Evaluation of magnetic resonance imaging for the differentiation of inflammatory, neoplastic, and vascular intradural spinal cord diseases in the dog

Evaluation of magnetic resonance imaging for the differentiation of inflammatory, neoplastic, and vascular intradural spinal cord diseases in the dogReceived: 25March 2016 Revised: 11 December 2016 Accepted: 16 January 2017
DOI: 10.1111/vru.12501
OR I G I NA L I NV E S T I G AT I ON
Evaluation ofmagnetic resonance imaging for the differentiation of inflammatory, neoplastic, and vascular intradural spinal cord diseases in the dog
Amanda E.Masciarelli1 John F. Griffin IV1 Geoffrey T. Fosgate2 Silke Hecht3
JosephM.Mankin4 Shannon P. Holmes5 Simon R. Platt6 Marc Kent6
Theresa E. Pancotto7 Annie V. Chen8 JonathanM. Levine2
1Departments of LargeAnimal Clinical Sciences,
College of VeterinaryMedicine andBiomedical
2TheDepartment of ProductionAnimal Studies,
3TheDepartment of Small Animal Clinical Sci-
ences, University of TennesseeCollege of Veteri-
naryMedicine, Knoxville, TN, 37996
VeterinaryMedicine andBiomedical Sciences,
30602
ofGeorgia College of VeterinaryMedicine,
ences, Virginia-MarylandCollege of Veterinary
Medicine, Blacksburg, VA, 24061
8TheDepartment of VeterinaryClinical Sci-
ences, College of VeterinaryMedicine,Washing-
ton StateUniversity, Pullman,WA, 99164
University,CollegeStation, TX77843.
Email: [email protected]
Abstract Magnetic resonance imaging (MRI) is a common test for dogswith suspected intradural spinal cord
lesions, however studies on diagnostic performance for this test are lacking. Objectives of this
multi-institutional, retrospective, case-control study were to estimate sensitivity and specificity
ofMRI for (1) distinguishing between histopathologically confirmed intradural spinal cord disease
versus degenerative myelopathy in dogs, (2) categorizing intradural spinal cord diseases as neo-
plastic, inflammatory, or vascular; and (3) determining tumor type within the etiologic category of
neoplasia. Additional aimswere to (1) determinewhether knowledge of clinical data affects sensi-
tivity and specificity of MRI diagnoses; and (2) report interrater agreement for MRI classification
of intradural spinal lesions. Cases were recruited from participating hospital databases over a 7-
year period. Three reviewers independently evaluated eachMRI study prior to and after provision
of clinical information. A total of 87 cases were sampled (17 degenerative myelopathy, 53 neopla-
sia, nine inflammatory, and eight vascular). Magnetic resonance imaging had excellent (>97.6%)
sensitivity for diagnosis of intradural spinal cord lesions but specificity varied before and after
provision of clinical data (68.6% vs. 82.4%, P= 0.023). Magnetic resonance imaging had good sen-
sitivity (86.8%) and moderate specificity (64.7–72.5%) for diagnosing neoplasia. Sensitivity was
lower for classifying inflammatory lesions but improved with provision of clinical data (48.1% vs.
81.5%, P = 0.015). Magnetic resonance imaging was insensitive for diagnosing vascular lesions
(25.0%). Interrater agreementwas very good for correctly diagnosing dogswith intradural lesions
(= 0.882–0.833), and good (= 0.726–0.671) for diagnosing dogs with neoplasia.
K EYWORDS
1 INTRODUCTION
intradural spinal cord lesions in dogs. Advantages offered by MRI
include noninvasive acquisition of images in three dimensions
and superior contrast resolution compared with other available
technologies.1–4 As intradural spinal cord lesions are often challeng-
ing to biopsy, especially those that are intramedullary and do not
Portions of this study were presented at the 2015 ACVR Annual Scientific Conference,
Minneapolis, MN.
contact the pial surface, MRI is considered to be a key test for making
diagnoses in the clinical setting. Over the past decade, several studies
in dogs have described the MRI appearance of common spinal cord
neoplasms,3,5–7 inflammatory diseases,8–11 and vascular lesions.12,13
However, controlled studies describing the sensitivity and specificity
of MRI for detecting intradural spinal cord lesions in dogs and for
differentiating neoplastic, inflammatory, and vascular etiologies are
not currently available.
