Functional Magnetic Resonance Imaging (fMRI) Beth Meyerand, PhD Department of Medical Physics University of Wisconsin-Medical School Madison, Wisconsin
Functional Magnetic Resonance Imaging (fMRI)
Beth Meyerand, PhDDepartment of Medical Physics
University of Wisconsin-Medical SchoolMadison, Wisconsin
Outline:
Introduction to fMRI physics (“BOLD”)
Overview of post-processing / analysis
Review of fMRI paradigms and procedures
Clinical applications
Blood Oxygen Level Dependent (BOLD) Imaging
* Requires a difference in oxygenation states
“ACTIVE” (increased oxyhemoglobin)
compared to
“RESTING” (decrease oxyhemoglobin)
• fMRI measures a blood flow (“hemodynamic”)effect, therefore it is only an INDIRECT measure of neuronal activity
How Are MR Images Affected By Oxygen Changes?
Hemoglobin: its magnetic properties depend on its oxygenation state.
•Oxygenated Hemoglobin-diamagnetic, like water and cellular tissue.
•Deoxygenated Hemoglobin- more paramagnetic than tissue, produces a stronger MR effect.
If an appropriate imaging sequence is selected, these differences in oxygen levels may be large enough to affect MR image intensity (T2*, or susceptibility).
Blood Oxygen Level Dependent (BOLD) Imaging
Cortical activation
Increase in cerebral metabolism
Increase CBF and CBV
Oxygen delivery exceeds oxygen demand
Oxygenation of venous pool increases
Concentration of dHb decreases
fMRI signal increase (< 5% at 1.5T)
BOLD signal time course
Signal(%) 3
0 20 40
delay
2
1
Time (seconds)
-1
Task, or stimulus timing
start stop
Echo Planar Imaging
•Imaging the whole braininvolves acquiring ~20 slicesthat are 6-7 mm thick spaced apart by 1 mm.
•The trade-off for fast scanning is low spatial resolution.
•To study brain function (changes in oxygen levels) we acquire over 100 whole brainimages collected over a time span of a few minutes.
fMRI Data Acquisition
x 100 repetitions = 100 time
points
time
signal
n slices
If 100 whole brain images are acquired in a scan, each voxel will be represented by a time series of 100
time points.
Overview of typical fMRI analysis procedures
- Motion correction
- Spatial smoothing
- Image co-registration
- Generation of statistical maps of BOLD response
- Selective interpretation of results
“pre-processing “ steps
fMRI “pre-processing”
uncorrected time courses
fMRI “pre-processing”
motion corrected time courses
fMRI “pre-processing”
motion corrected& spatially smoothed time courses
fMRI post-processingSpatial smoothing
- Increases SNR
- Some fMRI analysis software requires spatial smoothing as part of the analysis
Before After
fMRI post-processingMotion correction
- choose a time point in the middle of your time series
- register all images in time series to that time point
Before After
fMRI post-processingMotion correction
fMRI post-processingMotion correction
– effects on the time course
Before After
fMRI post-processing
EPI co-registration with high resolution anatomicals
- register (align in 3D space) EPI data to anatomicals
- sometimes very challenging due to susceptibility effects
- automatic or manual
fMRI post-processingImage registration
Left - Right
fMRI post-processingImage registration
Inferior - superior
fMRI post-processingImage registration
Roll
fMRI post-processingImage registration
Pitch
fMRI “pre-processing”EPI co- registration with high resolution anatomicals
How Do You Determine the “Active” Voxels?
Time
Sig
nal I
nten
sity
•After masking out the voxels outside the brain, we are left with ~15,000 voxel time series.
•The time series must be analyzed to determine the areas of activation during the task.
Task Performed:
task
rest
Which voxels best fit this model?
Time
Sign
al In
tens
ity
“Reference Function” of task performed:
Task Timing:
task
rest
Determining “Activation” Areas in fMRI
At every voxel, a test statistic is computed.
Functional Maps
A color intensity value is given corresponding to the value of this statistic.
Statistical Analysis
Paired t-test - used when 2 groups of data are correlated; thesame subject is scanned during activation (ON) and rest (OFF)periods.
NSXXt 21−=
Difference between the means of two groups divided by the standard error of the mean
Determining “Activation” Areas in fMRI
Functional maps after applying threshold
The statistically significant voxels are determined according to the sampling distribution.
Determining “Activation” Areas in fMRI
Functional maps after applying threshold
The statistically significant voxels are determined according to the sampling distribution.
