This study did not involve live animals but the analysis of MRI DICOM obtained for diagnostic purposes or for determining CM and syringomyelia status. Full consent was obtained from all owners and actual dogs remained anonymous.

There was very high intra-measurer reliability for the line AE data, with an ICC of 0.985, 95% CI (0.944, 0.996) and for the F-diameter data with an ICC of 0.880, 95% CI (0.591, 0.969). The 155 dogs in the study ranged from 1.1 to 12.6 years old (mean 4.5 years, standard deviation 2.5 years; median 3.8 years). 53 (34%) dogs were under three years of age. 39 (25%) were 3.0- to 4.9 years 63 (41%) were 5 years and over. Five dogs had insufficient imaging of the forebrain to identify the optic nerve and optic canal and, for 11 dogs, the quality of the scan prevented the olfactory bulbs from being identified with certainty so that line AH and Angle 5 were not recorded for these dogs.

The distribution of CM and syringomyelia affectedness is depicted in Table 2. Since syringomyelia may have a late onset, 20/53 dogs under three years with no central canal dilatation or syringomyelia were considered too young for a reliable syringomyelia status leaving a total of 135 dogs with confirmed syringomyelia status possible for this group.

Figure 2 and Table 3 illustrate eight representative dogs A-H and their CM and syringomyelia status with increased affectedness from left to right. The MRI images with significant lines and angles and three-dimensional reconstructions of the brain are depicted and dog’s age at time of MRI scan noted. Dogs with CM and syringomyelia tend to have more acute Angles 2 and 5 (colored yellow) and longer lines AE, BC and BD (colored blue). Interestingly this also applied for Dog F which had syringomyelia but was not CM affected. An explanation for this change could be cranial base shortening and increased proximity of the atlas to the occiput. In addition F-diameter was increased suggesting increased height of the rostral cranial fossa. A trend for a decrease in Angle 3 was not a consistent finding and in fact the two dogs with largest Angle 3 (colored red) was Dog E (CM0 SM1) with 1 mm central canal dilation and Dog H (CM2 SM2) with 9 mm syrinx. The three dimensional images illustrate that with increased syringomyelia affectedness the cerebellum is displaced ventral to the occipital lobes giving it a “pushed in” appearance (Dogs D, F, G, and H).

In Figure 3 and Table 3, a comparison is made between mesaticephalic breeds (Beagle and Australian terriers) and a second brachycephalic toy breed (Affenpinscher) which has common ancestors with the Griffon Bruxellois. None of the dogs had syringomyelia thus facilitating a comparison of CM only. Angle 2 becomes more acute with CM which can be appreciated when the normal Beagle (Dog I) and Australian terrier (Dog J) are compared to the Australian terrier with mild CM (Dog K). When the mesaticephalic breeds are compared with the brachycephalic Affenpinchers, Angle 5 becomes smaller and Angle 3 becomes larger. The three dimensional brain images of the CM affected Affenpinschers are similar to the Griffon Bruxellois dogs as the cerebellum becomes “tucked under” and ventral to the occipital lobes.

The three groups, CM0 (No CM), CM1 (Mild CM) and CM2 (CM) comprised 31, 32 and 92 dogs respectively. Initial analysis of all possible variables indicated that 8 of the 15 variables were significantly different across the three groups (Table 4). With three groups stepwise Discriminant Analysis produced 2 functions:

Function 1 [p<0.001] D1 = − 0.351* F-diameter +0.112* Angle 2+6.968

Function 2 [p = 0.197] D2 = 0.423* F-diameter +0.118* Angle 2–26.357

The stepwise selection confirmed the two most significant variables from the ANOVA results (F-diameter and Angle 2) and the functions themselves are ordered by their level of statistical importance hence Function 1 is the more powerful. Interpreting Function 1 is based on how its value can be increased or decreased hence larger Angle 2 values accompanied by smaller F-diameters will increase this function (larger positive value). Conversely Function 1 will be smaller (more negative) when F-diameter is larger and Angle 2 is smaller. Function 2, whilst statistically non-significant also selected F-diameter and Angle 2 as the important variables and allows for some separation in another dimension, with larger measures of both variables increasing Function 2 values. Figure 4 illustrates this, plotting Function 1 (horizontal) against Function 2 (vertical). The Group means (dark+signs) move more to the left as the CM0 to CM1 to CM2 i.e. CM absent to present. The success of this methodology assessed using cross-validation yielded an overall correct classification result of 52.9% however this success varied across the groups: CM 0 61.3% (19/31); CM 1 40.6% (13/32) and CM2 54.3% (50/92).

SM0 and SM1 dogs over 3 years were combined to form syringomyelia “SM NO” group ensuring there was no bias for the late onset nature of syringomyelia and SM2 became syringomyelia “SM YES”. Initial analysis of the data from those 135 dogs across all possible variables indicated that 5 of the 15 variables were significantly different (Table 5). Stepwise Discriminant Analysis produces only 1 function when only 2 groups are being compared: Function [p<0.001] D = 0.410 * F-diameter − 0.150 * Angle 5–10.757.

