XGBoost-SHAP-based interpretable diagnostic framework for knee osteoarthritis: a population-based retrospective...

Area of Science:

• Orthopedics
• Medical Informatics
• Machine Learning

Background:

• Knee osteoarthritis (KOA) is a prevalent degenerative joint disease.
• Accurate and early diagnosis of KOA is crucial for effective management and prevention strategies.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model for diagnosing KOA using routine demographic and clinical data.
• To identify key features contributing to KOA diagnosis.

Main Methods:

• A retrospective, population-based cohort study using questionnaire data from the Wu Chuan KOA Study.
• Feature selection, class balancing, and comparison of four ML classifiers (XGBoost with Boruta identified as best).
• Model performance evaluated using AUC, G-means, and F1 scores; feature importance determined by Shapley values.

Main Results:

• The study included 1188 participants, with 26.3% diagnosed with KOA.
• XGBoost with Boruta achieved the highest performance (AUC: 0.758, G-means: 0.800, F1: 0.703).
• The average experience of joint pain was identified as the most important feature for KOA diagnosis among the top 17 ranked features.

Conclusions:

• Machine learning models are effective in identifying crucial factors for KOA diagnosis.
• The findings can inform new prevention strategies for knee osteoarthritis.
• Further validation of this ML approach is recommended.