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

Overview

• This study developed and tested a machine learning model to assist in diagnosing temporomandibular joint osteoarthritis (TMJ OA) in horses.
• The model integrates computed tomography (CT) scan findings and synovial fluid biomarker data to distinguish healthy TMJs from those affected by osteoarthritis.

Background and Importance

• Temporomandibular joint osteoarthritis (TMJ OA) in horses is a difficult condition to diagnose clinically because the pain presentation is unique and complex.
• Traditional diagnostic methods may not be sufficient or accurate enough for reliable diagnosis.
• Computer-aided diagnostic tools that incorporate biological markers (biomarkers) and imaging data might improve diagnostic accuracy for this condition.

Study Objectives

• To develop a machine learning (ML)-based classification model that can differentiate between healthy and osteoarthritic equine TMJs.
• To evaluate the integration of CT imaging findings and synovial fluid biomarkers as input data for the ML models.

Study Design and Methods

• Sample:
  • 50 equine cadaver heads provided 82 TMJs to the study.
  • Each TMJ was classified as healthy or OA based on histological examination and CT findings.
• Data Collection:
  • Nine different CT imaging features/findings were evaluated for each TMJ.
  • Synovial fluid was collected from each TMJ for measurement of twelve biochemical biomarkers.
• Data Analysis:
  • Biomarker data were analyzed to determine correlations amongst biomarkers and with TMJ OA and age.
  • A mixed-effects logistic regression model was used to assess the influence of biomarker levels and age on OA classification.
  • Twelve ML models were tested, using combinations of:
    • Two feature selection methods to choose relevant variables.
    • Six classification algorithms to perform the binary classification task.

Key Results

• Several biomarker levels showed predominantly positive correlations with:
  • TMJ osteoarthritis presence.
  • Age of the horse.
  • Other biomarker levels.
• However, when accounting for all variables in the mixed logistic regression model, only age had a statistically significant impact on OA classification.
• The best-performing ML model achieved:
  • An accuracy of 82% in correctly classifying TMJs as healthy or OA.
  • An area under the curve (AUC) of 0.85, indicating good discrimination ability between the two categories.

Conclusions and Implications

• The study demonstrates that combining CT imaging data and synovial fluid biomarker information in a machine learning framework can effectively classify TMJ OA in horses.
• The performance of the ML model is superior to conventional diagnostic methods, suggesting its potential as a supportive clinical tool.
• Such a computer-aided diagnosis system may help veterinarians make more accurate diagnoses of equine TMJ OA, leading to improved management and treatment approaches.
• Age remains an important factor influencing TMJ joint health and should be considered in diagnosis and model interpretation.

Future Directions

• Further studies could validate the ML model in clinical settings with live horses and larger sample sizes.
• Integration of additional biomarkers or imaging modalities might enhance diagnostic accuracy further.
• Exploring the biological mechanisms linking biomarkers and TMJ OA progression could deepen understanding of disease pathology.