TheMRI features of spinal cord neoplasia, inflammation, and vascu-
lar disease have been reported to overlap in dogs, and this may impact
Vet Radiol Ultrasound. 2017;58:444–453. c© 2017 American College of Veterinary Radiology 444wileyonlinelibrary.com/journal/vru
MASCIARELLI ET AL. 445
meningoencephalomyelitis, and ischemic myelopathies can all exhibit
well-defined T2-weighted (T2W) hyperintense lesions with variable
contrast enhancement and variable spinal cord enlargement.3,8,11,14
There have been published cases of granulomatous meningoen-
cephalomyelitis with lesions in the cervical spinal cord exhibiting T2W
hyperintensity, equivocal T1-weighted (T1W) hypointensity, and no
contrast enhancement;8,15 all ofwhich are also reported to beMRI fea-
tures of ischemic myelopathy.12 Based on data from dogs with brain
disease, the ability of MRI to diagnose specific diseases within broad
etiologic categories may be imperfect, even for common conditions.
For example, in dogswith brain neoplasia,MRI has been reported to be
59.6% sensitive and 94.9% specific for the detection of meningioma.16
Currently, there are insufficient published data to estimate the reli-
ability ofMRI for diagnosing intradural spinal cord lesions in dogs. Sev-
eral studies have described interrater agreement for the MRI-based
detection of canine extradural spinal cord diseases, most prominently
intervertebral disk herniation.17–19 In general, interrater agreement
for the determination of the site of extradural compression and side
has been reported to be fair to good. However, limited data are avail-
able on repeatability and accuracy for establishing an etiologic diagno-
sis for extradural compressive diseases in dogs.
The purposes of this study were to (1) estimate sensitivity and
specificity of routine, high-field MRI to distinguish between dogs
with histopathologically confirmed intradural spinal cord disease and
a control population; (2) estimate sensitivity and specificity of MRI
to broadly categorize intradural spinal cord diseases as neoplastic,
inflammatory, or vascular; (3) estimate sensitivity and specificity of
MRI to determine tumor type within the broad etiologic category of
neoplasia; (4) evaluate whether knowledge of clinical data will affect
the sensitivity and specificity ofMRI diagnoses; and (5) calculate inter-
rater agreement for classification of intradural lesions into broad eti-
ologic categories. Our hypotheses were that sensitivity and specificity
of MRI for classifying lesions in the broad categories will be high, and
will be moderate for specific neoplastic diagnoses. We also hypothe-
sized that, as has been previously demonstrated in canine brain MRI
studies,16,20 the provision of clinical data will minimally impact test
sensitivity and specificity, and interrater agreement will be fair-good.
2 MATERIAL AND METHODS
tutional animal care and use committee approval was not required.
Medical records from 2007 to 2014 were searched for dogs with
intradural (intradural extramedullary or intramedullary) spinal cord
disease that were admitted to one of the following five veterinary
medical teaching hospitals: Texas A&MUniversity (TAMU), University
of Georgia (UGA), Washington State University (WSU), University of
Tennessee (UT), and VirginiaMaryland College of VeterinaryMedicine
(VMCVM). Dogs were included in this study if the following criteria
were met: (1) ante mortem spinal cord MRI available for review and
(2) underlying disease confirmed via histopathology as neoplastic (pri-
mary or metastatic), inflammatory (immune-mediated, infectious, or
unknown etiology), or vascular (ischemic or hemorrhagic). Databases
were also searched for dogs with degenerative myelopathy diagnosed
basedonSOD-1mutationhomozygosity, clinical signs, andnormalMRI
with or without confirmatory histopathology. These dogs were used as
a control group for the study.
2.2 Clinical data
Standard clinical data recorded for each dog were as follows: admit-
ting university, breed, sex, age at presentation (in years), weight
(in kilograms), duration of clinical signs (number of days between
onset of clinical signs and imaging), progression of clinical signs
(progressive/worsening—clinical signs worsened between onset and
MRI; static/unchanged—clinical signs remained static between onset
and MRI; improved—if clinical signs abated or appeared to have
resolved fully between onset and MRI), cerebrospinal fluid analysis
(if available), and necropsy or biopsy histopathologic diagnosis. Data
were entered into spreadsheets using commercially available software
(Excel, Office 2010,Microsoft Corp., Redmond,Washington).
2.3 Image analysis
Technical parameters for the MRI study acquisitions were also used
as criteria for subject inclusion in order to maximize homogeneity
of image data for readers. The MRI study requirements were as fol-
lows: (1) field strength ≥1.0T, (2) sagittal and transverse image planes
for T2W image sequences through lesion area, (3) a transverse image
plane for T1W pre- and postcontrast images, and (4) images in digi-
tal imaging and communications in medicine format. Additional image
planes and sequences were excluded from evaluation.
