.8 163). Thus, it was not coasidered fbrther.
The top saliences, which were negatively correlateci with the first singular value,
can also be seen in Table 11 (Appendices H.l to H.3 contain a complete listing). For the
purposes of the discussion, a threshold based on a Scree Plot (Tabachnick & Fidell, 1989)
TABLE 11: RESULTS SUMMARY FROM PLS ANALYSES
1 Latent Variable
Singular Value
1 - p-value
1 Subtest Saliences -0,2828 0.8788 -0.0557 0.3803
LF ln f 7-21 -0.2861
LF ln f 638 -0.2288
L FInf 6-11 4.2249
L F ln l 7-22 -0.2 143
R Flnf 7-2 4.1888
1, Flnf 7-20 -0.11186
LTemp 6-19 -0.IfUS
R Tcrnp 7-5 0.0980
LACing 1-7 0.1063
L ACing 2-23 0.1 103
LTernp 7-17 0.1 113
R Tcmp 7-6 0.1456
Draw Line Biseciion
0.6864 -0.0 1 75 0.3968 0.6092
L PSvp 3-14 4.1379
L PSup 4-14 4.1332
L Lat0 4-12 4.1 292
LPSup 3-13 4.1224
L Plnf 3-15 4.1 168
L Lat0 4-13 4.1 139
L Psvp 2-12 -0.1 I l 6
L Plnf 4-15
0.493 1 -0.6799
LAClng 2-23 -0.2294 R BG
-0.2053 R SM 2-4
-0.1921 R Plnf 5-8
4.1671 R TH
4.1609 L Tcmp 6-1 8
O. 1337 R Flnf 7 3
0.2436 L FSup 2-21
0.2628 L ACing 733
Top Image Saliences
according to Scree Plot
4.1526- R Plnf 6-8
4.14% R Plnf 3-8 -
-0,1476 R Plnf 4-8
-0.1475 n,s,
. . m
bottom, rcspcctivcly. L = Lcfl, R = Right, Rcfcr to abbrcviation list for codc to rcgions. L
Notc: For LV2 t LV3, lhe iKg le (in h l d ) and p s i c salicnccs arc in ordcr from thc top and
was used to impart "significance" for saliences. By plotting the saliences in descendhg
order, it was possible to locate the point (or salience) on the graph that departed fiom the
slope of the initial points. Stated differently, if a straight line was fit through the data
starting with the h t two highest saliences, then the salience that would change the slope
of that line was considered the threshold point. Al1 saliences above that point were
considered "significant" (Appendices H.5 to H.7 contain Scree Plots for LV1 to LV3).
Although the corresponding brain regions within this range can be regarded as influentid,
regions outside of this range should not necessarily be regarded as meaningless. Perhaps a
more appropriate method of evduating the significance of these saliences is to compare
the relative saliences of surroundhg regioas.
In addition, based on the saliences for the first singular value, a singular image
(SI) was produced (Figure 6). The singular image is displayed in a linear gray colour-
coded scale in which saliences can range fiom negative to positive and corresponds tu a
gray scale with shades going fiom black to white. Thus, regions that have high negative
saliences appear black and regions with positive saliences appear bright white. (This is
more pertinent for the second and third singular images.) Interpretation of the latent
variables requires knowledge of the relationship between the imaging and subtest
saliences. For example, a negative irnaging salience and a positive subtest salience means
that higher scores on the subtest battery will be associated with lower blood flow ratios.
From both the table and the singular image for L W , it can be seen that the left
parietal and lefi occipital regions emerge as the most salient regions associated with the
first singular value. In a similar way, the top saliences c m be viewed in Table 1 1 for both
the second and third singular values, with one exception. In contrast with the f h t singular
value, these two variables have both negative and positive saliences. Thus, the top
saliences for these two variables correspond to the "significant" negative saliences (h
bold), in descendhg values h m the top, and the top positive saliences, in descending
values fiom the bottom.
Singular images were also remapped for the second (Figure 7) and third (Figure 8)
singular values. For the second singular value, the lefi and right inferior frontal regiofls
emerged as the most wgatively salient regions whereas the lefi and right temporal md
lefi antenor cingulate regions emerged as the most salient in the positive direction. For
the third singular value, a number of different regions emerged, including the lefi anterior
cingulate, the right basal ganglia, the right sensorimotor, right inferior parietal, and the
right thalamus that al1 correlated negatively with the third singular value. On the other
hand, the left anterior cingulate, left supenor frontal, right inferior frontal, right supenor
parietal, and the lefi temporal regions al1 had positive saliences with the third singular
value.
The image and subtest scores for the first latent variable were placed in a
scatîerplot to visualize the relation and to see if any patterns emerge. The advantage of
this visual approach is that it facilitates identification of specific relationships within the
PLS output by graphically portraying the data. Detailed examination of each plot can lead
to the classification of specific subgroups or patterns within the dataset. By identimg
patients located at the perimeter of each graph, it is possible to recognize both cornmon
and unusual pattern trends within the dataset. The image (or subtest) scores, as described
Figure 7: Singular Image for the Second Latent Variable in the LHD Group Note: Regions with high ncgative s~liences look black, segments with psi l ive saliences apperir bright white, and near zero saliences appcnr gray.
Figure 8: Singular Image for the Third Latent Variable in the LHD Group Note: Regions with high ncgative saliencçs look black, segments with positivc saliences appcar bright white, and ncar zero saliences appcar gray.
earlier, were caicuiated by muitiplying the image (or subtest) saliences by their original
raw value in a patient and then summing the products for a given individuai. The resuitant
plots for the f k t , second, and third latent variables can be seen in Figures 9, 10 and 11,
respective1 y.
Examination of Figure 9 revealed a separation of patients with and without
neglect. The majority of patients with neglect (dark circles) were found on the right side
of the page while the majority of those without neglect (light squares) were found on the
lefl side of the page. The three points at the far right side of the page were patients with
higher neglect scores, who had abnormal subtest scores on the drawings, lule
cancetlation, and shape cancellation tasks. Thus, a pattern of neglect severity could be
seen across the page, extending fiom the rightmost side of the page with the most severe
patients, decreasing in a gradient fashion, to the lefi side of the page, which mostly
contained patients who had normal performance on the neglect battery.
Examination of Figure 10 revealed interesting subtest and image patterns for the
second LV, identifjmg specific subgroups. In a circular pattern around the penmeter of
the plot, the type of subtests on which patients were found to show abnormal performance
varisd. The two patients in the top right corner (group A) had abnormal scores on line
bisection and shape cancellation. Proceeding in a ciockwise direction, the next three
patients (group B) also had poor performance on line bisection and shape cancellation,
while the bottom of the three had an abnonnal line cancellation as well. The next two
patients (group C) were abnormal only on the shape cancellation task. The next patient
(group D) had poor performance on the drawhg, line and shape cancellation tasks.
, l . opogriipliy 01' I-iciiiisliiitinl Ncglcci
Figure 9: Image vs Subtest Scores for LV1 in LHD Group
Category
Cl No Neglec t -2 - 1 O 1 2 3 4 5
Siibtest Scores for the First Latent Variable
, l . opognipliy of I Icriiispoiiiil Ncglcci
Figure 1 0: Image vs Subtest Scores for LV2 in LHD Group
Group E
Group D
Group B
Category
Neglect
No Neglect -1.00 -. 50 0.00 .50 1 .O0 1.50
Subtest Scores for the Second Latent Variable
Finally the patient at (group E) only showed poor performance on line bisection. Thus the
pattern of abnormal subtest performance was line bisection in the upper quadrant, shape
cancellation in the lower right quadrant, line cancellation in the middle to lower part of
the page, and drawing in the middle to lower lefi quadrant.
With regard to the SPECT data in Figure 10 of the second LV, the top half of the
page contained patients with primarily anterior hypoperfbsion including the fiontal cortex
and basal ganglia, relative to the normal control data. The bottom half of the page
inc luded patients with primarily posterior damage.
Similady, examination of Figure 1 1 revealed distinct image and subtest patterns
for the third LV. Starting at the top right quadrant, the three patients (group A), one with
neglect and nvo without, had hyperperhion in both the right and left hemispheres,
compared to normal controls. The patient with neglect had abnormal performance on the
line bisection and the shape cancellation tasks. In a clockwise fashion, the patients (group
B) at the rightrnost side of the middle part of the page had poor performance on drawings
and line bisection, although their SPECT ratios were in the normal range. At the bottom
lefi of the page, the patient (group C) was identified as having poor performance on the
shape cancellation task and had decreased flow in their right occipital region. The next
patients above (group D) also had poor performance on the shape cancellation task and
had decreased perfiision in the occipital and parietal regions. Finally, the patient above
(group E) had poor performance on the shape cancellation and line bisection. This patient
had perfusion ratios on the left side within the normal range, but increased ratios in al1
regions on the right side. Thus, the pattern of abnormal subtest performance was line
bisection in the middle to upper quadrant, shape cancellation in the left side of the page,
and drawing in the lower right quadrant. The image pattern that seemed to ernerge was
hyperperfbsion in the upper quadrants, and hypoperfbsion in the lower quadrants.
In summav* a number of distinct patterns were seen in the three graphs above.
Figure 9 revealed a pattern of neglect per$omnce severiiy. Figure IO illustrated hvo
relationships berween imaging ratios and peflorrnance on the neglect subtests; one
between anterior hypope~i ion on SPECT and poorer performance on zhe Iine bisection
task, and the other between posterior hypoperjion and poorer performance on the
shape cancellation fask. Figure 1 II revealed a relationship of posterior ipsilateral
hypoperjusion with the drawing task. and an association of righf hemisphere
hyperpe&sion and the line bisection and shape cancellation tasks. Thus, examination of
the scatterplots of the image scores and subtest scores for each latent variable enabled
distinct subtest relationships to be idenrifed within the SN& despite the fact that the
subtesrs are highly correlated.
4.9.2. PLSFindings in the RHD Group
There were also four singular values computed for the RHD patients, but only the
first singular value emerged as significant. The fint singular value was 4.334 and
accounted for 95% of the SSCBC. Multiple linear regression of the subtests on the
associated latent variable for the imaging data was significant (p=0.0188) with a mode1
that accounted for 47% of the variance explained by the subtests. It is interesting to note
that compared with the LHD results, in which each of the three latent variables were
found to be significant, only one latent variable emerged in the RKD analysis.
The top saliences, which were al1 negatively correlateci with the first singular
value, c m be seen in Table 11 (Appendix H.4 contaias a complete listing). A singular
image was produced and can be seen in Figure 12. The most saiient regions included the
right lateral occipital, nght idenor and supenor parietal, and right parietal-temporal (see
Appendix H.8 for Scree Plot).
The scatterplot of the image and subtest scores for the first latent variable in the
RHD group can be seen in Figure 13. Similar to the LHD plots, a circula pattern around
the perimeter of the graph was seen with abnormal performance on the subtests. The
upper right quadrant contained patients (group A) with severe neglect who had poor
performance on al1 subtests. The bottom middle part of the page (group B) contained
patients with poor performance on the line bisection and shape cancellation tasks. Some
of these patients also had abnormal drawing or line cancellation scores. The patient (dark
circle) at the bottom left of the page (group C) had normal SPECT ratios, relative to the
normal controls, and an abnormal line bisection (score=6, mild neglect). The upper lefi
quadrant contained the majonty of patients who did not show abnomal performance on
the battery (group D). In sumrnary. Figure 13 revealed a relatioruhip of severity of
neglect peflormance and showed the heterogeneity of patient graups, even in severe
patients with respect to perjonnance on the subtests.
4.9.3. Summav
Partial L e m Squares was used with the SPECT data to explore the relationrhip
between counr ratios in the cortical rim and other regional segments and peflonnance on
the SM. The goal war to assess whether regions postulated to be pan of the network for
Figure 12: Singular Image for the First Latent Variable in the RHD Group Note: Rcgions wiih high ncgaiivc salicnccs look black, and segments with neat zero salienccs look white, with thc rcmaining segments in shades of gray.
Topograpliy of t-lciiiispiitial Ncglcci
Figure 13: Image vs Subtest Scores for LV1 in RHD Group
Group A
Group C
I I 1 w I w
Category
si No Neglect
Subtest Scores for the First Latent Variable
directeci attention wouid emerge as strong predictors of neglect, and whether any
additional regions not previousfy considered would emerge as well.
In the LHD group, three latent variables were produced that could be rcsed to
explore the brain-behaviour relationships. The first L V ident~ped the le3 superior and
inferior parietal regions and the lep lateral occipital cortex as the most salient regions.
mese regions were involved in predicting 3 of the 4 subtests, drawings. line and shape
cancellarion tash. Line bisection emerged as the most salient subtat in the second latent
variable. which was associated with lefC and right inferior fiontal regions, and the le9
temporal region. m i s second LV also revealed a second relationship behÿeen poor
pe~formance on the drawings with the lefi and right temporal cortex, and the left anterior
cingulate. Finalfy, the third L V revealed a relationship between poor performance on
drawings, line bisection and line cancellation with lower segment ratios in the lefi
anterior cingulate, rïght basal ganglia, nght sensorimotor, righ t inferior parietal. right
thalamus, and le9 in ferior fiontal regions. m e third L V also revealed an association
between shape cancellation and the left anterior cingulate. temporal and inferiorfiontal
regions. In addition, each of the LVs showed distinct subtest peflormance and imaging
ratio patterns, idennfling specl$c subgroups, when the imaging saliences were plotted
against the subtest saliences.
