Materials and Methods

Animals—Twenty-two dogs were prospectively studied. The experiment was conducted in accordance with the guidelines of the Animal Care Committee of Ghent University. Written owner consent was obtained prior to study enrollment. Sixteen of the 22 dogs were Doberman Pinschers. Other breeds included were Dalmatian (n = 2), Whippet (2), Weimaraner (1), and Bernese Mountain Dog (1). The clinical signs varied from cervical hyperesthesia only (n = 3) to ambulatory para-paresis and ataxia with or without cervical hyperesthesia (5), ambulatory tetraparesis and ataxia with or without cervical hyperesthesia (11), and nonambulatory tetraparesis with or without cervical hyperesthesia (3). These 22 dogs consisted of 10 males and 12 females between 4.6 and 10.8 years old (mean, 7.8 years; median, 7.4 years). For all dogs, a physical and complete neurologic examination, CBC, and serum biochemical analysis were performed. All Doberman Pinschers included underwent an additional echocardiographic examination and standard testing of buccal mucosal bleeding time. All physical and neurologic examinations were performed by 1 investigator (SD).

Imaging protocol—All examinations were performed under general anesthesia. Anesthesia was induced with propofol to effect and maintained with isoflurane vaporized in oxygen. Lactated Ringer's solution was infused IV at 10 mL/kg/h (4.55 mL/lb/h) throughout anesthesia.

In all dogs, cervical myelography was performed by use of iohexol^a (0.2 mL/kg [0.09 mL/lb] with a maximum dose of 10 mL) injected via cisternal puncture. In addition to lateral myelographic views with the neck in a neutral position, ventrodorsal myelographic views were obtained in all but 2 dogs. For these 2 dogs, variables that could only reliably be assessed with ventrodorsal myelographic views were not included in the statistical analysis. If the diagnosis of DAWS was confirmed, CTM was performed immediately after the myelographic procedure. For CTM, the dogs were positioned in dorsal recumbency with the head and neck positioned at the same level as the shoulders to avoid excessive extension of the cervical vertebral column; the thoracic limbs were fixed parallel to the chest wall. Contiguous CT slices were made from C4 to C7, parallel to the intervertebral disk spaces. A single row detector spiral CT scanner^b was used with a tube voltage of 100 kV (peak) and 100 mA. Slice thickness was 3 mm, and a bone algorithm was used. Two-dimensional multiplanar reconstructed images were made in the sagittal plane.

After CTM, the dogs recovered from anesthesia and were hospitalized. During hospitalization, the dogs were permanently monitored with special attention to the occurrence of seizures. If seizures occurred, these were treated with IV boluses of diazepam (0.5 to 1.0 mg/kg [0.23 to 0.45 mg/lb]). The next day, all dogs underwent a new complete neurologic examination by 1 investigator (SD). Immediately after this neurologic reevaluation, a permanent 0.2-T magnet^c was used to perform MRI in all dogs. Dogs were positioned the same as for CTM. The neck was positioned in a joint coil (circular transmit-receive coil) with an inner diameter of 19 cm. T1-weighted spin echo and T2-weighted fast spin echo studies were performed in all dogs in a sagittal, dorsal, and transverse plane. The images of this last plane were aligned perpendicular to the spinal cord. The vertebral column was imaged from C2 to C7 in the sagittal and dorsal planes and from C4 to C7 in the transversal plane. In all vertebral columns, the field of view was 29 cm in the sagittal, 24 cm in the dorsal, and 20 cm in the transversal planes. Slice thickness was 4 mm in the sagittal and dorsal images and 3 mm in the transversal sequences with no interslice gap in all studies. Dynamic studies were not routinely performed and were not provided to the different observers.

Observers—Images from the 22 myelographic, CTM, and low-field MRI studies were provided twice to 4 observers for evaluation. The observers consisted of 2 board-certified radiologists, 1 board-certified neurologist, and an academic staff member with > 10 years of experience in the interpretation of myelographic, CTM, and MRI studies. The observers were blinded to signalment and clinical information of the dogs. The images were provided as CD-ROMs that each contained a separate randomized sequence of 22 studies of a particular diagnostic procedure (ie, myelography, CTM, or MRI). The randomization sequence was determined by use of a randomization software program.^d Observers answered a predetermined paper questionnaire to report findings. After completion, the CD-ROM and respective questionnaire were returned, then a new CD-ROM and questionnaire were retrieved. In summary, 6 CD-ROMs containing a total of 132 imaging studies were interpreted by each observer. All images were interpreted with the same software,^e which allowed adjustment for brightness, contrast, window width, and magnification.