Before the MRI analysis, 1 investigator (A.M.) anonymized all
images by removing case identifiers using imaging software (Clear
Canvas, Inc, Toronto, ON, Canada). Sequential case numbers were
assigned to each study in a randomized fashion using a free online ran-
dom number generator (random.org). Two board-certified veterinary
radiologists (J.G. and S.H.) and 1 board-certified veterinary neurologist
(J.M.) independently reviewed and analyzed MR images using digital
imaging software. These investigators were not involved in case selec-
tion, review of medical records, or in medical record abstraction.
The three reviewers were asked to record the presence or absence
of an intradural spinal cord lesion, defined as an abnormality in spinal
cord or intradural-extramedullary compartment morphology or tissue
signal characteristics. Reviewersprioritized themost likelydisease cat-
egory by following published imaging criteria to aid differentiation of
inflammatory (single or multiple lesions, hyperintense on T2W images,
usually with an irregular or infiltrative pattern),8,11 neoplastic (usu-
ally single lesion, more well-defined shape, ± a mass effect, hyperin-
tense on T2W, exhibit contrast enhancement),3,5–8,22–32 and vascular
(ill-defined to triangular/wedge-shaped lesions, hyperintense on T2W
images, none to variable contrast enhancement, predominantly involv-
ing graymatter)12–14,33–35 spinal corddiseases of dogs. Theywere then
asked to specify the most likely spinal cord disease represented by the
446 MASCIARELLI ET AL.
received a copyof completed responses. Reviewerswere subsequently
given clinical data obtained from medical records and asked if they
wanted to modify their initial responses to the following: (1) normal
versus abnormal MRI study, (2) most likely etiologic category (neopla-
sia, inflammatory, or vascular), and (3) most likely specific diagnosis.
Reviewers recorded their new response if they elected to modify their
initial response based on the available clinical data.
2.4 Statistical analysis
manually entered formulas into a commercially available spreadsheet
program (Excel, Office 2010, Microsoft Corp., Redmond, Washing-
ton) and results were interpreted at the 5% level of significance. The
accuracy of MR to detect intradural spinal cord lesions was evalu-
ated by estimating sensitivity and specificity. Sensitivity of detecting
a lesion (in general) was estimated as the proportion of nondegen-
erative myelopathy cases correctly identified as having an intradu-
ral lesion. Specificity was estimated as the proportion of degenera-
tive myelopathy cases correctly identified as not having an intradural
lesion. Sensitivity for each broad etiologic category (neoplastic, inflam-
matory, and vascular) was estimated as the proportion of histologi-
cally confirmed cases within each category correctly identified by the
reader as having that lesion type. Specificity was estimated as the
proportion of cases within the other categories (excluding degenera-
tive myelopathy cases) correctly identified as not having that particu-
lar lesion type. Diagnosis-specific measures of MR performance were
calculated for neoplastic conditions in which more than a single con-
firmed casewas identified. Sensitivity was estimated as the proportion
of cases correctly identified and specificity was estimated as the pro-
portion of other etiologies within the broad neoplastic diagnostic cat-
egory correctly identified as not having that particular neoplastic con-
dition. The variance for a clustered sample (i.e. each MRI reviewed by
three reviewers) was estimated and used to adjust confidence interval
for sensitivity and specificity.36 Sensitivity and specificity were com-
pared between pre- and postclinical data using McNemar’s test while
accounting for the repeated observations on the same dog.37 Stan-
dard formulas were used to calculate the kappa statistic and its asso-
ciated confidence interval as a measure of interrater agreement.38
Kappa values of ≤ 0.20, 0.21–0.40, 0.41–0.60, 0.61–0.80, and 0.81–
1.00 were assumed to indicate poor, fair, moderate, good, and very
good agreement, respectively.39 Agreement between pairs of review-
ers was reported in tables.
