Razi University

Intervertebral disc calcification (IVDD) represents one of the most common spinal disorders in dogs, frequently culminating in debilitating pain, progressive motor deficits, and permanent paralysis if not diagnosed and managed promptly. Despite the indispensable role of survey radiography in initial triaging, anatomical superimpositions and observer-dependent perceptual errors—especially in subacute or incipient mineralized lesions—remain substantial diagnostic hurdles. This study aimed to develop and validate a deep learning-based computer-aided screening framework for automated IVDC detection in canine lateral spinal radiographs. A well-curated dataset comprising 260 lateral views (200 calcified disc lesions and 60 normal controls) was established and validated by board-certified veterinary expertise. To suppress surrounding soft-tissue artifacts and selectively enhance local contrast across narrow intervertebral spaces without inducing pixel saturation, Contrast-Limited Adaptive Histogram Equalization (CLAHE) was implemented. Hierarchical feature representations and binary classification were achieved utilizing a deep residual network (ResNet-50) via transfer learning pre-trained on ImageNet. To prevent data leakage, mitigate overfitting, and safeguard clinical generalizability, model performance was evaluated through stratified 5-fold cross-validation, with performance indices derived from aggregated out-of-sample inferences. The optimized model demonstrated exceptional and stable diagnostic capacity, achieving an overall accuracy of 92.3%, a sensitivity of 93.0%, a specificity of 90.0%, an F1-score of 94.9%, and an area under the receiver operating characteristic curve (AUC) of 0.956. Feature vector analyses confirmed the network’s competence in distinguishing subtle mineralized disc opacities from adjacent enthesophytes, spondylosis deformans, and overlapping osseous structures. These findings substantiate the potential of the proposed pipeline as a robust clinical decision support system (CDSS) to enhance diagnostic throughput, minimize human misinterpretation, and expedite patient management in veterinary clinical settings.