…and then overlaid on co-registered anatomical images
Determining “Activation” Areas in fMRI
Functional maps at varying thresholds:
t > 4 t > 5t > 2
Software for fMRI data processing & display
• BrainWave GE Medical Systems
• Advanced Neuro Siemens Medical Systems
• IView Phillips Medical Systems
MRI vendor-
specific software
Unix-based
shareware• AFNI National Institutes of Health
• SPM Wellcome Dept. of Imaging Neuroscience
Other fMRI-
specific software
• Brain Voyager Brain Innovation B.V.
• Med-X Sensor Systems
fMRI acquisition set-up
LCD projector
Laptop
Stereo system
headphones
Projection screen
RF coil mirror
MRI magnet Scanner control room
fMRI paradigms for clinical patients
Pick and choose:
• Motor • Language
• Sensory• Cognitive
memoryvisual organizationcomputation
• Visual
• Auditory
fMRI paradigms:alternating-hand finger tapping task
Right hand task cycles
Left hand task cycles
restrestrestrestleftleftleftleft
rightright
rest
rightright
• 4 cycles each of right hand, left hand, and rest
• 20-second task epochs
fMRI paradigms:alternating-hand finger tapping task
LR
Yellow/orange = right hand Blue/cyan = left hand
Comparative hemispheric responses in primary sensorimotor, thalamus, putamen, superior and inferior cerebellum
fMRI somatotopic motor mapping A B
Presurgical fMRI mapping with multiple motor paradigms:A) alternating-hand finger tapping taskB) foot/ankle movement task
Comparison between finger motor task & palm tactile stimulation
Finger motor task Right hand Palm tactile stimulation Right hand
LL
Sensory Paradigms
Propofol-sedated 3 year old with right parietal mass- left hand palm stimulation
fMRI language paradigms
• Antonym generation
• Word generation from letters
• Alphabet letter generation
• Word generation from categories
• Synonyms task
• Text reading / comprehension
• Narrated text
Expressive
language
Receptive
language
covert language generation paradigms
Word generation from antonyms, letters, or categories
Simple block paradigms for robust signal localizationtask
rest
Task is performed silently to minimize head motion
Covert performance precludes patient monitoring, and motor vocalization response
Language generation paradigms
• Left-frontal gliomaL L
category word generation antonym word generation
• Repeated tasks increase confidence
fMRI of language networks Text reading paradigm
Text reading cycles
Letter stringscycles
restrestrestrestlettersletterslettersletters
texttext
rest
texttext
…
…
…
• 10 cycles each of descriptive text, letter strings, and blank screen fixation
• 8-second task epochs
fMRI of language networks
L
Text reading paradigm
Primary visual stimulus paradigm
8Hz checkerboard
fixation
Primary visual stimulusLeft occipital lesion
R
• fMRI response absent in affected hemisphere
Interpretation of clinical fMRI for presurgical
mapping
Guidelines and caveats
Interpretation of clinical fMRI for presurgical mapping
Caveats and limitations:• Inspect mapping for technical adequacy
• Extent of BOLD response is not directlyrelated to functional extent of cortex
• fMRI relies on a hemodynamic effect, doesnot directly measure neuronal activity
Interpretation of clinical fMRI for presurgical mapping
Caveats and limitations:
• Some tumors have been reported toaffect hemodynamic autoregulation
• Disturbances in cerebral blood flow andmetabolism can affect BOLD fMRI
• EPI susceptibility-related signal loss canmask regions of eloquent cortex
Effect of Pathology - AVM
• Patient with extensive rightfrontal-temporal AVM
• Text listening paradigm, no functional response seen inright superior temporal gyrus
L
> 480
360
240
120
• Temporal signal-to-noise map of the EPI signal shows low SNR in AVM region
• BOLD response could bemasked by hemodynamic instability
R
Effect of Pathology - Susceptibility
Activationabsent
R
Patient with reoccurring left frontal glioma, previous resection
• Word generation paradigm, no functional response seen inleft inferior / middle frontal gyri, right hemisphere languagedominance indicated
Effect of Pathology - Susceptibility
• EPI signal intensity mask shows region of susceptibility-induced signal loss
Effect of Pathology - Susceptibility
• EPI signal intensity mask shows region of susceptibility-induced signal loss
The furture is here!
AcknowledgementsThank you to the faculty, staff and
students at Univ. Wisconsin-Medical School
Chad H. Moritz fMRI Research Program ManagerVictor Haughton professor of NeuroradiologyHoward Rowley professor of NeuroradiologyBehnam Badie professor of NeurosurgeryRobert Dempsey professor of NeurosurgeryBruce Hermann professor of NeurologySterling Johnson professor of Neurology
Contact: [email protected]