In the RHD group, on& one L V emerged as strongly associating aZl four subtests
wirh decreased jlow in the right Zateral occipital, and parietal and parietotemporal
regions. In addition, specifc subgroups, according to either subtest or imaging patterns,
were identified by examining the plot of irnaging vs. subtest salience for the jirst latent
variable. Table 12 contains a surnmary of the resuZ~fiom the PLS and MLR analyses.
TABLE 12: SUMMARY OF THE RESULTS OBTAINED FROM THE MLR AND PLS ANALYSES
I LHD Patients
MLR
CT & SPECT
CT - Exploratory 1 --- (Left Anterior Cingulate),
(Lefi Frontal [Superior])
SPECT-Hyp. 1 Le f t Parietal [Inferior], (Lefi Thalamus)
CT then SPECT 1 ---
CT-SPECT Forced 1 --- PLS I
SPECT
RHD Patients
Latent Variable 1
Latent Variable 2
Latent Variable 3
(Right Posterior White Matter), (Right Occipital)
Left Superior Parietal (-1, Left Lateral Occipital {-), Left lnferior Parietal (-1
Left Inferior Frontal (-1, Right lnferior Frontal 1-1, Left Temporal {-,+}, Right Temporal {+),
Left Anterior Cingulate {+) , Lefi Anterior Cingulate {-,+) , Left Temporal {-), Right Basal Ganglia (-1, Right Sensorimotor (-1, Right Inferior Parietal (-1, Right Thalamus {- ) ,
Left Inferior Frontal {+), Right Superior Frontal {+}
Right Anterior Cingulate,
Right Parietal [Supramarginal Cyrus), (Thalamus)
(Right Parietal)
Right Parietal (SPECT) , ( K i n g {CT) ), (Thalamus (CT))
(Right Parietal { S P E C ~ ) , (ACing {CT)), (Thalamus (CT))
Right Lateral Occipital (-1, Right Parietotemparal {-), Right Inferior Parietal (-1, Right Superior Parietal {-)
modality or positive or negativt salicnccs in thc PLS section. Arcas in bold rcfcr to significant rcgions in thc MLR (pCO.05) or PLS (according to a Scm Plot).
5. DISCUSSION
5.1. Intetpreîrtion of Resufts
5.1.1. General Overvitw
The structural and fiinctional results h m this study were supportive of an
anatomical network underlying hemispatial negiect and indicate that different neural
components may be important for each hemisphere. Overall, the parietal cortex emerged
as the brain region most correlated with hemispatial neglect in both hemispheres. Data
fiom the LHD group of patients supported an anatomical network whicti included the left
parietal, antenor cingulate, lateral occipital, temporal, and froatal cortical regions.
Evidence fiom the RHD gr ou^ analyses were supportive of an anatomical network
including the right parietal, anterior cingulate, lateral occipital, and parietotemporal
cortical regions.
In the LHD group, evidence fiom the hypothesis-driven MLR analyses of the
SPECT functional irnaging data demonstrated that the left parietal region was a predictor
of neglect, as measured by performance on the Sunnybrook Neglect Battery. Further
support came from the PLS analysis of the SPECT data which also revealed additional
distinct relationships. A strong relationship emerged between decreased perfusion in the
left inferior and supenor parietal, and lateral occipital regions and greater neglect as
indexed by an increased SNB score on 3 of 4 subtests. A relationship was also shown
between decrease in the lefi inferior frontal and temporal regions with higher scores on
the line bisection task. The drawing task was shown to be more highly associated with
decreased perfusion in the anterior cingulate and lefi and right temporal regions.
Simfiarly, the shape canceHation task was shown to be more associated with decreased
perfusion in the left antenor chgulate and inferinr fiontal, and right superior fiontal
regions. Results fiom the PLS anaIysis also revealed an unanticipated bilateral hding in
the LHD, which was in support of the theory of right hemisphere dominance for attention.
In the second and third latent variables, regions in both the Lefi and right hemisphere
emerged as negatively associated with performance on the subtests of the SNB. For
example, in the third LV, the left and right infenor fiontal cortices were negatively
associated with higher scores on the shape cancellation task.
in the RHD group, greater structural damage on CT in the nght parietal and
antenor cingulate cortical regions comelated with poorer performance on the SNB.
Damage to the posterior white matter fibre bundles, specifically, the FLi, posterior FLS
and FOF, also correlated with negkct. A combination of structural and fimctional data
from the same patients revealed that only decreased perfùsion in the parietal lobe on
SPECT was a significant predictor of SNB score, when combined with the CT data in a
MLR analysis. Further support came fiom the PLS analysis of the SPECT data which
revealed a strong relationship between decreased perfusion in the right infenor and
superior parietal, lateral occipital, and parietotemporal cortical regions and increased
score on the four subtests of the SNB.
Because different relationships emerged for the right and lefi hemisphere-
damaged populations, they will be discussed separately and then compareci. The
Although the fiontal cortex was found to be damaged to a greater extent on CT in
patients with neglect compared to patients without neglect (45% of patients compared to
22%, p<0.05), it did not surface in the MLR analyses. This is consistent with a previous
group study which showed that lefi hemispatial neglect was associated with right
posterior lesions and not anterior darnage (Vallar, 1993). Our data show4 thar al1
regions, with the exception of the anterior cingulate, that emerged in the MLR and PLS
analyses were pst-rolandic. This is aiso supported by the fact that 90% of patients with
neglect had darnage to posterior brain regions. Although the frontal lobe may have been
expected to show diaschisis on SPECT as a result of posterior damage (Perani, Vallar,
Paulesu, Alberoni, & Fazio, 1993), our data did not support this suggestion in relation to
the topography of hemispatial neglect. In the PLS analysis, the supenor frontal region did
show a negative salience (-0.0983) with the neglect battery, as did the regions above;
however, its salience was well below the designated threshold for distinction, and was not
regarded as supporting evidence.
The right basal ganglia also did not emerge as a significant predictor of neglect
performance on the SNB, as expected fiom the theoretical mode1 for directed attention.
The basal ganglia are involved in the neural programming of movement and have many
reciprocal connections with the fiontai lobe. Given that the frontal lobe regions did not
emerge strongly in any of the analyses, it is perhaps not surprising that the basal ganglia
did not enter as well. It was anticipated to emerge in the SPECT analyses, since it was the
only region in RHD patients to show a statistical significance @<0.002), afier correcting
for multiple comparisons, between the group with (0.793 mean ratio) and without (0.898
mean ratio) neglect. This finding may have been incidental, though, as a result of the fact
that patients with neglect had larger lesions. in the PLS analysis, the nght basal ganglia
on SPECT did show a negative salience (-0.0787) with the neglect latent variable,
although this value was well below the range considered to be significant. Thus, the data
fiom this study do not support a role for the basal ganglia in relation to hemispatial
neglect.
Although the frontal lobe, basal ganglia, and thalamus did not emerge fiom these
analyses, this does not necessarily mean that they are not involved in the anatomical
network for directed attention. The data for this study were derived fiom patients with
bnindamage, whose functional network for directed attention was disrupted. The fact
that a region did not emerge in our analyses cannot be used as evidence that those regions
are not involved in the network for directed attention in a normal functioning brain.
Recent studies by Gitelman et al. (1996) and Nobre et al. (1996) have provided
supporting evidence for the postulated cortical network for directed attention in the
normal bchaving adult human. Using functional MRI (MRI) in normal subjects, they
have s h o w that the frontal, parietal and cingulate cortices were activated in tasks
requiring directed attention and spatial orientation. The results fiom the current study are
based upon lesion localization of hemispatial neglect. Using such lesion data, inferences
may be drawn about regions necessary for the disruption of normal function, Le., the
regions that when darnaged cause abnomal fùnction of this network. The fact that a
region does not emerge in such analysis suggests it may not be critical for this fûnction,
but not that it does not participate in normal performance.
Another reason these regions rnay not have emerged rnay be that their influences
may be more subtle. The fact that these regions show smaller relationships (Le., smaller
saliences) does not mean that those relationships are meaningless. One way in which to
test the stability, and hence reliability, of a salience is to use a bootstrap technique. Using
this approach, by resampling the data and recalculating the saliences. an estimate of the
error associated with each salience can be determined. In this way, saliences which are
found to have small standard errors can be regardeci as reliable and probably reflect
regions with minor but significant influences, whereas those saliences with a large
standard error may be more unreliable. Mesulam (1990) postulated that the more regions
in the theoretical network damageci, the greater the severity the resultant neglect. It rnay
be that damage to the frontal, basal ganglia, and thalamus may not be in and of itself
sufficient to cause neglect but rather the combination of regions damaged may be
important both to the initial occurrence and to the persistence of neglect over time.
In summary, this is the first study to do a detailed in-depth examination of the
topograp hy of le) hemispatial neglect in a large population of consecutive stroke patients
rtirh unilateral lesions and to provide both sntctural and functional imaging cotrelates
of neglect from the same population of stroke patients. Since this study was comprised of
a large patient population. we were able to test the theoretical network for directed
attenriorz to see whether the same predicted regions would emerge in our analysis, which
utilized powerful statistical techniques such as MLR and PLS. The regions in the right
hemisphere rvhich emerged in accordance with the theoretical network for directed
attention were the parietal and anterior cingulate regions, but not the basal ganglia,
rhalamus or frontal regions. In addition, the lateral occipital and parierotemporal
cortical regions were also shown to be predictive.
5.1.5. Comparison of the Right and Lefi Hemisphere Networks
The results of this study supported the concept of a neuroanatomical network for
directed attention in both hemispheres, damage to which correlated with the measures of
hemispatial neglect. In addition, evidence was s h o w that the white matter fibre bundles
comecting these regions were of importance. The specific regions involved in this
network differed for each hemisphere. The inferior and superior parietal, anterior
cingulate, lateral occipital, and the parietotemporal regions were implicated in the rïght
hemisphere. The basal ganglia, thalamus, and frontal regions were not. For the lefi
hemisphere, the inferior and superior parietat, anterior cingulate, lateral occipital, and
inferior frontal regions, but not the basal ganglia and thalamus were irnplicated.
One interesting feature of the PLS analysis in the LHD groups was that the latent
variabIes produced shared the majority of variance across three latent variables, as
compared to the RHD analysis, which mainly loaded o d y on the first latent variable. One
possible reason for this may have to do with right hemisphere dominance for attention
(Weintraub & Mesularn, 1988; Posner & Petersen, 1990). The right hemisphere,
specificaliy the parietal lobe, has been demonstrated to activate in relation to both lefi-
and right-sided stimuli, although more so for contralateral stimuli, whereas the left
hemisphere is activated only by right-sided stimuli (Corbetta, Miezin, Shulman, &
Petersen, 1993; Heilrnan, Schwartz, & Watson, 1978). As discussed earlier, the parietal
lobe emerged as the most reliable and significant predictor of neglect performance, more
strongly in the right compared to the left hemisphere. in the case of RHD, there is little, if
any, compensation from the left hemisphere, so the neglect deficit is more severe, and the
Topopphy of Hemispatial Negiect
strong influence of the right parietal lobe could be discemed. It is Iikely that the parietal
region is a crucial region within the theoretical anatomical network for directed attention,
such that dysfunction in that region causes a more severe deficit. For LHD, on the other
hand, the ability of the right hernisphere to cornpensate, either completely or partially, for
the lefi-sided damage may have been the reason that right hemisphere regions emerged in
relation to neglect. For example, in the second latent variable in the LHD group, the lefi
inferior frontal regions had a high negative saliences (-0.2861), which is expected for
right sided neglect. In addition, the homologous infenor frootal region on the right
hemisphere also had a high negative salience (-0.1886), which is consistent with the
theory of right hemisphere dominance for attention.
Another intriguing finding, as briefly described above, concemed the first latent
variable From the PLS analysis and identified possible qualitative differences of neglect
within each hemisphere. In the RHD group, the four subtests positively correlated in
approximately equal arnounts in deriving the first latent variable. In contrat, ody three
(drawings, shape and line cancellation) of four subtests loaded on to the first latent
variable in the LHD group. Line bisection did not contribute to the first latent variable,
but it was the main contributor in the derivation of the second latent variable in the PLS
analysis. It may be that line bisection task is probing a different subcomponent of neglect
fiom the other tasks, or that a different combination of brain regions are cntical for the
line bisection task. For instance, on the first latent variable the regions on SPECT with
the highest negative saliences were the posterior ones while in the second latent variable
the regions with the highest negative saliences were more anterior, specifically the
inferior fiontal region. The fiontal region was related more to the line bisection task while
the parietal-occipital region came out as more related to the other three tasks. One reason
for this differentiation could be that the line bisection task requires more judgment, Le., it
is a cognitive estimation task, which is known to involve the fiontai lobe, while the other
three subtests may be more influenceci by perceptual discrimination processes primarily,
involving the parietal lobe. in addition, the anterior cingulate and temporal regions had
high positive saliences (0.1 103, 0.1456), on the second latent variable, which were
associated with omissions on the drawing task. The drawing task involves feature
detection in objects, which should require temporal lobe functioning (üngerleider &
Mishkin, 1982). This may be the reason that the drawing task did separate fiom the other
more visuospatial perceptual tasks (Goodale & Milner, 1992).
