Deep transfer learning can be used for the detection of hip joints in pelvis radiographs and the classification of their hip dysplasia status.

Reports of machine learning implementations in veterinary imaging are infrequent but changes in machine learning architecture and access to increased computing power will likely prompt increased interest. This diagnostic accuracy study describes a particular form of machine learning, a deep learning convolution neural network (ConvNet) for hip joint detection and classification of hip dysplasia from ventro-dorsal (VD) pelvis radiographs submitted for hip dysplasia screening. 11,759 pelvis images were available together with their Fédération Cynologique Internationale (FCI) scores. The dataset was dicotomized into images showing no signs of hip dysplasia (FCI grades "A" and "B", the "A-B" group) and hips showing signs of dysplasia (FCI grades "C", "D," and "E", the "C-E" group). In a transfer learning approach, an existing pretrained ConvNet was fine-tuned to provide models to recognize hip joints in VD pelvis images and to classify them according to their FCI score grouping. The results yielded two models. The first was successful in detecting hip joints in the VD pelvis images (intersection over union of 85%). The second yielded a sensitivity of 0.53, a specificity of 0.92, a positive predictive value of 0.91, and a negative predictive value of 0.81 for the classification of detected hip joints as being in the "C-E" group. ConvNets and transfer learning are applicable to veterinary imaging. The models obtained have potential to be a tool to aid in hip screening protocols if hip dysplasia classification performance was improved through access to more data and possibly by model optimization.

Radiographic, ultrasonographic, and anatomic assessment of femoral trochlea morphology in red foxes (Vulpes vulpes).

OBJECTIVE: To compare repeatability and equivalency of measures of femoral trochlea depth and trochlear angle in red foxes (Vulpes vulpes) determined by use of radiography, ultrasonography, and digital photography of cadaver limbs. SAMPLE: 24 pelvic limbs from 12 red fox cadavers. PROCEDURES: Cranioproximal-craniodistal oblique (skyline) and lateromedial radiographic views of the stifle joint and ultrasonographic images at 5 locations along the femoral trochlea were used in the study. Spacing of the 5 locations was determined on the basis of patellar position with the stifle joint at various caudal angles ranging from 96° to maximal extension (approx 170°). Ultrasonographic measurements were compared with those obtained at matched locations on photographs of anatomic preparations. Trochlear depth was assessed with all 3 image formats, and trochlear angle (measured between the trochlear ridges and sulcus) was assessed on radiographs and ultrasonographic images. Patellar thickness was measured on radiographs. Values obtained were compared by means of ANOVA, modified Bland-Altman plots, and repeatability testing. RESULTS: Depth measurement repeatability was considered good for all modalities. Small but significant differences between mean ultrasonographic trochlear depth and anatomic (photographic) measurements were found at 3 locations; 95% limits of agreement for paired anatomic and ultrasonographic measurements were wide. The ratio of trochlear depth to radiographic patellar thickness was approximately 30% for all modalities. Trochlear angle measurements were more variable than trochlear depth measurements, especially in the distal aspect of the trochlea. CONCLUSIONS AND CLINICAL RELEVANCE: Paired anatomic and ultrasonographic measurements did not appear equivalent in this study, possibly attributable to imprecise probe location, which could limit quantitative use of ultrasonography in assessing proximal trochlear depth in a clinical setting.

Measurement repeatability of tibial tuberosity-trochlear groove offset distance in red fox (Vulpes vulpes) cadavers.