3 RESULTS
The sample population consisted of 87 dogs from the five veterinary
medical teaching hospitals (31 from UGA, 24 from TAMU, 19 from
WSU, 11 from UT, and two from VMCVM) (Table 1). A total of 70 dogs
with intradural spinal cord lesions were assigned to the cases group
and 17 dogs with degenerative myelopathy were assigned to the con-
trol (degenerative myelopathy) group. A 1.0T MRI system was utilized
at TAMU from 2007–2011 and 3.0T was utilized from 2011–2014
(T2W: TR (repetition time) 2500–8,420ms, TE (echo time) 45–119ms,
slice thickness 2.0–4.0 mm; T1W: TR 500–2591.5 ms, TE 2.4–20, slice
thickness 2.5–4.0 mm). Both a 1.5T and a 3.0T MRI system were uti-
lized for UGA cases (T2W: TR 3,150–4,870 ms, TE 78–123 ms, slice
thickness 2.0–4.0 mm; T1W: TR 601–2752 ms, TE 9.6–16 ms, slice
thickness 2.0–4.0 mm). A 1.0T MRI system was utilized for all WSU
cases (T2W: TR 1,839–8087 ms, TE 96–130 ms, slice thickness 2.7–
6.0 mm; T1W: TR 317–1160 ms, TE 12–20 ms, slice thickness 3.0–
6.0 mm). A 1.0T MRI system was utilized at UT from 2007–2012 and
a 1.5T MRI system from 2013–2014 (T2W: TR 1450–3960 ms, TE
88–126 ms, slice thickness 2.0–10.0 mm; T1W: TR 300–987 ms, TE
17–20 ms, slice thickness 2.5–10.0 mm). A 1.5T MRI system was uti-
lized for VMCVM cases (T2W: TR 2646 – 6660 ms, TE 110–132 ms,
slice thickness 2.5–3.0 mm; T1W: TR 317–1160 ms, TE 12–20 ms,
slice thickness 3.0–6.0 mm). Postcontrast sequences were acquired
after intravenous administration of gadolinium-based contrast agent
(gadopentate dimeglumine, Magnevist, Bayer Healthcare Pharmaceu-
ticals,Wayne, NJ) at a dose of 0.1mmol/kg.
All of the degenerative myelopathy dogs were described as having
a progressive/worsening clinical course. Themedian duration between
onset of clinical signs and MRI was 90 days (range 1–270 days). There
were 53 cases of spinal cord neoplasia, nine cases of inflammatory
myelopathy, andeight cases of vascularmyelopathy (Fig. 1 and2).Diag-
nosis was based on antemortem biopsy (n = 39) or necropsy (n = 31).
A progressive clinical course was identified in 52/53 of the cases in
the neoplastic group, all of the cases in the inflammatory group, and
6/8 cases in the vascular group. The clinical course was described as
static/unchanged in 1/53 of the cases in the neoplastic group and 2/8
cases in thevascular group. Themediandurationbetweenonsetof clin-
ical signs and MRI was 30 days (range 0.5–200 days) in the neoplas-
tic group, 28 days (range 4–90 days) in the inflammatory group, and
1.5 days (range 0.5–10 days) in the vascular group.
Histopathology at each institution was performed by a board-
certifiedpathologist or anatomic pathology resident under their super-
vision. Tissue sections were stained routinely with hematoxylin and
eosin for histologic examination. Immunohistochemical staining with
cytokeratin, vimentin, CD34, Factor VIII, CD3 and CD79a, CD18 and
Iba-1, S100, and glial fibrillary acidic protein were used to further
characterize the histologic diagnosis when neoplasia was suspected.
All tumors were typed in accordance with World Health Organization
recommendations.21 For dogs with infectious myelitis, standard test-
ing for fungal, protozoal, bacterial, rickettsial, and viral etiologies was
performed.
gioma (n = 24), nephroblastoma (n = 7), unspecified/other (n = 6),
malignant nerve sheath tumor (n = 5), glioma (oligodendroglioma:
n = 2, astrocytoma: n = 2, gliomatosis cerebri: n = 1), ependymoma
(n=3),metastatic neoplasia (n=1), and one each of lymphoma and his-
tiocytic sarcoma.Histopathologic diagnoses in the inflammatory group
included granulomatous meningoencephalomyelitis (n = 7), immune
mediated (n = 1), and infectious (protozoal: n = 1). Histopathologic
diagnoses in the vascular group included ischemic myelopathy (n = 4) and hemorrhagic myelopathy (n = 4). All four dogs with ischemic
MASCIARELLI ET AL. 447
TABLE 1 Demographic data for sample population (N= 87)
Category n Age (Y) (range; median) Weight (kg) (range; median) Sex n Breed n
Degenerativemyelopathy 17 3–12; 10 9–51; 18 M 0 PembrokeWelsh Corgi 7
MN 5 Boxer 3
FS 12 Mixed breed 1
Other breeds (1 each) 4
Neoplastic 53 1–16; 7 3–82; 25 M 7 Mixed breed 9
MN 13 Boxer 9
F 4 Bulldog 4
Golden Retriever 2
Other breeds (1 each) 22
Inflammatory 9 2–8; 7 3–40; 10 M 2 Chihuahua 2
MN 2 Mixed breed 2
F 0 Other breeds (1 each) 5
FS 5
Vascular 8 1–13; 7.5 2–49; 25 M 0 Labrador Retriever 4
MN 4 Yorkshire Terrier 2
F 0 Other breeds (1 each) 2
FS 4
n, number; Y, years; M, male intact; MN, male neutered; F, female intact; FS, female spayed.
myelopathy had fibrocartilaginous embolism. The dogs with hemor-
rhagic myelopathy had intradural-extramedullary hemorrhage (n = 2)
or intramedullary hemorrhage (n= 2).