In summav, structural and functional data from this smdy provided evidence for
distinct anatomical networks underlying hemispatial neglect in the le@ and right
hemisphere. Evidence was put fomard to suggest a qualitative d i f l e n c e nof on&
regarding fie neural components of hemispatial negfect in each hemisphere, but also
t-egarding the measures used to capture the neglect phenornenon. m i s siudy h a provided
evidence from s~?rc~uraf and functional imaging in the same population and has clearly
s h o w distinct differences between the le> and right hemisphere in relation to the
ropography of hemispatial neglec f ,
Topography of Hcmispatial Ncglccf
5.2. Study Limitations and Future Directions
One limitation of this srudy was that the battery used to assess neglect was
composed only of visuoconstructive tasks. Each subtest captures a different component of
the neglect phenomenon, as shown by factor anaiysis. For exarnple, the spontaneous
drawing requires an interna1 representation of that object as well as complex
constmctional praxis skills. Shape and line cancellation tasks are prirnarily target
detection, visual search tasks, requiring sustained attention as well. Line bisection,
mentioned earlier, is a cognitive estimation task, and requires making a judgment about
distance. While many patients were impaired on al1 tasks, some showed dissociations,
even though analysis of the psychometric propenies of the battery revealed that ail four
su btests were al1 required in the factor analysis.
For these reasons it is certainly important to assess neglect on multiple tasks, a
shortcoming of many previous studies of neglect. By using four subtests to assess neglect,
this study was able to identiQ minor degrees of neglect, but as a result the underlying
brain correlations may have been more difficult to deconstruct. Another source of error
with regard to the battery was the fact that not everyone had a complete battery and a
composite partial score was exmpolated which probably underestimated the hue score
because a conservative formula was used. It was not always possible to obtain scores on
al1 subtests of a battery, which would have been preferable for correlation analysis. In this
study, only a smali nurnber of patients required interpolated scores. Aithough the
composite scores were shown to be highly correlated with the actual scores on the SNB,
linear regression analysis could be used to formulate an equation that could be used to
calculate better composite scores.
5.2.2. Testimg Dates
Although testing dates for this study were optimized as best they could, an
additional source of error could arise due to the tirne differences between procedures. For
example, although the difference in time between neglect testing and SPECT scanning
was minimized, not al1 patients could be scanned as desired during the same week that
neglect was tested. Only a minority could be scanned on the same &y. In many other
situations, such as PET activation studies, the subject actually perfonns the task of
interest d u h g scanning. in a traditional lesion study, the behavioural deficits are assessed
O ff-1 ine and correlated with structural andor functional damage documented within a
reasonable tirne interval. Because scanning was fiequently performed for clinical
purposes, this time interval was difficult to control. This is of less concem in the chronic
stable phase after recoveiy from stroke, but is an issue in the acute, evolving stage afier a
suoke.
Hemispatial neglect, as described previously, can recover quickly. Thus to
meaningfûlly correlate the hinctional imaging data with absence of neglect, the battery of
tests must be adrninistered within a few days of scanning, as was done in this study. For
the other situation, Le., severe persisting neglect, it would be possible to be more lenient
about the tirne interval between the behavioural testing date and scanning onset, on the
assumption that neglect was present throughout the interval. However, the greater the
time period between testing procedure and scanning, the greater the likelihood of adding
e m r to a correlation between brain and behaviour. The testing t h e interval used for this
snidy was not always ideal, but was a reasonable tmdesff that allowed for the inclusion
of many additional patients.
S. 2.3. CT Scan Limitations
CT scans were performed for clinical reasons generally within the first 48 hours
post-stroke. Negative scans occurred in a subset of individuals, and as part of the routine
investigation at the tirne of this study, many of these patients were rescanned. in general,
the scan which best captured the full extent of damage was used, if there were more than
one scan fiom which to select. it is known that the lesion on CT probably best represents
necrotic tissue alone, without the effects of edema, if the scan was done a few months
later. Scans a few months afier the stroke could not be done in our patients, however, for
financial, pragmatic, and ethical reasons. Thus, although the most ideal scan
demonstrating the true extent of the CT lesion was not available to us, the use of scans
obtained in the acute stage of stroke provided a reasonable index of the resultant
structural darnage.
Another source of error was that tracing and subsequent lesion localization was
performed on templates fiorn a stereotactic atlas, which required the assumption that scan
orientation was parallel to the orbitomeatal line. However, this was known to be off by a
few degrees in 42% of patients (CT scan tilt was known fiorn scout film). This was taken
into account by the lesion tracer as much as possible, but added some error, although, on
the whole it probably did not affect the results substantially.
Topography of Hemispatial Ncgicct
Finally, the method used to assess the extent of structural damage was not ideal, in
that it was able to capture the sire of the lesion in each region of interest in the verticai
but not horizontal plane. As described in the methods section in Chapter 2, damage in
each CT region was quantified as a percentage of the number of slices in which damage
was evident. While this approach provided more information than simply dichotomizing
into damage present or absent in a region, it would be preferable to estimate the tnie
percentage of damage in the volume occupied by that region in each patient. Some studies
have used a subjective estimate of damage on each slice, e.g., less than 10 percent, 1 1 -
49%, or greater than 50 percent (Ferro, Kertesz, & Black, 1987). The difficulty of this
approach is the subjectivity of such judgments. Generally CT does not show sufficient
anatomical detail to define different regions reliably. Magnetic resonance imaging would
be much more reliable in this context, but it is still not available for routine investigation
of stroke patients in most Canadian centres.
5.2.4. SPECT Scan Resolution
Perhaps a larger problem with this study was the resolution of the imaging
rnodaIity and of the regions-of-interest therefore available to us. This study used a single-
headed gamma camera with an inherent resolution of about 12mm (FWHM). It is known
that structures at the edge of an image will have better resolution than intemal structures,
as a result of the back filtration algorithm used to reconstruct the SPECT images
(Masdeu, Brass, Holman, & Kushner, 1994). Structures such as the basal ganglia and the
thalarnic nuclei, which are difficult to resolve on single-headed SPECT scans under
normal circumstances, may be even more difficult to see if perfusion is reduced. Thus, it
is possible that the counts measured in those regions were significantly higher than the
m e counts, but not necessarily different by the same proportion in al1 patients. Noise is
therefore added to the image, making it more difficult to differentiate between groups.
The fact that the thalamic nuclei and basal ganglia did not emerge significantly in the
analyses in this study may have more to do with poor resolution that with the absence of
differences between the groups. Currently, there are dual and triple head SPECT carneras
that can be used. which have much higher resolution and would improve accuracy
(Devous, 1995).
5.2.5. Regional Measurements
The regions used to measure brain activity also had inherent noise as a result of
being an automated procedure. Ideally, it would have been better to have had MR scans
on each person and then trace the regions of interest with the MR anatomy as a guide.
This, of course, was not possible in this study for a number of reasons. For one, the
financial expenditure would have been enormous for a study of this size. The time
necessary to trace each scan, and the software and hardware needed to store this
information would have made this project much more expensive and complex. More
importantly, even if fhding had been available, it would not have been feasible in many
of Our patients to require them to undergo an MRI in addition to CT, which was the
clinical procedure of choice. Many of the patients were il1 and would not have been able
to undergo the additional procedure.
To reduce observer
was used to capture counts
error and increase
in segrnented brain
tirne efficiency, an automated procedure
regions, afier correcting for brain size by
linear scaling, standardking the angle of orientation, and correcthg for head tilt. An
alternative approach would be to use a template with preset regions-of-interest
comesponding to anatomy, rather than placing ROIs that are not anatornically guided.
This study utilized both techniques in an effort to measure brain activity. For the cortical
regions, an automated circular annulus was used, without specific regard to lobar
divisions. The segmental divisions were then localized anatomically. This allowed the
segments to be combined in an objective fashion prior to analysis. In addition, preset
ROIs were used to capture activity in subcortical as well as cerebellar regions.
To ven@ the anatomical designation of the regions we used a template fkom the
MR-SPECT superposition in a few subjects, where both MR and SPECT were available.
It would have been desirable, if t h e had permitted, to base the template on a larger
sample of ten to fifteen individuals. Even with a larger sarnple though, some e m r would
still be present as a result of the individual differences in brain size and lobar
differentiation between people. No single template could ever be perfectly matched for
every brain. An alternate approach could be to use a nonlinear deformation technique that
would warp each brain into a standardized stereotaxic space (Friston, Frith, Liddle, &
Frackowiak, 199 l), which could be used to reduce intersubject variation.
5.2.6. SPECT Reference Region
Another source of error using the irnaging data was the amount of within group
variation. As a result of the characteristics of SPECT, the counts measwed are not
absolute measures of blood flow, but are only relative to that particulru reference region.
Xenon inhalation techniques (Masdeu, Brass, Holman, & Kushner, 1994) can be used to
provide absolute blood flow measures on SPECT, but with marked Ioss of resolution. To
deal with this issue, SPECT counts fiom each segment were made into ratios by dividing
them by the counts in a region of the brain thought to be the least affected in the majority
of individuals in the group. in this study the cerebellar hemisphere with the higher counts
was used. This was the ipsilesional cerebellum in over 90% of patients, because it is
unaffected by direct or indirect effefts in the majority of hemisphenc stroke patients. The
cerebellum with higher counts was used rather than the ipsilesional cerebellurn as a rule
because it was determined that in association with occipital lesions, fiom postenor
circulation stroke in this study, approximately 3% of patients also had direct darnage to
the ipsilesional cerebellwn.
Although there was no mean difference between the counts in the cerebelli
benveen groups, examination of the intragroup variation found that it was high compared
to the intergroup variation, since high standard deviations were assoçiated with each
SPECT region. Part of the reason for the high intragroup variation stems fiom the fact
that each group was composed of stroke patients with lesions of varying size and location.
Another source of variation may corne fiom the fact that the cerebellum was not an ideal
reference region. A different reference region that was considered was the average counts
in the undarnaged hemisphere, which would have the advantage of comparing
homologous regions with each other. This is not an ideal approach in acute stroke,
however, since transhemisphenc diaschisis is present especially in larger lesions (Dobkin,
Levine, Lagreze, Dulli, Nickles, & Rowe, 1989). To compensate for this, it may be
possible to calculate the mean average in the undamaged hemisphere, remove any pixel
values that were greater than two standard deviations from the mean and then recalculate
the average value. This could result in a less biased and more reliable reference source.
Finding an ideal reference is not an easy task and is one of the inherent limitations of
current SPECT tracer technique.
6. CONCLUSIONS AND FUTURE DIRECTIONS
This study supports the idea that the neural correlates of hemispatial neglect
involve a network of anatomical regions subse~ ing directeci attention including the
frontal, parietal, and anterior cingulate cortices, basal ganglia and thalamus. The regions
actually found in this CT-SPECT study to be correlated with hemispatial neglect were
different in the left and right hemispheres. In LHD patients, the significant regions were
the frontal, parietal, antenor cingulate, Iateral occipital and temporal cortical regions. Ln
RHD patients, the significant regions were the parietal, anterior cingulate, lateral
occipital, and parietotemporal cortical regions. It was M e r determined that the role of
each region may not be equally important. This study reaf fms the primary role of
damage to the parietal lobe in hemispatial neglect, as suggested in the earliest clinical
pathological case reports. The shidy also provided evidence suggestive of a qualitative
difference of the neglect phenomenon in each hemisphere. Further investigation of the
nature of this difference between the hemispheres may provide further understanding of
qualitative differences of neglect and the neural components responsible. The need to
adopt different statistical techniques depending on the nature of the data and the questions
posed has also been illustrated. Both conventional MLR and a new technique PLS were
used to test hypotheses that guided the study design. Finally, this study has shown the
value of complementary structural and functional imaging techniques, such as CT and
SPECT, in conjunction with neuropsychological tests of behaviour in attempting to
elucidate brain-behaviour relationships.
Albert, M. L. (1973). A simple test of visual neglect. Neurolow. 23,658-664.
Amersharn International Place. (1987). The Clinical Amlication of rCBF Imaging; bv SPECT. High Wycombe: Brier Press Ltd.
Behrmann, M., Moscovitch, M., Black, S. E., & Mozer, M. (1990). Perceptual and conceptual factors in neglect dyslexia: Two contrasting case studies. Brain. 1 13(4), 1 163- 1183.
Belsley, D. A., Kuh, E., & Welsch, R. E. (1980). Reaession Diamostics: Identitifing influential data and sources of collinearity. New York, USA: John Wiley and Sons.
Bender, M. B., & Furlow, C. T. (1944). Phenomenon of visual extinction and binocular rivalry mechanism. Tram Am Neurol Assoc. 70, 87-93.
Bender, M. B., & Furlow, C. T. (1945). Phenomenon of Msual extinction and homonymous fields and psychologicai principals involved. Archives of Neuroloev and Psvchiatry, 53,29-33.
Binder, J., Marshall, R., Lazar, R., Benjamin, J., & Mohr, J. P. (1992). Distinct syndromes of hernineglect. Archives of Neurolow. 49(11), 1 187- 1 194.
Bisiach, E., Luzzatti, C., & Perani, D. (1979). Unilateral neglect, representational schema and consciousness. Brain. 102, 609-6 18.