Interpretation of images—The following variables were assessed on the different images: number of spinal cord compressions, site of worst spinal cord compression, direction (ventral, dorsal, or dorsolateral) of spinal cord compressions, direction of worst spinal cord compression (ventral, dorsal, or dorsolateral), number and sites of narrowed or collapsed intervertebral disk spaces, number and sites of abnormally shaped vertebral bodies, number and sites of abnormally positioned vertebral bodies (craniodorsal tilting), number and sites of intervertebral disk spaces with spondylosis deformans, and number and sites of abnormal or hypertrophied articular facets. Additionally, the number and sites of stenotic intervertebral foraminae were assessed on CTM and MRI studies. In an attempt to reflect clinical practice and to avoid subjective bias, no detailed instructions or guidelines for image interpretation were provided.

Data analysis—For intraobserver agreement, each observer interpreted the 22 myelographic, CTM, and MRI studies twice. For interobserver analysis, the original studies rated by all 4 observers were included. Calculation of intermethod agreement and comparison were based on all assessed studies of the 4 observers. The intraobserver, interobserver, and intermethod agreement in rating the myelographic, CTM, and MRI images were summarized by use of κ and weighted κ statistics.²³ Weighted κ values were used for ordinal data with > 2 possible scores. Intraobserver agreement was based on the double interpretations of each observer. Interobserver agreement was based on the 22 original measurements of each observer. Intermethod agreement was based on all interpretations of all observers. The strength of agreement was interpreted on the basis of the κ values suggested by Altman,²⁴ as adapted from the method of Landis and Koch²³: κ values of 0.81 to 1.00 indicated very good agreement, κ values of 0.61 to 0.80 indicated good agreement, κ values of 0.41 to 0.60 indicated moderate agreement, κ values of 0.21 to 0.40 indicated fair agreement, and κ values of 0.20 or lower indicated poor agreement. The evaluated diagnostic modalities were further pairwise compared by use of the χ² test. Values of P < 0.05 were considered significant.

Results

In 6 of the 22 dogs, seizures occurred during the anesthetic recovery from the myelographic and CTM procedure, and in 3 dogs, neurologic deterioration was observed the day following the myelographic and CTM procedure. Neurologic deterioration was transient in 2 of the 3 dogs and consisted in each case of more pronounced ataxia and paresis of the pelvic limbs. In none of the MRI studies were artifacts caused by the preceding myelographic study encountered.

Intraobserver agreement—The κ or weighted κ values for both overall intraobserver agreement and intraobserver agreement for each observer separately were determined (Table 1). There was very good or good intraobserver agreement for most variables assessed by myelography. Intraobserver agreement was moderate for the number and site of abnormally shaped vertebral bodies and articular facets assessed by myelography.

There was moderate intraobserver agreement for most variables assessed by CTM; intraobserver agreement for the site of worst spinal cord compression, the directions of these compressions, and the direction of worst spinal cord compression was very good. There was good intraobserver agreement for the number and site of intervertebral disk spaces with spondylosis deformans, number of intervertebral disk spaces with abnormally shaped articular facets, and site of vertebral bodies with craniodorsal tilting assessed by CTM.

There was moderate intraobserver agreement for most variables assessed by low-field MRI; intraobserver agreement was very good for the direction of worst spinal cord compression. There was good intraobserver agreement for the number of spinal cord compressions and the site of worst spinal cord compression assessed by MRI; intraobserver agreement was only fair for the number and site of abnormally shaped articular facets.