This confirmed the two most significant variables from the ANOVA results (F-diameter and Angle 5). Interpreting the function, as previously, is based on how its value can be increased or decreased hence larger F-diameters accompanied by a smaller Angle 5 will increase the value of the function (more positive) and conversely smaller F-diameters accompanied by a larger Angle 5 will decrease the value of the function (more negative). Figure 5 illustrates these function values for the SM NO and SM YES groups and illustrates that on average the SM YES dogs have slightly higher values of the function (more positive). A few SM YES dogs can be seen to be outliers (*) with unusually high function values.

The success of this methodology was assessed using cross-validation and yielded an overall correct classification result of 69.8%. This success was similar across the groups: SM NO 71.0% (44/62) and SM YES 68.9% (46/67). It was noted that six dogs were dropped from original 135 during the analysis as a result of missing data hence these results are based on 129 dogs.

The F-diameter, Angle 2 and Angle 5 variables have been shown to be the significant parameters when aiming to discriminant between the sub-categories of CM and syringomyelia but the power of the discriminatory functions varies considerably, both within a condition and between conditions (Table 6).

As F-diameter was the most significant variable in each of the previous discriminant analyses, the strength of this variable alone as a diagnostic tool was investigated further. For the 111 dogs where a clear indication of CM NO/CM YES [n = 28, n = 83] and SM NO/SM YES [n = 56, n = 55] was known, very significant statistical differences were found in the average length of F-diameter (p<0.001) for both CM and syringomyelia groupings. From inspection of mean differences a first estimate of an appropriate cut-off for having the condition (or not) was an F-diameter of 42.5 mm. Consequently this estimate was used with 0.1 increases and decreases as a starting point in assessing the sensitivity, specificity, PPV and NPV for both CM (n = 155) and SM (n = 135). ROC analyses produced optimum CM sensitivity of 70%, with specificity 71% and, for SM, a sensitivity of 77%, with specificity of 54%, both occurring at F-diameter of approximately 41.8 mm. Figure 6 details the results. The results indicate high levels of sensitivity, above 70%, are possible for both the diagnosis of CM and syringomyelia. These high sensitivity values are however paired with much lower specificity values around 54% for SM. The PPV results for CM diagnosis was consistently high for F-diameter values in the range but were accompanied by, in general, low NPV. For SM the PPV were in general lower and comparable to the NPV, both being mainly between 60–70%.

This study supports the multifactorial nature of CM and syringomyelia and that this condition cannot be explained by a simple defect in the development of a single skull bone; it is a more complex disorder involving cranial base shortening, craniocervical junction abnormalities and other, as yet undetermined, factors not investigated in this study. As an example, raised intracranial pressure and/or a mismatch in cerebrospinal fluid (CSF) production and absorption could influence development of syringomyelia but these parameters have yet to be investigated in detail [36], [37]. This study found that decreased Angles 2 and 5, increased diameter of the occipital lobe circle (F-diameter) and, to a lesser extent, decreased Angle 3 and increased AE, BC, BD and FG were the most significant variables. However discriminant analysis could not always predict CM and syringomyelia status suggesting there are other anatomical or environmental factors which affect the development of these disorders. Dogs with CM and syringomyelia were more likely to have increased F-diameter i.e. greater cranial height. This is likely an apparent compensation for cranial base shortening.

It is well recognized that brachycephalic dogs have a compensatory increase in cranial height to allow accommodation of the brain in the foreshortened skull [19], [38]. Indeed in their breeding programs, many Griffon Bruxellois breeders select for a “bombe” (rounded forehead) characteristic (Figure 7) (Lee Pieterse personal communication); by doing so it is possible they are selecting for increased risk of CM and syringomyelia.

Line AE was also significantly greater for dogs with CM and syringomyelia. It is possible that this reflects increased height of the rostral part of the caudal cranial fossa (pars rostralis).

It has been shown previously that Cavalier King Charles spaniels with CM have a relatively large volume pars rostralis and relatively small volume pars caudalis [39]. The authors suggested that, in this breed, occipital bone development is insensitive to changes in hindbrain volume and that there is a compensatory bulging of the tentorium cerebelli in a rostral direction [39]. However, an increase in line AE could also occur because of its dependence on the position of the occipital lobes relative to the cerebellum. In dogs with CM the cerebellum appeared to be invaginated under the occipital lobes for example, compare the three dimensional images of Dog A with Dogs D and H in Figure 2. A more acute Angle 2 and longer Line AE with increased rostral cranial fossa height (F-diameter) was also observed in dogs with syringomyelia but without CM e.g. Dog F (Figure 2). Using the term ‘mild’ CM is perhaps misleading if applied to risk of developing syringomyelia. Figure 8 is a schematic representation of the compensatory changes in skull dimensions when there is craniosynostosis or shortening of certain skull bones. The sequence demonstrates how brain parenchyma reorganizes with an increase in cranial height and axial tilt so that the olfactory bulbs shift ventrally and the cerebellum is pushed rostrally. This rotation can mean that measurements between certain fixed points can increase, perhaps explaining the increased length of AE and BC in this study.