Black, S. E., Ebert, P., Leibovitch, F. S., Szalai, J. P., Bondar, J., & Blair, N. (1995). Recovery in hemispatial neglect [Abstract]. Neurolow. 45. A 178
Black, S. E., Vu, B., Martin, D. K., & Szalai, J. P. (1 990). Evaluation of a bedside battery for hemispatial neglect in acute stroke [Abstract]. Journal of Clinical and Experimental Psvchologv. 12. 109
Bogousslavsky, J., Miklossy, J., Regli, F., Deruaz, J. P., Assal, G., & Delaloye, B. ( 1 9 8 8). Subcortical neglect : neuropsychological, SPECT, and neuropathological correlations with antenor choroidal artery temtory infarction. Annals of Neurolow, 23(5), 448-452. -
Bookstein, F. L. (1994). Partial least squares: A dose-response mode1 for measurement in the behavioural and brain sciences. Psvcoloauv. 5(23)
Bookstein, F. L., Streissguth, A. P., Sampson, P., & Barr, H. (1 990). Measuring 'dose' and 'response' with multivariate data using partial least squares techniques. Commun Stat Theor Methods. 19, 765-804.
Brain, W. R. (1 941). Visual disorientation with special reference to lesions of the right cerebral hernisphere. Brain, 64,244-272.
Cappa, S. F., Perani, D-, Bressi, S., Paulesu, E., Franceschi, M., & Fazio, F. (1993). Crossed aphasia: a PET follow up study of two cases [see comrnents]. Journal of Neurolom, Neurosurgew & Psvchiatry, 56(6), 665-67 1.
Cappa, S. F., & Vallar, G. ( 1 992). Neuropsychological disorders afier subcortical lesions: Implications for neural models of language and spatial attention, In G- Vailar, S. F. Cappa, & C. W. Wallesch (Eds.), Neuropsvcholo~cal disorders associated with subcortical lesions. (pp. 7 4 1 ). Oxford: Oxford University Press.
Cohen, J. ( 1 988). Statistical Power Analvsis for the Behavioural Sciences. (Second ed.). Hillsdale, New Jersey: Lawrence Erlbaum Associates.
Cohen, J., & Cohen, P. (1983)- A~plied Multi~ie Reszression/Correlation Analvsis for the Behavioural Sciences. (Second ed.). Hillsdale, New Jersey: Lawrence Erlbaum Associates.
Corbetta, M., Miezin, F. M., Shulman, G. L., & Petersen, S. E. (1993). A PET study of visuospatial attention. Journal of Neuroscience. 1 3(3), 1202- 1226.
Coslett, H. B., Bowers, D., Fitzpatrick, E., Haws, B., & Heiiman, K. M. (1990). Directional hypokinesia and hemispatial inattention in neglect. Brain. 1 13(Part 2), 475- 386.
Critchley, M. (1966). The Parietal Lobes. New York: Hafner.
Damasio, A. R., Damasio, H., & Chui, H. C. (1980). Neglect following damage to frontal lobe or basal ganglia. Neuro~s~choloaia. 18(2), 123- 132.
Damasio, H. (1 983). A computed tomographic guide to the identification of cerebral vascular territones. Archives of Neuroloav, 40(3), 1 38- 142.
Damasio, H. ( 1995). Human Brain Anatomv in Com~uterized Images. New York: Oxford University Press, inc.
de la Sayette, V., Bouvard, G., Eustache, F., Chapon, F., Rivaton, F., Viader, F., & Lechevalier, B. (1 989). Infarct of the antenor Iimb of the right intemal capsule causing le fi motor neglect: case report and cerebml blood flow sîudy. Cortex. 25( 1 ), 147- 1 54.
Devous, M. D. (1 995). SPECT functional brain imaging. Journal of Neuroimaeinq, S(Supp1 l), S2-S 13. -
Dobkin, J. A., Levine, R. L., Lagreze, H. L., Dulli, D. A., Nickles, R. J., & Rowe, B. R. ( 1989). Evidence for transhemispheric diaschisis in unilateral stroke [see comments]. Archives of Neurolo~,r. 46(12), 1333- 1336.
Ebert, P. L., Black, S. E., & Lee, L. (1996). Recovery patterns and lesion correlates in sensory extinction [Abstract]. Canadian Journal of Neurolopical - Sciences. 23. S78
Edgington, E. S. (1 980). Randomization Tests. New York: Dekker.
Ferro, J. M., Kertesz, A., & Black, S. E. (1 987). Subconical neglect: quantitation, anatomy, and recovery. Neurolow. 3 7(9), 1 487- 1492.
Fiorelli, M., Blin, J., Bakchine, S., Laplane, D., & Baron, J. C. (1991). PET studies of cortical diaschisis in patients with motor hemi-neglect. Journal of the Newological Sciences, 1 04(2), 135- 142.
Friedman, P. J. ( 199 1). Clock drawing in acute stroke. Ane - & A~eing;. 20(2), 140- 145.
Friston, K. J., Frith, C. D., Liddle, P. F., & Frackowiak, R. S. J. (1991). Plastic de formation of PET images. Journal of Com~uter Assisted Tomom~hv . - 1 5,634-639.
Gitelman, D. R., Nobre, A. C., Meyer, J. R., Panish, T. B., Callahan, C., Russell, E- R., & Mesulam, M. M. (1 996). Functional magnetic resonance imaging of covert spatial attention [Abstract]. Neuroimage. - 3. S 180
Good, P. ( 1 994). Permutation Tests. New York: Springer-Verlag.
Goodale, M. A., & MiIner, A. D. ( I 992). Separate visuai pathways for perception and action. Trends in Neuroscience. 15,20-25.
Guariglia, C., & Antonucci, G. (1992). Personal and extrapersonal space: a case of neglect dissociation. Neuro~svcholopia. - - 30(11), 100 1 - 1009.
Halligan, P. W., & Marshall, J. C. (1991). Left neglect for near but not far space in man [see comments]. Nature. 350(63 18), 498-500.
Halligan, P. W., & Marshall, J. C. (1992). Left visuo-spatial neglect: a meaningless entity? Cortex. 28(4), 525-535.
Hecaen, H. (1 962). Clinical symptomatology in right and lefl hemisphere lesions. In V. B. Mountcastle (Ed.), Interhemis~henc Relations and Cerebral Dominance. Baltimore: The John Hopkins Press.
Heilman, K. M., Bowers, D., & Watson, R. T. (1983). Performance on hernispatid pointing task by patients with neglect syndrome. Neurolow. 33(5), 66 1-664.
Heilman, K. M., Schwartz, H. D., & Watson, R. T. (1978). Hypoarousal in patients with the neglect syndrome and emotional indifference. Neurolow. 28(3), 229-232.
Heilman, K. M., & Valenstein, E. (1972). Frontal lobe neglect in man. Neurolow. 22, 660-663.
Heilman, K. M., & Valenstein, E. (1979). Mechanisrns underlying hemispatial neglect. Annals of Neurolom. 5(2), 166- 1 70.
Heilman, K. M., Watson, R. T., & Valenstein, E. (1993). Neglect and related disorders. In K. H. Heiiman & E. Valenstein (Eds.), Clinical Neuro~svchology. - ('p. 279- 336). Oxford, England: Oxford University Press, Inc.
Heilman, K. M., Watson, R. T., & Valenstein, E. (1994). Localization of lesions in Neglect and related disorders. In A. Kertesz (Ed.), Localization and Neuroimaainrr in Neuropsvcholog-y. @p. 495-524). California, USA: Academic Press, Inc.
Hellman, R. S., Tikofsky, R- S., Collier, B. D., Hothan, R. G., Palmer, D. W., Glatt, S. L., Antuono, P. G., Isitman, A. T., & Papke, R. Z. (1989). Alzheimer disease: Quantitative analysis of [l"~]iodoamphetamine SPECT brain imaging. Radiolow. 172, 183- 188.
Hier, D. B., Davis, K. R., Richardson, E. P., & Mohr, 1. P. (1977). Hypertensive putaminal hemorrhage. Annals of Neurolow. 1, 152-159.
Hier, D. B., Mondlock, J., & Caplan, L. R. (1983). Behavioral abnormalities after right hemisphere stroke. Neurolom. 33(3), 337-344.
Hier, D. B., Mondiock, J., & Caplan, L. R. (1 983). Recovery of behavioral abnormalities after nght hemisphere stroke. Neurolow. 33(3), 345-350.
Howell, D. C. (1982). 5. (Second ed.). Boston: PWS- Kent.
Kertesz, A., & Dobrowolski, S. (198 1). Right-Hemisphere deficits, lesion size and location. Journal of Clinical Neuro~svchology, 3(4), 283-299.
Kimura, D. (1 986). NeuropsychoIow Test Procedures. London, Ontario: DK ConsuItants.
Lassen, N. A., Anderson, A. R., Friberg, L., & Paulson, O. B. ( 1 988). The retention of [ 9 9 m ~ c ] - d . l - ~ ~ - ~ ~ ~ in the human brain afier intracarotid bolus injection: A kinetic analysis. Journal of Cerebral Blood Flow & Metabolism. 8, S l3-S22.
Levine, D. N., Warach, J. D., Benowitz, L., & Calvanio, R. (1986). Lefi spatial neglect: effects of lesion size and premorbid brain atrophy on seventy and recovery following nght cerebral infarction. Neurolow. 36(3), 362-366.
Masdeu, J. C., Brass, L. M., Holman, B. L., & Kushner, M. J. (1994). Brain single- photon emission computed tomography. [Review]. Neurolow. 44(1 O), 1970- 1977.
McFie, J., Piercy, M. F., & Zangwill, O. L. (1950). Visual spatial agnosia associated with lesions of the right hemisphere. Brain. 73, 167- 190.
-McIntosh, A. R., Bookstein, F. L., Haxby, J. V., & Grady, C. L. (1996). Spatial Pattern Analysis of Functional Brain Images Using Partial Least Squares. Neuroimage. 3, 143- 157.
Mclntosh, A. R., & Gonzalez-Lima, F. (1 993). Network analysis of functional auditory pathways mapped with fluorodeoxyglucose: associative effects of a tone conditioned as a Pavlovian excitor or inhibitor. Brain Research, 627(1), 129- 140.
Mesulam, M. M. (1 98 1). A Cortical Network for Directed Attention and Unilateral Neglect. Annals of Neurolom. 10(4), 309-325.
Mesularn, M. M. (1 985). Attention, Confiisional States, and Neglect. In M. M. Mesulam (Ed.), Princ_iDles. @p. 125- 168). Philadelphia, USA: F.A. Davis Company.
Mesulam, M. M. ( 1 990). Large-Scale Neurocognitive Networks and Distributed Processing for Attention, Language, and Memory. Annals of Neurolow. 28(5), 597-6 1 3.
Minoshima, S., Berger, K. L., Lee, K. S., & Mintun, M. A. (1992). An automated method for rotational correction and centering of three-dimensional functional brain images. Journal of Nuclear Medicine. 33(8), 1579- 1585.
Mishkin, M., & Ungerteider, L. G. (1982). Contributions of striate inputs to the visuospatial fùnctions of parieto-preoccipital cortex in monkeys. Behaviowal Brain Research. 6( 1 ), 57-77.
Morecrafi, R. J., Geula, C., & Mesulam, M. M. (1993). Architecture of connectivity within a cingulo-fkonto-parietal neurocognitive network for directed attention. Archives of Neurologv. 50(3), 279-284.
Nobre, A. C., Mesulam, M. M., Sebestyen, G. N., Frackowiak, R. S. J., & Frith, C. D. ( 1996). Functional localisation of the neural network for visual spatial attention by positron ernission tomography [Abstract]. Neuroirnaee. 3. S 192
Nyberg, L., Mchtosh, A. R., Houle, S., Nilsson, L.-G., & Tulving, E. (1996). Relation of hippocarnpal and medial temporal activity with episodic memory retrieval in individual subjects. Nature. 380, 71 5-717.
Ogden, J. A. (1985). Contralesional neglect of constructed visual images in right and left brain-damaged patients. Neurovsvcholopia. 23(2), 273-277.
Ogden, J. A. ( 1985). Antenor-Postenor interhemispheric Difierences in the Loci of Lesions Producing Visual Hemineglect. Brain & Cosmition. 4,59-75.
Ogden, J. A. ( 1 987). The 'neglected' lefi hemisphere and its contribution to visuospatial neglect. ln M. Jeannerod (Ed.), Newo~hvsiolo~cal and neuror>svchoIo~cal -
Pandya, D. N., Karol, E. A., & Heilbronn, D. (1971). ï h e topographical distribution of interhemispheric projections in the corpus callosurn of the rhesus monkey. Brain Research. 32(1), 3 1-43.
Pandya, D. N., & Kuypers, H. G. (1969). Cortico-cortical connections in the rhesus monkey. Brain Research. 1 3 ( 1 ), 13-36.
Pandya, D. N., & Vignolo, L. A. (1969). hterhernisphenc projections of the panetal lobe in the rhesus monkey. Brain Research. 15(1), 49-65.
Pandya, D. N., & Yetenan, E. H. (1990). Prefiontal cortex in relation to other cortical areas in rhesus monkey: Architecture and connections. In H. B. M. Uylings, C. G. Van Eden, J. P. C. De Brum, M. A. Corner, & M. G. P. Feenstra (Eds.), Promess in Brain Research. @p. 63-94). Elsevier Science.
Perani, D., Di Piero, V., Lucignani, G., Gilardi, M. C., Pantano, P., Rossetti, C., Pozzilli, C., Gerundini, P., Fazio, F., & Lenzi, G. L. ( 1988). Remote effects of subcortical cerebrovascular lesions: a SPECT cerebral perfusion study. Journal of Cerebral Blood Flow & Metabolism. 8(4), 560-567.