Interobserver agreement—The κ or weighted κ values for both overall interobserver agreement and interobserver agreement between pairs of observers were determined (Table 2). There was fair interobserver agreement for most variables assessed by myelography. Interobserver agreement was very good for the site and direction of the worst spinal cord compression assessed by myelography, and interobserver agreement was good for the number and site of intervertebral disk spaces with spondylosis deformans assessed by myelography. There was moderate interobserver agreement for the number and site of narrowed intervertebral disk spaces assessed by myelography; interobserver agreement for the number and site of abnormal articular facets was only poor.

There was fair interobserver agreement for most variables assessed by CTM. Interobserver agreement was very good for the direction of worst spinal cord compression assessed by CTM, and interobserver agreement was good for the site of worst compression and for number and site of intervertebral disk spaces with spondylosis deformans assessed by CTM. Interobserver agreement was moderate for the site of narrowed intervertebral disk spaces and for number and site of vertebral bodies with craniodorsal tilting assessed by CTM. There was only poor interobserver agreement for number and site of both intervertebral foraminal stenosis and abnormal articular facets assessed by CTM.

There was fair interobserver agreement for most variables assessed by MRI; for the direction of worst spinal cord compression, interobserver agreement was very good, and for the site of worst spinal cord compression, interobserver agreement was good. There was moderate interobserver agreement for the number and site of vertebral bodies with craniodorsal tilting assessed by MRI. Interobserver agreement was poor for site of narrowed intervertebral disk spaces, number and site of intervertebral foraminal stenosis, and abnormal articular facets assessed by MRI.

Intermethod agreement—The κ or weighted κ values for overall agreement among the various imaging modalities were determined (Table 3). There was moderate intermethod agreement for most variables between assessment with myelography and CTM; for the site and direction of worst compression, intermethod agreement was very good, and for the number and site of intervertebral disk spaces with spondylosis deformans, intermethod agreement was good. There was poor intermethod agreement for the number and site of abnormal articular facets between assessment with myelography and CTM.

There was moderate or fair intermethod agreement for most variables between assessment with myelography and low-field MRI; for the direction of worst spinal cord compression, intermethod agreement was very good, and for the site of worst spinal cord compression, intermethod agreement was good. There was poor intermethod agreement for the number and site of abnormal articular facets between assessment with myelography and low-field MRI.

There was fair intermethod agreement for most variables between assessment with CTM and low-field MRI; for the direction of worst spinal cord compression, intermethod agreement was very good, and for the site of worst spinal cord compression, intermethod agreement was good. There was moderate intermethod agreement for number of spinal cord compressions, site of abnormal vertebral bodies, and number and site of intervertebral disk spaces with spondylosis deformans between assessment with CTM and low-field MRI.

Difference between the various imaging modalities—Spinal cord compressions and craniodorsal tilted vertebral bodies were diagnosed significantly more often by use of MRI, compared with myelography and CTM, and by use of myelography in comparison with CTM. Narrowed intervertebral disk spaces and abnormally shaped vertebral bodies were diagnosed significantly more often by use of MRI, compared with CTM and myelography, and by use of CTM in comparison with myelography. Spondylosis deformans was diagnosed significantly more often by use of CTM, compared with myelography and MRI, and by use of myelography in comparison with MRI. Intervertebral foraminal stenosis was diagnosed significantly more often by use of MRI, compared with CTM. Abnormal articular facets were diagnosed significantly more often by use of CTM, compared with MRI and myelography, and by use of MRI in comparison with myelography. All significant differences between these imaging modalities were characterized by values of P < 0.001.

Discussion

In the present study, we determined the interobserver, intraobserver, and intermethod agreement for results of myelographic, CTM, and low-field MRI studies of a selected population of dogs with clinical signs of DAWS. Although previous veterinary reports have compared myelography with CTM²² or myelography with MRI²¹ in Doberman Pinschers with caudal cervical spondylomyelopathy, no previous studies have compared these 3 diagnostic modalities in the same population of dogs.

In general, there was very good to good intraobserver agreement for variables assessed by myelography, but there was only moderate intraobserver agreement for most variables assessed by CTM and low-field MRI. For the 3 diagnostic techniques, there was moderate to fair interobserver and intermethod agreement for most assessed variables. These findings of moderate to fair intraobserver, interobserver, and intermethod agreement for variables assessed by CTM and MRI in the present study are similar to findings of several human studies,^25–27 suggesting a disturbing variability in image interpretation when either of these common imaging techniques are used for assessment of cervical spondylomyelopathy.