Magnetic resonance imaging was highly sensitive for the identi-
fication of intradural spinal cord lesions (Table 2). Within the pop-
ulation of dogs with intradural lesions, sensitivity was highest for
neoplasia and lowest for vascular disease. Provision of clinical data sig-
nificantly improved the specificity for diagnosis of intradural lesions
(P = 0.023) and the sensitivity for the detection of inflammatory
disease (P = 0.015). Specificity was highest for vascular disease
and lowest for neoplasia. Within the neoplastic group, sensitivity
was highest for diagnosis of meningioma and nerve sheath tumor
(Table 3). Individual reviewer sensitivity and specificity is reported
in Appendix 1. There was very good agreement in the classifica-
tion of dogs into control and positive intradural lesion groups (Table
4). There was good agreement in the classification of dogs into the
neoplastic group versus the inflammatory or vascular group. Agree-
ment between pairs of reviewers is reported in Appendix 2 and
Appendix 3.
4 DISCUSSION
To the authors’ knowledge, this is the first published report analyz-
ing diagnostic performance for MRI in dogs with intradural spinal
cord lesions. Findings from the current study indicated that MRI had
excellent sensitivity and moderate specificity in the diagnosis of dogs
with intradural spinal cord lesions. The suboptimal specificity for the
diagnosis of dogs with intradural spinal cord disease indicated a high
likelihood for false positive diagnoses. Prior to the provision of clini-
cal data, each reviewer made five or six false positive diagnoses. Out
of 16 instances (i.e. two reviewers making five false positive diag-
noses and one with six), the reviewers misdiagnosed the dogs as hav-
ing inflammatory myelopathy 12 times, neoplasia three times, and
vascular myelopathy once. The improved specificity following provi-
sion of clinical data was largely due to reviewers changing their ini-
tial diagnosis of inflammatorymyelopathy to degenerativemyelopathy
(no lesion). This happened in five instances involving four dogs, three
of which had CSF analysis results available. These dogs had a median
nucleated cell count of 0 cells/L (range 0–7 cells/L) and a median
duration of clinical signs of 60 days (range 54–240 days). We believe
that reviewers had difficulty deciding whether or not subtle changes
in signal intensity were real. The problem was likely exacerbated by
variable image acquisition parameters and an inconsistent relation-
ship between imaging features and histopathology.40 Future studies
are needed to further evaluate this finding.
As expected, there was good sensitivity in the diagnosis of neo-
plasia. The neoplasia group primarily included dogs with meningioma,
nephroblastoma, and nerve sheath tumor. The imaging features of
these conditions have been well described and they tend to cause
an intradural-extramedullary mass in a typical location.6,23,25,28,29,41
Specifically, meningioma frequently occurs in the C1-C4 region,
448 MASCIARELLI ET AL.
F IGURE 1 T2-weighted transverse and sagittal images of the cervi- cal vertebral column, exhibiting a hypointense intramedullary lesion in the dorsal aspect of the spinal cord at C3-4 (arrowhead). The adjacent portion of the spinal cord ventral to the lesion is hyperintense. This lesion was classified as neoplastic by all three evaluators, who subse- quently identified this as an inflammatory lesion after provision of clin- ical data. Histopathology on necropsy of this patient revealed granulo- matousmeningoencephalomyelitis
nephroblastoma frequently occurs in the T9-L3 region, and nerve
sheath tumors frequently occur at the brachial or lumbosacral
plexus.6,23,25,28,29,41
ditions was poor. For example, sensitivity for the diagnosis of
glioma was 13.3–26.7% and for ependymoma was 11.1%. Pub-
lished descriptions of imaging features of these conditions are
sparse, their predilection sites are poorly understood and vari-
able, and their MRI features overlap with those of inflamma-
tory or vascular disease.3,5,8,12 This is a crucial area for future
study. In a large survey of 331 dogs with histopathologically
confirmed spinal cord tumors, 16% were intramedullary. These
lesions are infrequently biopsied and are treated differently than the
inflammatory or vascular lesions that theymimic.5,8,42
Provision of clinical data improved sensitivity for the diagnosis of
inflammatory disease in the current study. The improved sensitivity
following provision…