Perani, D., ValIar, G., Cappa, S., Messa, C., & Fazio, F. (1987). Aphasia and neglect afier subcortical stroke. A clinicaVcerebra1 perfiision correlation study. Brain. 1 1 O(Pt 3, 121 1-1229.
Perani, D., Vallar, G., Paulesu, E., Alberoni, M., & Fazio, F. (1993). Left and right hemisphere contribution to recovery fiom negiect aller nght hemisphere damage--an [ 1 8FIFDG pet study of two cases. Neuropsvcholorria. 3 1 (2), 1 15- 125.
Poppelreuter, W. K. ( 19 1 7). Die Dsvchisen Schdieuneen dwch Kopfschuss in k e i g 1 9 1 4- 1 9 1 6: Die Storungen der niederen und hoheren Leistuneen dwch Verletzuneen des Okzi~italhirns. Leipzig: Leopold Voss.
Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. [Review] . Annual Review of Neuroscience. 1 3,2542.
Posner, M. I., Waker, J. A., Friedrich, F. J., & Rafal, R. D. (1984). Effects of parietal injury on covert orienting of attention. Journal of Neuroscience. 4(7), 1863- 1874.
Riddoch, M. J. (1 99 1). Neglect - and the peri~heral - dvslexias. Hove: Lawrence Erlbaum Associates,
Rizzolatti, G., & Berti, A. (1990). Neglect as a neural representational deficit. Revue Neuroloo;ique, 146,626-634.
Robertson, 1. H., Tegner, R., Tham, K., Lo, A., & Nimmo-Smith, 1. (1995). Sustained attention training for unilateral neglect: Theoretical and rehabilitation implications. Journal of Clinical and ExDerimental Neuro~svcholoev. 1 7(3), 4 16-430.
Russ, J. C. (1992). The Image Processing Handbook. Boca Raton, Flonda: CRC Press.
Schenkenberg, T., Bradford, D. C., & Ajax, E. T. (1980). Line bisection and unilateral visual neglect in patients with neurologic impairment. Newolow. 30, 509-5 17.
Seltzer, B., & Pandya, D. N. (1984). Further Observations on Parieto-Temporal Connections in the Rhesus Monkey. Ex~enmental Brain Research. 55,301-3 12.
Shallice, T. (1 988). From Neuro~svchologv to Mental Structure. Cambridge, England: Cambridge University Press.
Shepherd, G. M. (1 994). Neurobioloq. (Third ed.). New York: Oxford University Press.
Sorenson, J. A. (1 974). Quantitative measurement of radioactivity in vivo by whole body counting. In G. J. Hine & J. A. Sorenson (Eds.), instrumentation in Nuclear Medicine. (pp. 3 1 1-348). New York: New York Academy Press.
SPSS [Computer Software]. ( 1995). Chicago, Illinois: SPSS Inc.SPSS Inc.
Stevens, J. (1 986). &died Multivariate Statistics for the Social Sciences. Hillsdale, New Jersey: Lawrence Eribaum Associates.
Stone, S. P., Patel, P., Greenwood, R. J., & Halligan, P. W. (1992). Measuring visual neglect in acute stroke and predicting its recovery: the visual neglect recovery index. Journal of Neurolom, Neurosurgerv - & Psychiatry, 55(6), 43 1-436.
Stuss, D. T., Eskes, G. A., & Foster, J. K. (1994). Experimental Neuropsychological Studies of Frontal Lobe Functions. In F. Boller & J. Gramian (Eds.), Handbook of Neuro~svchology. @p. 149- 1 85). Elsevier Science B.V.
Stuss, D. T., Shallice, T., Alexander, M. P., & Picton, T. W. (in press). A multidisciplinary approach to anterior attentional functions. Annals of the New York Academv of Sciences.
Tabachnick, B. G., & Fidell, L. S. (1989). Using Multivariate Statistics- (Second ed-). New York, USA: Harper and Row.
Talairach, J., & Toumoux, P. (1988). Co-~lanar Stereotaxic Atlas of the Human Brain. New York: Thieme.
Tegner, R., & Levander, M. (1991). Through a looking glass. A new technique to demonstrate directional hypokinesia in unilateral neglect. Brain. 1 14(Part 4), 1943- 195 1.
MATLAB [Cornputer Software]. (1 994). Natick, MA: The MathWorks, 1nc.The MathWorks Inc.
Ungerleider, L. G., Desirnone, R., Gaikin, T. W., & Mishkin, M. (1984). Subcortical projections of area MT in the macaque. Journal of Com~arative Newolow. 223(3), 368- 3 86.
Ungerleider, L. G., & Mishkin, M. (1982). Two cortical visual systems. In D. J. ingle (Ed.), Analvsis of visual behaviour. @p. 549-586). Cambridge, MA: MIT Press.
Vallar, G. (1993). The anatornical basis of spatial hemineglect in humans. In 1. H. Robertson & J. C. Marshall (Eds.), Unilateral Neglect: Clinical and Ex~erirnentd Studies. (pp. 27-57). New York: Lawrence Erlbaum Associates.
Vallar, G., & Perani, D. (1 986). The anatomy of unilateral neglect afier right- hemisphere stroke lesions. A clinicaVCT-scan correlation study in man. Neuro~sycho1o~ia. 24(5), 609-622.
Vallar, G., Perani, D., Cappa, S. F., Messa, C., Lenzi, G. L., & Fazio, F. (1988). Recovery fiom aphasia and neglect after subcortical stroke: neuropsychological and cerebral perfusion study. Journal of Neurologv. Neurosur erv & Psychiatry. 5 1 ( 1 O), 1 269- 1276.
Vallar, G., Rusconi, M. L., Geminiani, G., Berti, A., & Cappa, S. F. (199 1). Visual and nonvisual neglect afier unilateral brain lesions: Modulation by visual input. International Journal of Neuroscience. 6 1,229-239.
Van der Linden, M., Seron, X., Gillet, J., & Bredart, S. (1980). Heminegligence par lesion frontale droite. A propos de trois observations. Acta Neurolo~ca Belfzica, 80, 298- 3 10.
Van Essen, D. C., & Maunsell, J. H. (1980). Two-dimensional maps of the cerebral cortex. Journal of Comparative Neuroloev. 19 1 (2), 255-28 1.
Vogt, B. A. (1 984). Afferent specific localization of muscarinic acetylcholine receptors in cingulate cortex. Journal of Neuroscience. 4(9), 2 19 1-2 199.
Vogt, B. A., & Miller, M. W. (1983). Cortical comections between rat cingulate cortex and visual, motor, and postsu bicular cortices- Journal of Com~arative Neurolow, 2 l6(2), 192-2 1 O. -
von Giesen, H. J., Schlaug, G., Steinmetz, H., Benecke, R., Freund, H. J., & Seitz, R. J. ( 1994). Cerebral network underlying unilateral motor neglect: evidence from positron emission tomography. Journal of the Neurolopical Sciences. I25(1), 29-38.
Von Monakow, C. ( 19 14). Die Lokalisation im Grosshirn und der Abbau der Fwiktion durc h Cortikale. Wiesbaden: Herde.
Von Monakow, C. (1969). Diaschisis. In K. H. Pnbrarn (Ed.), Brain and Behaviour, Mood. States and Mind. @p. 37-36). Baltimore, Maryland: Penguin Book.
Wam, P. D., Gini, A., Tucker, D., Roeltgen, D., & Holmes, R. (1988). Changes in regional cerebral blood flow associated with recovery fiûm hemispatial neglect [Abstract]. TINS, SUD&
Watson, R. T., & Heilman, K. M. (1979). Thalamic neglect. Neurolom, 29(5), 690- 694.
Watson, R. T., Heilman, K. M., Cauthen, J. C., & King, F. A. (1973). Neglect after cingulectomy. Neurolow, 23, 1003- 1007.
Weiller, C., Willmes, K., Reiche, W., Thron, A., Isensee, C., Buell, U., & Ringelstein, E. B. (1 993). The case of aphasia or neglect after striatocapsular infarction. Brain. 1 I6(Pt 6). 1509- 1525.
Weintraub, S., & Mesulam, M. M. (1 987). Right cerebral dominance in spatial attention. Further evidence based on ipsilateral neglect. Archives of Neurolom. 44(6), 62 1 -625.
Weintraub, S., & Mesulam, M. M. ( 1 988). Visual hemispatial inattention: stimulus parameters and exploratory strategies. Journal of Neuroloay. Neurosur~erv & Psvchiatry, 5 1(12), 1481-1488. -
W iner, B. J. (1 97 1 ). Statistical Pnnciples in Ex~enmental Desim. (Second ed.). New York, USA: McGraw-Hill, Inc.
Woods, R. P., Mazziotta, J. C., & Cherry, S. R. (1 993). MN-PET Registration with automated algorithm. Journal of Cornputer Assisted Tomom~hv. 17(4), 536-546.
A P P E m K : TABLE OF CONTENTS
A . SLNNYBROOK NEGLECT BATïERY APPENDIX." ...H...HH........ ................................. -115
A . 1 . DRAWMG AND COPYING OF A DAISY AND A CLOCK ............................................................................. 115 A.2. LWE BISECTION TA% .......................................................................................................................... 116 A.3. LWE CANCELLATIONTASK ................................................................................................................... 117 .4.4. SHAPE C~LVCELLATION TASK ................................................................................................................ 118 A.5. SWYBROOK N E G L E ~ BAITERY SCORMG SHEET ........................................................................... 119 A.6. SUMMARY TABLES FOR SurwYBROOK NEGLECT BATTERY vs . VISUAL SEARCH BOARD ANALYSES .. 120
B . 1 . CT SCAN OF A P A T I ~ T WH HEMISPATIAL NEGLECT ......................................................................... 121 B.?. CT TEMPLATE OF A PAT~EXT m HEMISPATIAL NEGLECT ............................................................. 1 2 2 B.3. BREAKDOWN OF CT REGIONS BY STEREOTACTIC SLICES . QUANT~~ATION APPROACH ........................ 123
C . SPECT APPENDCX ................... ............. .............. ...................................................................... 124
C . 1 . ACUTE SPECT SCAN OF A RHD PATIENT WITH HEMISPATIAL N E G L E ~ ............................................ 124 C.2. SPECT SCAN ANGLE ROTATION .......................................................................................................... 125 C.3. EXAMPLE OF A CORTICAL RIM METHOD ............................................................................................... 126 C.3. EXAMPLE OF THE ROI METHOD ........................................................................................................... 127 C.5. MR-SPECT SUPERPOSITION ................................................................................................................ 128 C.6. BREAKDOW OF SPECT LOCALLZATION APPROACH ................................................. ................... 129
D . CORRELATION MATRICES .................................................................................................. 1 3 2
D . 1 . CT CORRELATION WTRIX . LHD ....................................................................................................... 132 D.2. CT CORRELATION U4TRIX - RHD ...................................................................................................... 133 D.3. SPECT CORRELATION WTRIX - LHD ................................................................................................ 135 D.3. SPECT CORRELATION ~ T W X - RHD ................................................................................................ 136 D.5. CT-SPECT CORRELATION ~ T R I X - LHD .......................................................................................... 137 D.6. CT-SPECT CORRELATION MATRIX - RHD .......................................................................................... 138
E . AXOVA TABLES FOR CT REGRESSION ANALYSES ................................................................. 139
H . COiMPLETE TABLE OF PLS SALXENCES .... ........................... .......................... ........................... 149
H . 1 . THE FIRST LATENT VARJABLE OF THE LHD GROUP ............................................................................ 149 H.Z. THE SECOND LATENT VARIABLE OF THE LHD GROUP ........................................................................ 150 H.3. THE THIRD LATE~T VARIABLE OF THE L m GROUP .......................................................................... 151
cxüi
H.4. THE FIRST LATM VARIABLE OF THE RHD GROUP ............................................................................. 152 H.5. SCREE PLOT FOR IMAGE SALIENCES LHD LV1 .................................................................................... 153 H.6. SCREE PLOT FOR IMAGE SALIENCES LHD LVZ ................................................................................... 153 H.7. SCREE PLOT FOR IMAGE SALIENCES LHD LV3 .................................................................................... 153 H.8. SCREE PLOT FOR IMAGE SALIENCES RHD LVl ...................... ... .................................................... 153
A. 5. Sunicybrook Neglect Battery Scoriig Sheet
f figuru Ornlsslon of any flpure on contralateral slde of DW. &ncdIation l a I
Normal performance:I l omlsslon 1 Scorc:I 1 pl. per omllted figure > I (mer.40 flgs.)