OBJECTIVE: To describe CT image reconstruction criteria for measurement of the tibial tuberosity-trochlear groove (TT-TG) offset distance, evaluate intra- and inter-reconstruction repeatability, and identify key sources of error in the measurement technique, as determined in vulpine hind limbs. ANIMALS: 12 red fox (Vulpes vulpes) cadavers. PROCEDURES: CT images of each hind limb in intact cadavers were obtained; at 1-week intervals, 3 reconstructions were performed that were based on 1 plane passing through the centers of the femoral head and medial condyle and parallel to the caudal femoral condyles, 1 plane aligned with the femoral trochlea, and a third orthogonal plane. Randomized and anonymized reconstructions were assessed for TT-TG offset distance with a single-image technique by 1 observer, and inter-reconstruction repeatability and intra- and inter-reconstruction measurement repeatability were assessed via the repeatability coefficient and intraclass correlation coefficient. RESULTS: Multiplanar reconstructions of hind limb images were repeatedly made to within a few degrees of each other. Intra- and inter-reconstruction repeatability for TT-TG offset distance measurement was good. Repeatability was most affected by accurate identification of the tibial tuberosity and femoral trochlea landmarks. CONCLUSIONS AND CLINICAL RELEVANCE: Results obtained from vulpine hind limb CT images indicated that reconstructions can be made with a high degree of repeatability when based on strictly defined and applied criteria. The TT-TG offset distance has potential as an objective assessment of alignment of the distal portion of the quadriceps mechanism; its use as an aid in case selection for corrective femoral osteotomy among dogs with medial patellar luxation warrants investigation.

A new technique for modeling of hematogenous osteomyelitis in pigs: inoculation into femoral artery.

A new inoculation technique has been developed and applied in a porcine model of juvenile hematogenous osteomyelitis. Following the success of the model, we describe the inoculation technique in detail to enable its replication in future studies. The technique was based on an anatomical feature of the femoral artery that enables inoculation into the artery using a simple surgical procedure. Inoculation in the femoral artery is advantageous because the localization of lesions constitutes a discriminative model of the naturally occurring hematogenous osteomyelitis in long bones, usually involving femur and tibia in children. The procedure was performed under general anesthesia and consisted of five major steps: (1) Exposure of the right femoral artery, (2) retrograde catheterization, (3) inoculation of bacteria, (4) hemostasis of the arterial puncture site using compression, and (5) suturing of the wound in two layers.

Five patellar proximodistal positioning indices compared in clinically normal Greenland sled dogs.

Patellar luxation in large-breed dogs is associated with abnormal proximodistal patellar positioning. Using a clinically normal population of Greenland sled dogs, measurement reliability and the effect of limb position were compared for five patellar proximodistal positioning indices based on the Insall-Salvati (IS), modified Insall-Salvati (mIS), de Carvalho (dC), patellotrochlear (PT) and Blackburne-Peel (BP) indices. Indices were measured at one knee angle in 44 dogs and two knee angles in 10 dogs. Index susceptibility to error was modelled for different errors in knee angle estimation. Two reported techniques for determining knee angle were compared in a fox hind limb model. Indices dC and PT were significantly affected by knee angle (P<0.001). Error susceptibility was the lowest for IS and the greatest for PT. Intra- and inter-observer agreements were moderate to substantial for all indices. Measurement precision was good for all indices except BP. Patellar ligament laxity significantly affected IS, mIS, dC and BP (P<0.05). Knee angle measurements were technique dependent, with a bias of 9° to 13° and limits of agreement of ±5°. All five indices were reliable, but precision varied. For pre- and post-operative comparison, dC showed less error susceptibility than PT and better precision than BP. An ideal index for clinical use remains to be defined; the best option is dC, although with limitations. The method of knee angle measurement must be defined prior to index measurement and comparison of index values. Index values may vary with species and/or body size.

Comparison of reliability of five patellar position indices at various stifle joint angles in pelvic limbs obtained from cadavers of red foxes (Vulpes vulpes).

OBJECTIVE: To compare 5 patellar position indices at various stifle joint angles in cadavers of red foxes, determine measurement reliability, and assess the suitability of these indices for clinical use. SAMPLE: Pelvic limbs from cadavers of 12 red foxes (Vulpes vulpes). PROCEDURES: Patellar position in each limb at 7 stifle joint angles (30° to 148°) was assessed by use of the Insall-Salvati (IS), modified Insall-Salvati (mIS), de Carvalho (dC), patellotrochlear (PT), and Blackburne-Peel (BP) indices. RESULTS: Values for all indices varied significantly on the basis of joint angle, but for IS and mIS indices, this was minor and nonsignificant between 52° and 130° and between 52° and 148°, respectively. The dC index increased linearly, and PT and BP indices varied polynomially with increases in stifle joint angle. Stifle joint angles measured from radiographs agreed well with the goniometrically set stifle joint angles up to approximately 100° and diverged thereafter. Intraobserver and interobserver agreement was substantial for all indices, and IS index was the most precise. CONCLUSIONS AND CLINICAL RELEVANCE: IS and mIS index values were effectively independent of stifle joint angle, in contrast to dC, PT, and BP indices. The BP index varied nonsignificantly across a range of joint angles. To maximize angular accuracy, radiographs should not be obtained at joint angles > 100°. Although dC, PT, and BP indices appeared to be suitable for preoperative and postoperative evaluation of patellar position, BP index appeared to have the most promise for determination of patellar position in clinical applications.