Dogs with CM and syringomyelia were also more likely to have smaller Angles 2, 3 and 5. A more acute Angle 2 could result from a short basicranial axis in the caudal cranial fossa and (even more) premature fusion of the spheno-occipital synchondrosis. Angle 2 was uniquely significant to CM supporting this observation. Angle 2 could also be more acute if the atlas were closer to the occiput either because the supraoccipital bone had lost is convexity i.e. was less rounded or in the event of atlanto-occipital overlapping/proximity. Proximity of the atlas to the occiput would decrease the overall volume of the craniocervical junction; indeed, atlanto-occipital overlapping would not be a prerequisite for this to occur. Reduction in the volume of the craniocervical junction would decrease the size of the subarachnoid space and cisterna magna. This in turn could increase impedance to normal biphasic and pulsatile CSF flow resulting in inhomogeneous flow with abnormal CSF peak velocities predisposing to symptomatic CM and development of syringomyelia. However this hypothesis is not supported by the finding of a significant increase in length of line BC (basion of basioccipital bone to the cranial dorsal laminae of the atlas) which tended to be longer for both CM and syringomyelia groups. In addition Angle 3 tended to be smaller with CM and syringomyelia affectedness - a more acute Angle 3 would occur if the atlas was further away from occiput. However, both BC and Angle 3 was less useful for discriminating dogs with CM and syringomyelia and there are possible alternative explanations for the increase in line BC (Figure 8). A more acute Angle 5 would occur if there was foreshortening of the sphenoid complex suggesting possible insufficiency of this bone or craniosynostosis of the intersphenoidal synchondrosis. A more acute Angle 5 was significant for both syringomyelia and CM and this finding suggests that these disorders in the dog are not just a consequence of a small volume caudal cranial fossa but shortening of the entire cranial base.

The conformation changes with CM and syringomyelia are summarized in figure 9 which was constructed from the MRI of ‘normal’ Dog A (CM0 SM0) with the significant lines and angles colored blue. The significant lines and angles of Dog D (CM2 SM2) colored red are superimposed on top. One can appreciate the increased height of the rostral cranial cavity (greater diameter of the occipital lobe circle) and decreased in size of angles 2 and 5 (yellow) in compensation for shortening of the entire cranial base and increased proximity of the atlas to the occiput. In a previous study we showed that there is also compensation by lengthening of the parietal bone [3]. Other studies have shown that brachycephalic dogs are more likely to have ventral orientation of the olfactory bulbs [40], [41]. This ventral orientation can be appreciated in the three dimensional images (Figure 2). Our study also suggests that with CM the cerebellum is invaginated under the occipital lobes.

One of the aims of this study was to develop a system of measurements that could be used on MRI obtained by machines of differing field strength and which do not include the entire forebrain. This is necessary because for “low cost” MRI screening a limited but economic imaging is obtained to ascertain CM and syringomyelia status prior to breeding [31]. A typical “low cost” MRI covers an area from the inter-thalamic adhesion to the C4/C5 intervertebral disc space and is often performed on a low field MRI machine. Only sagittal images of the brain are obtained; transverse images are obtained of the spinal cord. Consequently the rostral forebrain is often absent and there are no transverse or dorsal images of the cranium meaning that many measurements are not possible; for example, accurate determination of the skull and parenchyma volume as described in other studies [6]–[9], [30], [39], [42]. We found that using diameter of the occipital lobe circle (F-diameter) allowed an assessment of the height of the cranial cavity even if only the occipital lobes are present. F-diameter alone was a more useful tool for identifying syringomyelia than for CM. A cut-off of 41.8 mm for the F-diameter measure correctly diagnosed around two-thirds of syringomyelia cases with PPV of 61.9% and a NPV of 70.6%. For CM although 85.5% of those known to have the malformation would have been correctly diagnosed when the F-diameter cut-off was 41.8 mm, the NPV was less valuable with approximately half [42.9%] of dogs without CM correctly identified.

Angles proved more useful than point-to-point distant measurements because the latter are affected by head and body size, which is variable even within the same breed. For example we found that the standard error of the caudal skull base length (AB) was almost as large as the actual mean difference hence it was a non-significant. By contrast angles may represent a more accurate representation of the position of one anatomical feature relative to another. However a possible weakness of this study is that the size, weight and body condition of the dogs was unknown and sexual dimorphism was not analyzed which might have affected some parameters.

This study achieved its aim which was to establish a method of qualitative analysis of MRI DICOM images to enable better phenotyping and therefore genotyping. However, it failed to identify a sensitive and specific measurement(s) to improve the current British Veterinary Association and Kennel Club MRI health scheme and better diagnose CM and the risk of syringomyelia at an early age thus precluding the need for dogs to undergo multiple MRI evaluations throughout life. This is most likely because diseases like CM and syringomyelia have a complex etiology and disease expression probably occurs when there is a threshold combination of genetic and environmental factors. Therefore further work is required and we recommend that this study be repeated and/or modified for other breeds with a high prevalence of CM and syringomyelia such as Cavalier King Charles spaniels and Chihuahuas.