1 I 1 1-
A
A. 6. Surir nt a ry Tablcs For S m iiybrook Neglect Battery vs. Usual Search Board Aiaulyses
Segment Area Region ' Are8 1 Region Area 'Brodmana, Region Region Side SI-0 SFG. MFG F SFG F GFs 8 F A c h e R i ~ h t s 1-1 ReCG M Pre&PostCG : SM ZFm.GRC.GPâ 6.4.3.1 2 SM SM Mght s 1-2 POKG S SPL c P LPi M P P-Inf ' Right S 1 3 SPL. K U P SPL KU P LPS.PCU 7 P P_Suo Rinht SI3 SPL. K u P SPL. K u P LPs.PCu 7 P P-Sup Ltft SI-5 POSICG S SPL P LPi M P P I n f Left Si0 R K G M Pre&PostCG SM SFm.GPiC.GPoC 6.43.12 SM SM Left SI-7 SFG. MFG F SFG F GFs 8 F ACinn Left S 2 0 SFG F SFG F GFs 8 F ACing Rigbt S2-1 MFG F MFG F GFm 8 F F-Sup Rigbt s2-2 MFG F MFG ' F GFm 8 F F-Sup Right S 2 3 MFG F RcCG M GFm 9 F FSup Right S 2 4 M G M PostCG S GFm 9 F S M Rieht S2-5 PostCG S SMG P GnC 6.4 M S M Rigbt s2-6 SMG P SMG P GPoC 3.12 S P_hf R i ~ b t s2-7 SX1G P SMG P LPi 10 P PJnf Rigbt s2-8 SMG P SMG P LPI 40 P P 1nf W ~ b t s2-9 SMG P ' SMG P LPi 40.19 P P S U P Right s2 10 SPL P SPL P 0 . K u 19.7 PO P SUD R i ~ b t s2-11 SPL. K u P K u P 0.PCu 19.7 PO P S U P Wgbt S2-12 SPL. PCU P K U P O.KU 19.7 PO P-Sup Lefî S2-13 SPL P SPL P O .Ku 19.7 PO P s u p Left s2-14 SMG P SMG P LPi 40.19 P P-Sup , Left s2-15 SMG P SMG P LPi 40 P P-In f Left S2-16 SMG P SMG P LPi 40 P P-hf Lefî s2-17 SXtG P SMG P GPoC 3 - 1 2 S P Inf Left s2-18 PostCG S SMG P GPrC 6.4 M SM Left S2 1 9 Pr& G M P O ~ G s G F ~ 9 F SM Le f î S2-20 MFG F M G M GFm 9 F F-Sup Left s2-2 1 hlFG F MFG F GFm 8 F F SUD Left sz-22 M FG F MFG F GFm 8 F F-Sup Left S 2 2 3 SFG F SFG F GFs 8 F ACing Le ft
-
S 3 0 SFG F SFG F GFd 9 F AChg Right S3-1 hl FG F SFG F GFs 9 F F-Sup Rigbt S3-2 MFG F ,MFG F GFs 9 F F-SUP Rigbt S3-3 I FG F M G M GFm 9 F SM R i ~ h t s3-4 PrcCG M P m C G S GF i 44 F S M Rigbt S3-5 PostCG S SMG P GRC 6.4 M P h f Mgbt S 3 6 SAMG P SMG ' P GPoC 3.12 S P I n f Rigbt s3-7 SMG P A G 1 P LPi 40 P P I n f Rigbt s3-8 SMG P AG P LPi. Gsrn 40 P P S U D R i ~ b t s3-9 AG P AG P ' Ga 39 P P-sup Rigbt
S3-1 O SPL P SPL P I G O S 19 O MedO Rinbt S3-11 SPL. PCU P CU O 1 G0s.c~ 19 O MedO Le ft S3-12 sPL.PCu P CU O ' G0s.c~ 19 O P SUD Left S3-13 SPL P SPL ' P i G O s 19 O 1 P-SUP Left s3. -14 AG P I A G ' P I Ga 39 P P Inf Lefî s3-15 SMG P AG P LPi.Gsm 40 ' P 1 P-hf Left S3-16 SMG P ~ A G P I L P ~ I ~ O ~ P , P-Inf Left s3-17 SMG P SMG P ' GPoC 1 3 . 1 2 s SM Left S3-18 P O K G S , SMG i P GRC 6.4 M 1 SM Left s3-19 P d G M PostCG 1 S 1 GFi 44 I F 1 F-Sup Left
C.6. Breakdown of SPECT Localization Approach - conf.
Segment Area Region Are. ! Region Are. Broâmann~ Region Region Side s3-20 ! FG F ReCG !bl GFm 9 F F SUD Le ft S3-2 1 .M FG F MFG I F GFs 9 F F-Sup 1 S 3 2 2 MFG F SFG F GFs 9 F ACing 1 S3-23 SFG F SFG F GFd 9 F F-Sup R - 0 SFG F SFG ' F GFs 10 F ACing Il s4-1 MFG F SFG. MFG F GFm 1 O F F-Sup R 53-2 hl FG F MFG F GFi 46 F F-Sup Il S4-3 I FG F M G M GFÎ 45 F F-Sup Il 9 - 4 ReCG M PosCG S GFi 44 F SM R s4-5 PostCG S SMG P GK-GPoC 6-4-53 SM SM R
eft
igbt
ight
5 4 6 SMG P SMG P LPi 40 P P-hf Rigbt 53-7 SMG P AG 1 P GTs -- 7 7 T P-hf Right s4-8 SMG P AG P GTsLPi 22.39 Pl- P-hf Right a-9 AG P AG P GTm 19 TO P-Sup Right 54-1 0 SPL P SPL P GO^ 19 O Lat0 Right 54-1 1 CU O CU O GO~,CU 18 O MedO Right S4-12 CU O CU O G O ~ C U 18 O MedO Left S4-13 SPL P SPL P GOm 19 O Lat0 Left S C 1 4 AG P , AG P GTm 19 TO P SUD Left 54-1 5 SMG P AG P GTS.LP~ 22.39 PT PJnf . Left a-16 SMG P AG P G T s -- 17 T P-hf Le ft S1-17 SMG P SMG P L P ~ JO P P-1nf Left s3-18 PostCG S SMG P GfK.GPoC 6.1.13 SM S M Left S4-19 PreCG M PosiCG S GFi -44 F SM Left s4-20 1 FG F M G M GFi 45 F FSup Left
MFG F MFG , F GFi 46 F F S U D Left S4-22 MFG F SFG-MFG F GFm 10 F F-Sup Left !M-23 S FG F SFG F GFs 10 F ACing Lefi SS-0 SFG F SFG F GFs 1 O F ACing Mgbt S5-1 hl FG F MFG F GFm 10 F F-hf Rigbt SS-2 1 FG F .MFGJFG : F GFi 10.16 F F-In f Right SS-3 PrcC G M [FG.ReCG . M GFi 45 F F-hf Rigbt SS-4 PosrCG S PostCG S G K 44.6 M SM Rigbt SS-5 STG T STG T GTT 6.42 T Temp Rigbt SS-6 STG T STG T GTS 4222 T Tem p Right ss-7 AG P AG P GTs - 7 7 T P-1 nf Right ss-8 AG P AG P GTm 39 PT P-hf Mgbt SS-9 LOG O lat OG O GOm 19 O Lat0 Right
S5-10 LOG O fat OG O GOm 18 O Lat0 Right SS-Il CU O , LG.CU O Ci0m.C~ 18 O MedO Rieht ss-12 Cu O LG.Cu O GOm,Cu 18 O MedO S5-13 LOG O lat OG O GOm 18 O Lat0 S5-14 LOG O ' latOG O GOm 19 O Lat0 s5-15 AG P AG P GTm 39 PT P-hf ! s5-16 AG P AG P GTs 22 T P-In f
.-
S5-17 STG T SïG T , GTs 4222 , T Temp
ss-18 STG T STG ' T GTT 6.42 . T r e m p S5-19 PostCG S PostCG S i GRC 44.6 ' M SM '
1 SS-20 RCG M IFG.R~CG' M ! GFI i 45 ' F , F-1.1 Left 1
C- 6. Breakdown of SPECT Localkation Approach - cont.
Cori Rim Darnasio O De1p1ees : Damasio 15 Degrees Tdaimch-Toumow O k e e s , Sclectcd 1 Segment Area Region Area Region Arta Brodmanni Region - Region SMe
S5-2 1 IFG F MFG.iF G F G E 10.46 F F-Inf Left SS-22 MFG F MFG F GFm 10 F F-Inf Left s5-23 SFG F SFG ' F GFs 10 F ACing Lef i S6 0 FPolc F SFG F GFs 10 F ACinn R i ~ h t s6-1 MFG F MFG F GFm 10 F F-hf Right S6-2 1 FG F MFG.iFG ' F GFi 46 F F Inf Rinht S6-3 1 FG F 1FG.RcCGi M GFi 45 F F-hf Right S6-4 STG T POSCG S GTs 7 7 T Temp Right S6 5 STG T STG T GTs - 7 7 T Temn Rinht S6-6 STG T STG T GTm - 7 7 T Temp Right s6-7 MTG T MTG T GTm 2 1 T T e m ~ Rinht S6-8 AG P MTG T GTm 37 PT E T Right S6-9 LOG O IÎL OG O GO^ 19 O L a t 0 R i ~ h t S6-10 LOG O ~ a t OG O ciom 18 O L a t 0 Right s6-11 LG. Cu O LG. CU O GOmCu 17 O bled0 Wght S6-12 LG. CU O LG. CU O GOmCu 17 O MedO Loft S6-13 LOG O ht OG O GOm 18 O L a t 0 Left s6-1 4 LOG O Iat OG O GOm 19 O L a t 0 Left S6-15 AG P MTG T GTm 37 PT Temp Left , s6-16 MTG T blTG T GTm 2 1 T Temp Le ft S6-17 STG T STG T GTm 22 T Temp Left s6-18 STG T STG T GTs - 7 7 T Temp Le f t s6-19 STG T PostCG S GTs - 17 T T e m ~ Left S6-20 IFG F IFG.PreCG : M GFi 45 F F-hf Le f t s6-2 i IFG F MFGJFG F GFi 46 F F h f Left S6-22 MFG F ,UFG F GFm 10 F F-hf Left s6-23 FPolc F SFG F GFs 10 F A C i n g k f t s7-0 FPok F ' SFG F GFs 10 F ACing Right S7-1 IFG F MFG F GFrn 1 O F F-1 n f Wght s7-2 I FG F iFG F GFi 1 O F F-1 n f Right S7-3 STG T 1 FG.PrcCG M G Fi 47 F F-hf Right s7-4 STG T STG T GTs 22 T Temp Right S7-5 STG T STG T GTs 2 1.22 T T e m ~ Rinht S7-6 MTG T MTG T GTs 21 T Temp Right
LITG T MTG T GTrn 2 1 T Temp Right s 7-8 MTG T MTG ' T GTi 37 PT Temp Right S 7-9 LOG O Lat (Xi 0 ci01 19 O L a t 0 Right
s7-10 LOG O Lat OG O GOi 18 O L a t 0 Right S7-11 LG. Cu 0 Cu 0 G0i.c~ 17 0 MedO Right S7-12 LG. CU O CU O G01.cu 17 L O MedO Left S7-13 LOG O ht OG O GO1 18 O L a t 0 Le ft
LOG O Iat OG O GOi 19 O L a t 0 ~ e f t MTG T MTG ' T GTi 37 PT T e m ~ k f t
S7-16 MTG T MTG T , GTm 2 I T Temp Le ft s7-17 X1TG T MTG : T GTs 2 1 T Temp Left s7-18 STG T STG T GTs - 71 - 77 T Temp Left S7-19 STG T I STG T ' GTs 22 T Temp Left s7-20 STG T ' 1FG.PreCG ' M ' GFi ' 47 F F-1 n f kft S7-2 1 IFG F , IFG F GFi ' 10 F F-t nf Le ft S7-22 IFG F MFG F 1 GFrn 1 10 F F-hf k f t s7-23 FPole F SFG F 1 GFs 10 , F ACing k f t
Topognphy of Hemispatial Beglcct
D. Correlation Matrices
D. 2. CT Correlation Mat& - LHD
CIU3
-265 -0::
.378
. OGO
.O94
.2 12
- 502 .O00
-527 .O00
:.cou
-. 052 -217
. $07 -000
-680 .O00
-101 .O61
.?O4 -00;
.6 11
.O00
- .28$ .O01
-.136 -122
-356 -00:
S N S O R Y O C I P E A L TEKPüRAL E S L A R U S
CT Correlation Matth - M D
Correlation, 1 - tailed Sig:
N a M G
AGE
VOLUME
ArL'ITC I N G
iINTWM
BG
CRAD
DEEPTPO
FRONTAL
MOTOR
OCIPITAL
PAR 1 ETAL
P0SrnvW
SENSORY
TEM W RAL
THALAMUS
AGE
-203 .O13
1.000
- -267 .O02
- -116 -103
- -127 . O83
- -248 . O03
- -130 .O78
- .O18 -4 24
- -128 -082
- .O54 -278
- .O37 -346
- -125 . O88
- -186 . O21
- .O56 -27 1
- . O83 -184
- - 059 -262
Page 133
D.2 CT Correlation Munir - W . - c0n.t
. * . .
PJaLOG
AGE
VOLUME
M i T C 1 NG
A P m
BG
CRAD
CEEPTPO
FRONTAL
MOTOR
O C I PITkL
PAF1 1 =AL
POSrn'M
SZNSORY
TEMPORAL
TFSUA-WS
M U L T
FRONTAL
-128 .O81
- -128 .O82
- 560 . O00 -727 . O00 -673 .O00
-186 .O21
-348 . O00
- .O95 -151
1.000
-573 . O00
- -164 . O36
.181
. O24
-087 .172
.404
. OOG ,120 . O9 5
- . 070
I P L E R E G R E S S I O N ' * * *
MOTOR O C I PITAL PARIETAL
Page 134
SPECT Correhtion M a h - LHD
=:CG
l .CO0
- 5 4 1 - 0 0 0
- . 094 .2CO
- .O13 . I 6 2
- . cg7 -233
- .12E . :6E
- - 1 5 5 - 2 7 0
- .O71 - 2 9 7
- . O44 . 2 I C
. . C I € .365
- . 128 - 1 6 7
. ! < G . - d d . -. . - L - L ?T
- . 094 - 2 4 0
- .299 -0::
.513
.OC0
.62E
. ûPO
- 6 2 1 - 3 0 0
.;O1 - 0 0 0
-7:: . OC0
- 8 4 6 . O O O
1. CO0
.64E
.O00
- 7 4 7 - 0 0 0
-73; . GCO
V O L L ?
