Computer-Aided Diagnosis of Equine Temporomandibular Joint Osteoarthritis Using Machine Learning Integrating Computed Tomography Findings and Synovial Fluid Biomarkers

Abstract

Horses presenting with temporomandibular joint (TMJ) dysfunctions are often clinically evaluated for TMJ osteoarthritis (OA). Due to the unique characteristic of TMJ-related pain, the clinical diagnosis of equine TMJ OA is challenging; however, it may be supported by computer-aided tools incorporating biomarker data. This study aims to evaluate a machine learning-based approach to address a binary classification distinguishing healthy TMJs from TMJ OA. Among 50 equine cadaver heads, 82 TMJs were included and annotated as healthy or OA based on histological and computed tomography (CT) findings. For each TMJ, nine CT findings were assessed, and synovial fluid was collected for the evaluation of twelve biomarkers. Using a biomarker dataset, correlations among biomarkers were calculated and supported with a mixed-effects logistic regression model. Using a combined dataset, twelve machine learning models, incorporating two feature selection methods and six classification algorithms, were evaluated. Specific biomarker levels showed predominately positive correlations with TMJ OA, age, and with each other; however, only age had a significant effect on OA assignment in the mixed model. The best-performing machine learning model achieved an accuracy of 0.82 and an area under the curve (AUC) of 0.85 for binary TMJ classification. The proposed classification model outperforms conventional diagnostic methods and may therefore be considered beneficial in aiding the diagnosis of equine TMJ OA.