The quadriceps angle: reliability and accuracy in a fox (Vulpes vulpes) pelvic limb model.

OBJECTIVES: To evaluate the effect of measurement technique and limb positioning on quadriceps (Q) angle measurement, intra- and interobserver reliability, potential sources of error, and the effect of Q angle variation. STUDY DESIGN: Cadaveric radiographic study and computer modeling. ANIMALS: Pelvic limbs from red foxes (Vulpes vulpes). METHODS: Q angles were measured on hip dysplasia (HD) and whole limb (WL) view radiographs of each limb between the acetabular rim, mid-point (Q1: patellar center, Q2: femoral trochlea), and tibial tuberosity. Errors of 0.5-2.0 mm at measurement landmarks alone and in combination were modeled to identify the effect on Q angle. The effect of measured Q angles on the medial force exerted on the patella (F(MEDIAL)) was calculated. RESULTS: The HD position yielded significantly (P < .001) more medial Q angles than the WL position. No significant difference was observed between Q1 and Q2, but Bland-Altman plots indicated they were not equivalent. Intra- and interobserver agreement was substantial. Q2 errors were inherently greater than Q1: the mid-point and tibial tuberosity are the most important sources of Q angle variability. Increasing Q angles significantly increased the exerted F(MEDIAL) (P < .0001, gradient 1.7%). CONCLUSIONS: Measurements are reliable, but Q2 is more prone to error than Q1, and the 2 measurement techniques are not interchangeable. Positional errors must be kept below 1.3 mm (Q1) or 0.8 mm (Q2).

Modulation of the intramedullary pressure responses by calcium dobesilate in a rabbit knee model of osteoarthritis.

BACKGROUND AND PURPOSE: The presence of bone marrow edema in patients with osteoarthritis is associated with pain and disease progression. Management of bone edema with the synthetic prostacyclin iloprost may be complicated by side effects. Calcium dobesilate, a treatment for chronic venous disease, shares some pharmacological actions with iloprost but appears to be better tolerated. Anecdotal reports have suggested that calcium dobesilate may be useful for medical management of osteoarthritis, possibly by reducing bone marrow edema, and this study was performed to investigate possible benefits of treatment. METHODS: The effects of a 6-week period of oral calcium dobesilate administration on tibial intramedullary pressure dynamics and physical joint characteristics were evaluated in 20 rabbits with unilaterally induced knee osteoarthritis that were randomly allocated to either a treatment group or a placebo control group. Treatment or placebo started 8 weeks after induction of osteoarthritis, and was followed by a 4-week washout period. RESULTS: Calcium dobesilate did not affect joint thickness or range of motion, nor individual pressure measurements, compared to placebo. Pressure ranges in the operated limb were greater than in the intact limb after 8 weeks, and approached those of the intact limb after 6 weeks of treatment with calcium dobesilate but not with placebo. Inter-limb differences were lower (p = 0.02) in the dobesilate group following the washout period. INTERPRETATION: Calcium dobesilate had a detectable effect on pressure dynamics in the subchondral bone of osteoarthritic joints in this model. The significance of these effects for pain and function should be established.

Measurement of body composition in cats using computed tomography and dual energy X-ray absorptiometry.