In the present study, the site and direction of worst spinal cord compression were the most reliable variables assessed, with always very good or good intraobserver, interobserver, or intermethod agreement, regardless of the imaging modality used. Abnormal articular facets and intervertebral foraminal stenosis were the least reliable variables assessed, with poor interobserver agreement.

Although the assessment of the myelographic studies resulted in relatively better intraobserver and interobserver agreement values for several of the assessed variables, compared with CTM and MRI, this imaging modality is associated with considerable complications and limitations.^2,19 This rather invasive procedure is not completely without risk.^2,28 As demonstrated in the present study, seizures and transient neurologic deterioration are the most important complications following myelography.^17,28,29 A higher risk of postmyelographic complications has been found for Doberman Pinschers with caudal cervical spondylomyelopathy than for dogs with other cervical spinal cord lesions.²⁹ Other reported risk factors for the development of postmyelographic complications are total volume of iohexol administered, higher body weight (> 20 kg [44 lb]), cerebellomedullary injection site (vs lumbar injection site), multiple cerebellomedullary injections, sex, and breed.^29,30 Six of 22 dogs in the present study developed seizures during anesthetic recovery, and 3 dogs were neurologically worse the day after the myelographic and CTM examinations. This is in agreement with a recent veterinary study²¹ in which 5 of 18 Doberman Pinschers with cervical spondylomyelopathy developed postmyelographic seizures. Although myelography allows screening of the entire cervical vertebral column for compressive lesions, the results of the present study suggest that the lack of transversal imaging limits the diagnostic potential of this technique. Lack of transversal imaging resulted in the inability to evaluate intervertebral foraminal stenosis and complicated reliable assessment of articular facet abnormalities on myelography. The latter is illustrated by the poor interobserver and intermethod agreement values of this variable with the use of myelography.

Based on the best intraobserver agreement, findings of the present study suggest that CTM is the most reliable imaging technique to evaluate articular facet abnormalities. This is in agreement with findings in human studies,^27,31 where CT is considered the gold standard in predicting articular facet abnormalities. This can probably be explained by the superior visualization of bony detail by CTM, compared with MRI.^31,32 Although an optimal CTM image is obtained with a lower dose of intrathecal contrast medium than in a conventional myelographic study,³³ the same complications can occur as with myelography.² Because CTM is initially limited to the transversal plane, it is not as efficient as myelography and MRI in screening the entire cervical vertebral column and is therefore generally used as a supplementary examination at specific levels of interest.³⁴ This was also true in the present study, where myelography served to localize the lesion for subsequent CTM examination.

Similar to human medicine,³⁵ it is believed that intervertebral foraminal stenosis can be a contributing factor or even a primary cause of clinical signs in dogs with cervical spondylomyelopathy.¹⁸ In the present study, the interobserver agreement for intervertebral foraminal stenosis was very poor by use of CTM and low-field MRI. It has been postulated that intervertebral foraminal stenosis is best assessed on oblique radiographic images of the neck,³³ by use of IV enhanced contrast CT,^22,33 or on gradient-echo MRI sequences with magnetization transfer.^18,36,37 However, none of these studies were performed in the present study. It is possible that higher agreement values would have been reached if these studies could have been provided to the observers. In agreement with findings in human studies,^{27,32,38–40} intervertebral foraminal stenosis was diagnosed significantly more often by use of MRI than by CTM. This is probably caused by the presence of a susceptibility artifact,³⁷ which is manifested as signal loss at the outer edges of bone because of the abrupt difference in magnetic susceptibility at bone–soft tissue interfaces. As a result, the bone appears larger than it really is and thus artifactually narrows adjacent soft tissue. This artifact can be misread as foraminal stenosis.³⁸ Addition of magnetization transfer to the image acquisition decreases the signal arising from soft tissue and thus increases the contrast between structures and reduces the artifacts.^36,37,41,42