- SC :. . 000
:. O00
- .O95 - 2 3 7
.O88 -253
- .O87 - 2 5 6
- - 037 - 5 9 0
- .306 . O09
- - 1 5 5 . O70
- - 2 9 9 .O11
- - 163 . I O 9
- - 2 9 6 - 0 1 1
- . i 0 5 .215
Lsn
- . OC6 .365
- .163 . 109
-703 .O00
. 7 4 7
. OOC
.870
. O03
- 6 6 6 . O00
- 4 0 8 .O00
.a05
.O00
- 6 4 8 .300
1 .000
.763
.O00
,726 .O00
I Z G
- - 0 9 4 - 2 4 O
- .O95 - 2 3 7
1.000
.77 8
. O00
- 8 0 5 -009
- 4 6 9 . O00
. 2 5 7
.O25
- 5 9 6 .O00
- 5 1 3 .O00
- 7 0 3 .O00
- 7 0 0 .O00
.a02
.O00
LTEK?
- . 1 2 8 . 1 6 7
- - 2 9 6 . O 1 1
- 7 0 0 . O00
.61C
.O00
- 7 4 8 . O00
- 5 6 7 . O00
,466 . O00
.685
.ooc
- 7 4 7 .O00
- 7 6 3 .O00
1 .000
.696
.O00
SCING
- .O13 .CO2
. 088 - 2 5 3
- 7 7 8 .O00
1.000
- 8 6 6 .O00
- 7 7 5 .O00
- 4 9 1 - 0 0 0
- 7 6 1 .O00
- 6 2 8 .O00
- 7 4 7 .O00
.6 14
. O00
.862
.O00
LTH
- - 1 5 5 . 1 2 1
- - 1 0 5 .215
- 8 0 2 .O00
- 8 6 2 .OOG
- 7 7 5 .O00
- 7 8 2 .O00
- 5 3 0 .O00
- 8 0 9 .O00
.734
.O00
.726
.O00
.696
.O00
L.000
SO
- - 1 9 5 .070
- .!O6 . 009
- 2 5 7 - 0 2 5
- 4 9 1 .000
- 4 8 2 -306
- 794 . 000
1.000
.656
.000
.7 11
. O00
.468
. O00
-468 . OOC
- 5 3 0 - 3 0 0
Li?
- - 6 7 1 - 2 5 7
- -195 . O70
- 5 5 6 .000
.76 1
.000
.752
.c00
- e 17 .000
. € 5 6
.000
1.000
.846
.000
. eos
. O00
- 6 8 5 .000
.a09 - 0 0 0
Page 135
SPECT Correlation Mu& - RHD
C c r r e L a t ~ o n , l-talle& S i g : NE LOG
1 - o o c
. 197 - 3 3 3
- -- - 2 0 ,
- OOC)
- - 3 8 2 - 3 0 0
- .19E .O32
- - 2 3 3 - 3 14
- - 2 6 3 . GO7
- . ? 4 2 -90:
- -232 .O01
. .?GO . O00
- .264 . O07
- - 3 3 9 - 2 3 1
- .323 -89:
3 P
- - 3 3 2 . O01
- 1 9 7 .O33
- - 2 8 6 . CO3
-6 15 - 0 0 0
.0:7
. GO0
- 7 3 7 .O00
- 6 4 5 . o o c
- 6 3 6 .O03
1 - 0 0 0
- 7 2 2 . û00
. E 1 5
.O00
- 7 6 6 . OC0
- 6 4 7 .O00
AGE
.197
.O33
1 .000
- ,321 . g o 1
.O26
. iOS
.O50 - 3 2 2
. OS4
.?O8
- 1 3 4 .LE7
-238 .O26
. '9 7
.O33
- 0 4 8 - 3 2 7
- 1 1 5 . .. . -..-. .O02 - 4 9 4
.O78 - 2 3 6
RPT
- - 3 8 0 .O00
. O48 - 3 2 7
- - 1 7 2 - 0 5 5
- 4 5 6 - 2 0 0
- 4 0 0 - O00
- 5 14 .O00
- 4 2< .O00
.598
. O00
* 7 2 2 . O00
1 .000
- 4 7 3 . O00
- 8 9 9 . O00
- 4 7 8 .O00
VOLCJIiO
- 3 6 7 .O30
- -32: .O01
i . 0 0 0
- . 346 . O 00
- .O53 - 3 1 3
- - 1 0 6 - 1 5 8
- -111 - 1 5 2
- - 1 5 5 .O75
- - 2 6 6 .O03
- - 1 7 2 - 0 5 5
- - 3 7 0 ,000
- - 2 5 1 . O09
- - 2 4 4 .O11
RSM
- .264 - 5 0 7
.115
.TC4
- - 3 7 0 . O00
- 5 2 7 . O00
- 5 1 0 .O03
- 6 7 5 .O00
. 4 5 1
. O00
. 4 0 1 * O00
. a 1 5
. O00
- 4 7 3 .O00
1 .000
.621
.O00
.O88 . O00
RC ING
- , 1 9 8 - 0 3 2
- 0 5 0 - 3 2 2
- .O53 - 3 1 3
- 6 6 9 . o a o
1 .000
- 7 7 6 .O00
- 7 2 8 . O00
- 5 5 4 . O00
- 6 1 7 . O00
- 4 0 0 .O00
- 5 1 0 .O00
- 3 3 2 .O0 1
- 7 4 6 ,000
XTH
- - 3 2 3 .O01
. O76 -236
- - 2 4 4 -0::
- 7 7 5 .O00
- 7 4 6 .O00
- 6 3 9 .O00
- 7 6 3 . O00
-521 .O00
- 6 4 7 - o c 0
- 4 7 8 .O00
- 4 8 6 - 0 0 0
.cc:
.O00
1.000
Page 136
D. 5. CT-SPECT CorreIation Mut* - LHD
C o r r e l a t i c c , 1 -:azled Sig:
NJLOG
1- O00
-541 . OOC
- -091 ,279
- -303 -491
- -116 -216
- .O75 -257
- -358 -347
.25e
. O2G
.2C3
. O83 -425 .O01
-302 . a 19
-053 -265
CTSG
.20?
. CE3 -235 . O10
- -492 .O00
- -225 .O62
- .293 . 3 22
- -25.4 . O1 1
- .226 . C6 1 -164
7 -3-3
1.000
. ? 05
.O16
.2i5
.O25
.529
. go0
LBG K I N G
Page 137
VOLUME
-332 .O0 2
'-000
. 5 64 -000
.392
.O00
.590
.O00
-560 .OC0
.144 -199
- -305 .O04
.Ot7 ... .....-. - . O34 -386
- .227 .O25
- .19C . o;e RF
- .123 -147
- .O34 -386
- .oe5 .234
-205 .O39
- -139 -116
- .O33 -391
.O71
.272
.674
. O00
-736 . O00
1.000
.67G
.O00
.553
.O00
CTaG
.2e7
.O06
-392 -000
-245 . 0 17
1-000
-229 -02;
-145 -107
-435 .O00
- -212 .O34
-134 .226
.205 -039
.O74 -265
- .O36 -378
RTH
- -254 .O14
- -194 -048
- .O23 -423
- .O36 -378
- .O60 .3 04
- -105 - 2 8 5
- -204 .O40
.731
.O00
.667
. O00 -553 . O00 -537 -300
1.OGO
Page 138
E. ANOVA Tables for CT Regression Analyses
E. 2. CT A N 0 VA Table - Exploratory Metliod - LHD
* * * * M U L T I P L E R E G R E S S I O N " ' *
Multiple R -51882 R Square .26917 Adjusted R Square -25916 Standard Error -4 129 1
Analysis of Variance DF Sum of Squares Mean Square
t t t t M U L T I P L E X E G R E S S I O N * * * *
M u l t i p l e R .50000 R Square -25000 Adjusted R Square -23060 Standard Error -50446
iinalysis o f Variance D F Sum of Squares Mean Square
Reqress ion 3 9.83975 3 -27992 Residual 116 29.51998 -25448
F = 12.88857 Siqnif F = -0000
- - - - - - - - - - - - a - - - - - - - - - Variables in the Equation - - - - - - - - - - - - - - - Variable B SE B 95% Confdnce I n t r v l B
AGE -012618 -003296 . 006089 -019146 V O L W -002256 9.9583E-O4 2.83599E-04 . 004228 P O S m i . 006997 . 002539 . 001968 . 012027 (constant . 096707 -246385 - -39 1290 - 584705
Appendix H. 1: Complete Table of PLS Saliences for the First Latent Variable of the LHD Croup
sUbTe8t S~ôte8î -1. Une Bire«ion -0-0 175 UnetCincelktion O. 396û -Sb- C8ncdktiori 0.6092 A
Dmwlngs O. 6864 Segment Side Anatomy Image Saliencar- Seg. Skie Anatomy Image 9 1 . çs0. Side * An8tomy- lm Sa
Lefî P-SU~- Lefî P-Sup Lefî Lat0 Left P-Sup Left P-lnf Lefi Lat0 Lefî P-Sup Left P-lnf Left Medû - Left Me& Right P-Sup Left P-Sup Right P-Sup Right M e d û Right SM Right P-Sup Lefi P-lnf
Right Medû Right Temp Lefi Temp Left P-Sup Right P-lnf Right P-lnf Right ACing Left LatO
Right Temp Right F-lnf Left P-lnf Left TH .
Rig ht P-Sup Left P-lnf Left P-lnf Left Temp Left TH
Right ACing Left LatO Right P-Sup Left P-lnf
Right ACing Right Temp Right P-lnf Left ACing
Right F-Sup Right F In f Right MedO Right SM Left ACing Left F-Sup Left Temp
Right F-Sup Rig ht F-lnf Right Temp Right SM
9-10 Right LatO -0.0835 S6-11 Right MedO - -0.0683 - S4-7 ~ i g h t - P-lnf -0.0821 S6-23 Left ACing -0.0669
S4-2 Right F-Sup -0.0821 S3-18 Left - SM -0.0669 S5-8 Right P-lnf 4.082 52-1 8 Left SM -0.0669 S6-17 Left Temp -0.082 S2-21 Left ~ 3 u p -0.0664 : S2-6 1 ~ i g h t -
- 55-3 Right' S7-3 - Right. : S7-0 : Right
- S5-9 Right; S7-4 Right
j S5-22 Left : S6-8 Right- S2-1 Right SI-1 Right - 55-23 - Left -
SS-1 Right - S c 1 ' Right- - 54-9 - ~ i g h t - S2-23: Left '
52-2 Right- S3-19 Left S2-7 . Right' S5-10- Right S6-14- Left S2-16 Left : S7-8 ' Right S2-19 Left -
bg r l Right : S5-4 : FIight- S3-7 Right. th rO 1 Right. S2-3 . Right
' S5-2 Right . S5-17 Left
1 S6-7 ~ i g h t - S2-20, Left S6-O Right
- S6-9 ' Right . : S4-5 ' Right 1 9 - 8 Right S2-4 Right
- S7-6 . ~ i g h t ' . 53-2 . Right' S6-22 Left .
S4-6 , Right. th r l Right'
' ~6-12' Left S6-18 L e y
' S7-19. Left S3-5 : Right. S4-19, Left S5-5 Right
- . P-lnf F-lnf F-lnf
ACing
Lat0 Temp F-lnf p-T
KSUP . F-Sup - ACing F-lnf F-lnf
P ~ U P ACing F-SUP
SM P-lnf - LatO LatO . P-lnf Temp SM BG SM
P-lnf TH
F-SUP A
F-lnf Temp Temp F-Sup ACing LatO ,
SM P-lnf SM
Temp .
ESup - F-lnf P-lnf TH
MedO Temp Temp SM SM
Temp
S2-22 1 Left : S3-1 Right
' ~7-16 1 Left 1 S7-2 Right S3-6 Rig ht S6-4 ~ i g h t -
' ~7-17 Left 54-17 Left S6-13 Left
1 51-3 1 ~ i g h t ' Si-O Right
- Sl-1 ~ i g h t ; S I 4 . Left -
S3-23 Left SI-2 ~ i g h t . S2-17 Left S4-18 Left si-22. Lefi S3-17 Left _ Si-7 Left S7-9 . ~ i g h t - S3-21 Left S7-7 Right : bg rO : Ftight : bg 10 Left
53-20 Left S3-22 Left SI-5 Left S5-18 Lefl S7-12 Left Si-6 . Left S7-21 Left S7-15 Left
1 SS-i 9 : Left I S7-13 Left . S6-21 Left bg II Left
' ~7-14 Left 2 1 Left . 85-21 Left S7-10 Right S4-20 Left S7-20- Left : S6-19. Lefl .