Dual energy X-ray absorptiometry (DEXA) is a reference method for assessing body composition but is seldom `accessible in veterinary settings. Computed tomography (CT) can provide similar body composition estimates and we propose that it can be used in body composition studies in animals. We compared CT and DEXA data from 73 healthy adult neutered domestic cats. Three approaches for measuring adipose tissue percentage from full-body CT scans were explored. By examining the frequency distribution of voxels by Hounsfield unit (HU) value, it is possible to calculate a fat index (Fat%) that is in close agreement with the fat percentages obtained from DEXA scans. Fat% values obtained by the best of the methods had a mean difference of 0.96% (95% confidence interval 0.33-1.59%) from the DEXA results. Fat% obtained by the other two methods were characterized by good correlation but poor agreement and in one of the methods, the difference between the values from the two modalities was proportional to their mean. By using CT, it is possible to obtain body composition estimates that are in close agreement with those available using DEXA. While the significance of individual Fat% measurements obtained from CT can be difficult to interpret and to compare between centers, CT can contribute to research studies concerned either with nutrition or with obesity-related disorders.

Web-based teaching tool incorporating peer assessment and self-assessment: example of aligned teaching.

OBJECTIVE: Our objective was to create an electronic learning tool that incorporates the pedagogic advantages of peer assessment and self-assessment into report-generating skills. CONCLUSION: The tool was created using Web programming software. It was tested with 12 veterinary students and provided peer-peer and tutor-student interactions that supported learning. The tool is suited for training any professional concerned with the interpretation of images, particularly resident radiologists.

Computer tomographic investigation of subcutaneous adipose tissue as an indicator of body composition.

BACKGROUND: Modern computer tomography (CT) equipment can be used to acquire whole-body data from large animals such as pigs in minutes or less. In some circumstances, computer assisted analysis of the resulting image data can identify and measure anatomical features. The thickness of subcutaneous adipose tissue at a specific site measured by ultrasound, is used in the pig industry to assess adiposity and inform management decisions that have an impact on reproduction, food conversion performance and sow longevity. The measurement site, called "P2", is used throughout the industry. We propose that CT can be used to measure subcutaneous adipose tissue thickness and identify novel measurement sites that can be used as predictors of general adiposity. METHODS: Growing pigs (N = 12), were each CT scanned on three occasions. From these data the total volume of adipose tissue was determined and expressed as a proportion of total volume (fat-index). A computer algorithm was used to determined 10,201 subcutaneous adipose thickness measurements in each pig for each scan. From these data, sites were selected where correlation with fat-index was optimal. RESULTS: Image analysis correctly identified the limits of the relevant tissues and automated measurements were successfully generated. Two sites on the animal were identified where there was optimal correlation with fat-index. The first of these was located 4 intercostal spaces cranial to the caudal extremity of the last rib, the other, a further 5 intercostal spaces cranially. CONCLUSION: The approach to image analysis reported permits the creation of various maps showing adipose thickness or correlation of thickness with other variables by location on the surface of the pig. The method identified novel adipose thickness measurement positions that are superior (as predictors of adiposity) to the site which is in current use. A similar approach could be used in other situations to quantify potential links between subcutaneous adiposity and disease or production traits.

Influence of age and position on the CT number of adipose tissues in pigs.

The location of adipose tissue depots is important in determining their significance. Research into the physical and chemical differences between these depots is therefore of interest. Using image analysis, this paper examines the influence of location on the linear attenuation coefficient of adipose tissue for X-rays, in computed tomography (as indicated by CT number) at three time points. Nine pigs were CT scanned on three separate occasions approximately 1 month apart. The mean CT number was -78, -100, and -104 for visceral adipose tissue (VAT) from the first to the final scan, respectively. The corresponding CT numbers for subcutaneous adipose tissue (SAT) were -80, -101, and -106. There was a significant difference between the CT numbers at each location at each scan (P values from 0.025 to <0.001) and between the CT numbers for each location at different times (P < 0.05). In a separate analysis of the final scan session, the mean CT number of adipose tissue at increasing distances from a mathematically defined center of the animal was determined. Regression analysis showed that the CT number of adipose tissue decreases with increasing distance from the animal's center (y = -102.7 - 0.04 x, P < 0.001, where y is the predicted CT number for adipose tissue, from the animal center (x = 0) to the skin (x = 100)). It can thus be expected that the overall mean CT number for adipose tissue can be used as an indicator of the relative quantities of adipose tissue at each location if the mean for each is known.