Deciphering Knee Osteoarthritis Diagnostic Features With Explainable Artificial Intelligence: A Systematic Review

Existing artificial intelligence (AI) models for diagnosing knee osteoarthritis (OA) have faced criticism for their lack of transparency and interpretability, despite achieving medical-expert-like performance. This opacity makes them challenging to trust in clinical practice. Recently, explainable artificial intelligence (XAI) has emerged as a specialized technique that can provide confidence in the model's prediction by revealing how the prediction is derived, thus promoting the use of AI systems in healthcare. This paper presents the first survey of XAI techniques used for knee OA diagnosis. This survey identified 78 AI-based primary knee OA diagnostic test accuracy studies, of which 70 (89.7%) employed XAI. In 34 out of 70 (48.6%) of studies, XAI was utilized for the goal of visualization of predictions. Gradient-weighted class activation mapping (GradCAM) is the most common technique, being used in 24 out of 70 studies (34.3%), followed by SHapley Additive exPlanations (SHAP), being used in 9 out of 70 (12.9%) studies. All included studies analyzed the outcomes generated by XAI methods through qualitative analysis. However, only three studies utilized quantitative measures to evaluate the reliability of the XAI outcomes. We also observed that 64.3% of the studies utilized widely-circulated dataset, namely Osteoarthritis Initiative (OAI) extensively.The XAI techniques are discussed from two perspectives: data interpretability and model interpretability. Our paper provides an overview of XAI's potential towards a more reliable knee OA diagnosis approach and helps to encourage its adoption in clinical practice.