Detection and Classification of Canine Hip Dysplasia According to FCI Grading System Using 3D CNN’s

Dysplasia refers to abnormal growth or development that worsens with time. The current standard diagnostic techniques involves harsh radiation methods, ultrasound imaging of the hip, and metric extraction. This has been shown to be unreliable due to human error, in probe positioning, resulting in misdiagnosis, time-consuming, and tiresome labor. In comparison to a normal 2D CNN, MRI with 3D CNN has been offered as a more effective option since it can examine the whole set of 3D MR images as a single unit. In this paper, we developed a deep learning model that can classify canine hip dysplasia according to its severity using 3D sequences of hip joint magnetic resonance data. The severity of each hip was graded on a scale of A–E by the Federation Cynologique Internationale (FCI). We used the Danish Kennel Club dataset, which included 11,759 ventro-dorsal pelvic images (23 518 hip joint images), with X-ray and MRI images accessible for each hip joint. In addition, to assist breeders discover better and healthier parents from their stock and to prevent hip dysplasia in future generations, Another model was trained by reclassifying the samples into "non-dysplastic" (A+B) and "dysplastic" (C–E) groups. When compared to earlier models, our models attain an accuracy of 89.7% and 70.0% respectively, and outperform in terms of computing time and performance. This also shows that 3DCNN’s have a greater potential of improving diagnostic accuracy and may be employed as a clinical help in veterinary medicine for hip dysplasia than traditional X-ray approaches.