Low-field MRI also provided substantially lower intraobserver, interobserver, and intermethod agreement for the assessment of intervertebral disk spaces with spondylosis deformans, compared with CTM and myelography. This high variability of spondylosis deformans is in agreement with a recent veterinary study⁴³ and several human studies^27,40 and is presumably related to the variable presence and composition of bone marrow in osteophytes.⁴⁰ Osteophyte signal can vary from a markedly decreased intensity when the bone is dense to an intensity homogenous with either the disk or the adjacent vertebral body when fatty marrow is more abundant.⁴⁰

There was poor interobserver agreement for the site of narrowed intervertebral disk spaces assessed by low-field MRI and a moderate interobserver agreement for these variables when assessed by myelography and CTM. Additionally, narrowed intervertebral disk spaces were diagnosed significantly more often by use of MRI than by CTM and myelography. Although the exact reason for this remains unclear, it can probably be explained by difficulties in discriminating the hypointense signals from a completely degenerated disk and the adjacent vertebral cortices on sagittal MRI images. This can potentially result in increased variability and overinterpretation of intervertebral disk space narrowing on MRI.

Overinterpretation is a recognized problem that complicates the interpretation of MRI images of the cervical vertebral column in dogs.^18,41–43 Some reports^18,42 have described the presence of degenerative abnormalities on MRI images of clinically normal dogs. With this in mind, it is not surprising that spinal cord compression was diagnosed significantly more often by use of MRI than by CTM and myelography. Despite this risk of overinterpretation, MRI offers several advantages over CTM and myelography. In contrast to myelography and CTM, MRI allows direct, noninvasive, multiplanar imaging without the need for reconstruction and an excellent soft tissue characterization with an absence of ionizing radiation.^20,44 Because MRI is independent of intrathecal contrast administration, it not only is safer than myelography and CTM but also undergoes no image degradation because of CSF block distal or proximal to a severe compressive lesion.^40,45 It is worthwhile to mention that limitations concerning sagittal and coronal reconstructions are less important with the use of the newer-generation multislice CT units. A distinct advantage of MRI is the ability to correctly assess the spinal cord parenchyma.⁴⁴ Although degenerative disease, especially DAWS, was suspected as the cause of myelopathy in the present study, a large number of disease entities must be included in the differential diagnosis of middle-aged to older large-breed dogs with cervical hyperesthesia or myelopathy. With the use of MRI, myelopathy secondary to intrinsic spinal cord disease can be easily distinguished from myelopathy secondary to compressive disease.³² Although intraparenchymal signal changes were not evaluated in the present study, MRI also allows for their assessment.⁴⁴ These changes are considered nonspecific for a certain disease entity,^46,47 but their presence has been suggested to be a reliable indicator of clinically relevant cervical spinal cord compression.^18,43

The authors recognize several limitations of the present study. All the included dogs had a clinical diagnosis of cervical hyperesthesia or myelopathy secondary to DAWS. As a result, it is likely that the imaging findings were influenced by the presence of specific pathological changes in most patients. Therefore, the authors suggest that the results of the present study should ideally be confirmed in studies where other cervical lesions are also included. All MRI images were obtained with a low-field (0.2-T) MRI unit. It is possible that the use of a magnet with greater strength could have resulted in higher agreement and lower variability. Additional variation in image interpretation could be caused by the fact that the different observers did not assess the imaging studies at the same workstation. To reflect clinical practice, the observers did not receive detailed guidelines for interpretation. Although findings in human MRI studies^48,49 have suggested that operational guidelines do not necessarily improve intraobserver or interobserver agreement in patients with cervical myelopathy, this had not yet been investigated in veterinary medicine. Although efforts were made to perform each imaging study with the cervical vertebral column in a neutral position, intermethod variability due to differences in neck position cannot be excluded.

In summary, the results of the present study suggest that there is considerable variation in image interpretation among observers and between the use of various imaging modalities in dogs with DAWS. The observed discrepancies in interpretation call into question the reliability of comparisons among observers and various imaging techniques. Analysis of the results of the present study indicate that it seems inappropriate to consider one of the evaluated diagnostic modalities as the imaging modality of choice for dogs with DAWS and that they should rather be considered as complementary to each other. Although not addressed in the present study, plain CT or survey radiography could perhaps complement MRI in demonstrating bony detail and intervertebral foramina, thereby substantially decreasing the need for intrathecal contrast administration.