SubTert A S u M W SII. Dirwings -0.2828 - Une Cincelktion 4.0557
- Shape Cancelktion 0.3803 Une Biseaïon 0.8788
Segment Side ~natomy-image Satiences- Seg. Si& Anatomy Image al. Seg. Side Anatomy Image 5.1 S5-6 Right Ternp -0.0194 S6-12 Left MedO 0.0417 th 10 Left - TH
:ç3_i6 Lef t P-lnf . S c 4 Right SM S5-O Right. ACing S6-13 Left - LatO ~3-14 Left P-Sup S c 6 'Right, P-lnf
1 S4-2 Right F-Sup , bg r l Right] BG S2-16 : Left P-lnf -
SS-1 Right F-lnf S4-5 ~ igh t : SM :
- S<I 6 Left - P-~nf - ~2-18 1 Left SM .
S6-10 Right- LatO 1 S5-4 Right SM
S4-1 Right F-Sup .
S7-10:~ight1 LatO 53-1 2, Left , Medo 53-21 Left F-Sup S5-10Right LatO SC23 - Left _ ACing S7-14- Left LatO S2-15 Left . P-lnf
:s~-Is: L e t p-~nf 1 Sô-4 Right Temp . S5-9 Right MedO .
S5-5 ~ i g h t Temp S7-18 Left Temp S6-9 - Right . LatO S5-15 Left P-lnf bg rû Right BG
' S4-û Right ACing S2-14. Left . P-Sup -
S3-6 Right P-lnf S2-5 Right SM S3-4Right SM . 57-8 ' Right Ternp ~ 3 - 2 -Right F-Sup thrû Right TH -
S2-4 Right SM S2-21 Left F-Sup S2-3 Right. F-Sup ,
S5-11 Right- MedO Si-1 Right SM S3-11' Right ] MedO
. 52-6 Right P-lnf .
S6-11 Right MedO . : S6-15 Left : Temp -
S6-5 Right- Temp - S5-16 ' Left . P-lnf
S i 3 Right S3-O Right S6-8 Right S7-9 - Right SS-7 Right S7-3 Right S4-9 Right S6-6 Right S c 7 ' Right : ~ 1 5 : Left j S2-2 - Right S2-10. Right 1 S7-12 Left S3-9 . Right S5-8 Right SI-5 Left S4-11 Right 53-22 Left : S3-5 Right 3 Right - S6-16 Left : S2-7 : Right - S7-11 Right' S4-10 Right 1 S2-13 Left - S7-13 Left 57-15 Left ,
S7-7 Right S2-22 Left S3-7 Right S6-7 Right . S3-10 Right SI-6 Left Si-2 Right S2-9 Right : SI-4 Left SI-O Right -
S3-8 Right S2-1 Right S7-4 Right
S7-16 Left . S2-12 Left S4-8 Right S2-O , Right ,
S3-23 Left S2-8 . Right , S7-5 Ri'ht SI-7 , Lely S2-23. Le/t 57-17 Left S7-6 Right '
p-sup ACing p-T ,
LatO P-i nf F-lnf
P-SUP Temp P-lnf P-l n f ESup - P-Sup Medo - P-Sup P-lnf P-lnf MedO F-Sup
SM F-Sup .
Temp P-l nf MedO Lat0
p-sup LatO Ternp Temp E S u p P-I nf Temp p-sup .
SM P-I nf
P-Sup -
p-sup ACing P-lnf F 3 u p - Temp Temp F S u p .
P-lnf ACing ESup P-1 n f T-P- . ACing
- T-P .
T ~ P S2-11 Right P-Sup 0.0389 ,
. -
S5-2 Right F-lnf ' th r l * ~ i q h t - TH 0.0408
Page 150
Appendix H.3: Complete Table of PLS Saliences for the Third Latent Variable of tbe LHD Group
Segment Side Anatomy Image ~aliencas- Seg. Side AMtomy Image SaI. Seg. Si& ' Anatomy I m ~ e Sal 52-23 Left ACing bg rl Right BG
~ i s h r Right Right Left Left Left Left Left Left Left
Righ t Righ t Right Right Right Right Left
Right Le ft Le ft
Right Right Right Le ft Leit
Right Le fi Le ft Left
Right Right Left
Right Left
Right Right Right Right Left
Right Left Le ft Left
Right Left Le ft Left Le ft Left Le ft
Right
SM P-lnf TH
Temp Temp LatO SM
P-SUP SM
Temp P-lnf
P ~ U P F-s u P P-1 nf F-Inf F-I nf LatO
F-sup P-I nf
p-sup Tem p LatO P-I nf Temp FSJp p-T P-lnf ACing P-lnf SM
p-sup p-sup LatO F-I nf Temp P3up ACing
SM Temp F-I nf
F-SUP P3up
SM P-lnf p-s u p F-lnf SM SM SM
Temp Temp
] S&l l j Right MedO # S7-21 Left F-tnf
] 53-8 : Right P-lnf S2-1 , Right. F-Sup S3J , Right ' ACing -
- S5-5 .Right- Temp S3-9 Right- P-Sup
- S2-3 1Right- F-Sup -
S2-8 .Right- P-lnf -
bg 11 Left BG 1 S6-16- Left 1 Temp S c 1 0 Right Lat0 - S2-O I ~ i g h t - ACing S5-4 Right- SM S5-22 Left F-lnf S5-6 - Right Temp S7-15 Left Temp S2-2 Right. F-Sup - S2-19 Left , SM S3-17 Left P-lnf S5-23 Left ' ACing : S5-10 Right - LatO : ~5-13: ~ e f t LatO S3-5.Right SM
-S4-f 1 Right- MedO -
S4-3 Right F-Sup S2-5 R i g h t SM S3_6 ~ i g h t P-lnf -
S7-13. Left LatO S6-13, Left , LatO S6-21. Left F-lnf : SI-2 Right P-lnf -
S c 6 1Right' P-lnf -
S5-2 Right. F-lnf bg IO Left . BG
S2-10. Right P-Sup S7-8 ~ i g h t Temp S7-19. Left ' Temp S7-3 ~ i g h t ' F-Inf '
S5-16 Left ' P-lnf '
1 ~6-22. Left ' F-lnf .
~ 7 - 6 . Right Temp ' ~ 1 2 ' Left : Lat0 . S2-16 Left , P-lnf SS-21 Left , F-lnf S c 2 ' Right F-lnf - S5-O ' Right: ACing S2-7 , Right P-lnf S7-5 ~ i g h t ' Temp S2-9 ' Right P-Sup S4-21' Left F-Sup ~5-20' Left , F-lnf S5-7 Right P-lnf
-0.0126 ~5-11 . Right -0.0126 S4-5 Right -0.0122 S3-23' ~ e f t -0.01 18 S-20 : Left : -0.0114 S3-16- Left -0.01 13 S U - Right 1 -0.01 1 th I l Left -0.01 03 S6-5 - Right 1 -0.0094 54_16- Left -
-0.0083 S4-2 Right -0.0063 S7-9 Right 4.0057 S6-6 Right -0.0057 S5-3 , Right -0.0057 S4-19 Left -0.0051 S7-18 Left -0.004 S I 4 . Left -0.0001 S7-12 Left .
0.0004 . S4-17 Left 0.002 S3-2 Right 0.0036 , S3-22 Left 0.0057 S4-O Right 0.0064 S5-1 Right 0.064 52-17 Left 0.0089 S5-9 Right 0.0097 ' S3-12 Lefî 0.0109 S7-17 Left 0.0136 56-1 7 Left 0.01 37 bg rO - Right ' 0.01 56 53-3 Right 0.0168 S7-4 Right 0.01 82 55-12 ' Left . 0.0194 S7-14, Left -
0.0203 S3-14 Left 0.0205 S7-11 Right 0.0216 S3-11 Right- 0.021 8 S6-14 1 Left 0.0222 . S4-22 . Left 0.0224 thrO .Right 0.0245 S6-23. Left +
0.0266 th lO . Left ,
0.0267 S4-15 Left . 0.0288 S3-4 Right 0.0298 SZ-22 Left '
0.031 7 S6-12, Left - 0.032 57-20 Left -
0.0342 S3-21 . Left 0.0344 . S7-1 Right 0.0356 SI-7, Let! 0.036 . s2_ r i aight
0.0378 S7-2 Right j 0.038 S2-21 Lefl 0.0385 S7-23 Len 0.0389
MedO SM
ACing F-sup P-lnf ACing
TH Temp P-lnf F3up LatO Temp F-lnf SM
Temp P-SUP MedO P-l n f
F ~ U P F S J P ACing F-lnf P-lnf LatO
MedO Temp Temp BG
F-Sup Temp MedO LatO
t s u p MedO MedO Lat0
F-Sup TH
ACing TH
P-lnf SM
F B J p MedO F-I nf
ESup F-I nf ACing p,Sup F-/nf
F,Sup A Cing
S6-9 ~ i g h t ~ a t 0 -0.0201 S2-15 ' Left P-lnf 0.041 6
Page 151
Right Right Right Right Right Right Right Right Right Right Right Right Right Right Right Right Right Right Right Right Right Righ t Right Right Left Right Right Right Right Right Right Le ft
LatO S5-1 Right F-lnf -0.0752 S6-14 Left Lat0 -0.0566
Page 152
56-13 Left LatO : Sô-4 a Right S3-3 . Right. S6-23. Left . a 4 Right
157-13 Left S5-4 Right S3-5 ~ i g h t - S5-5 Risht
- S2-5 ~ i g h t - th rO Right
: ~ 3 _ i 4 Lett S2-10 Right S6-2 , Right
. SC3 , ~ i g h t : th r i Right S2-4 , Right
' S3-1 Right. : ~3-13: Left : . S2-14 - Left .
56-3 - Right S2-12 Left S I 4 Left S6-12 Left S6-0 Right 3 -6 Right- 57-4 'Right
. S7-11 ~ i g h t S3-11 ' ~ i ~ h t '
1 5 2 2 ~ igh t . S3-20 Left
- S4-1 ' Right Si-O ~ i g h t - S4-19 Left S2-22 Left S4-20 Left S7-3 Right S2-3 Right SI-7 Left S3-15 Left S4-13 Left
Temp F-Sup ACing SM
LatO SM SM T
SM TH
P-SUP P-Sup F-l n f
ESup TH SM
F-Sup P-sup p-Sup F-lnf
p-sup p-sup LatO
ACing P-lnf Temo MedO MedO F-Sup ESup ESup ACing
SM ESup F-Sup F-1 nf
ESup ACing P-I nf LatO
. s4-14 1 Left 1 SI-5 Left : S6-1 ~ i g h t - S6-15 Left Si-1 Right S3-12 Left -
S7-14 Left S4-5 Right. S3-22, Left S4-15- Left ,
S2-11 Right - S4-22' Left
Appendix H.4: Complete Table of PLS Saliences for the First Latent Variable of the RHD Croup
SubTert - Subtesî Sal. Olrwings 0.6382 -
Une Biisction 0.4836 _ Cine Clncellrtion 0.4426 -
Shrpe CInc+tion 0.4036 -
Segment Side Anatomy . Image Saliant& Seg. S i g Anatomy image 5.1 .
1
1
1
1
l
l
l
t
l
!
t
4
! d
4 . 4 . !
c . c
< a
:
1
Seg. ' Side - Anatomy - Image Sa1 S3-19 Left SM -0.0565 S3-18 Lef! SM -0.0562 S5-22 Left F-lnf -0.0558 S3-17: Left P ln f - -0.0555 SI-2 Right- P-lnf - -0.0535
S2-15 Left P-lnf -0.0534 S3-16. Left P-lnf -0.0532 S5-15 Left . P-lnf -0.0526 S2-21 Left . F-Sup , -0.052 ~5-20- Left F-lnf -0.052 SS-19' ~ e f t SM -0.051 9 S5-14 1 ieft ' LatO . -0.051 5 S2-18 Left SM -0.051 1 S I - Left SM -0.0509 S2-20. Left F-Sup -0.0505 SI-3 .Right' P-Sup ' -0.0499 S7-15, Left Temp -0.0488 S4-16, Left P-lnf -0.0486 S4-21. Left , F-Sup -0.0486 S3-21. Left F-Sup -0.0479 52-17 Left P-fnf -0-0461 S4-17 Left P-lnf -0.0458 55-21 Left F-lnf -0.0457 S5-16 Left P-lnf -0.0456 56-22 Left F-lnf -0.0439 57-1 2- Left MedO , -0.0424 52-1 6 - Left P-lnf -0.042 52-19. Left ' SM - -0.042 S4-18.Left' SM - -0.041 4 bg IO , Left BG - -0.0412 S6-21,Left F-lnf -0.0398 S6-16 Left Ternp , -0.0397 S7-23 ' Left ACing -0.0393 S7-2 Right i l n f -0.0379 th 10 Left TH -0.036
56-20 Left F-ln f -0 .O349 35-17, Left Temp -0.0346 36-19 Left Temp -0.0322 35-1 8. Left Temp - . -0.0306 57-16, Left Temp -0.03 56-1 8- Left , Temp -0.0292 bg11 Left. BG -0.0255 S7-1 . Right F-lnf -0.0254 37-20' Left : F-lnf ' -0.0254 37-19 Left Temp -0.0227 36-17 Left Temp -0.0215 th11 Left- TH -0.01 94 37-18: Left Temp : -0.0156 S7-O , Right, ACing -0.01 25 57-17. Left Temp -0.0062 57-22 Left F-lnf ~. -0.0046 S7-21: Left F-lnf 0.0039