A robust and interpretable deep transfer learning framework on knee acoustic emissions for osteoarthritis...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Musculoskeletal Health

Background:

• Knee osteoarthritis (OA) diagnosis often relies on invasive methods.
• Knee acoustic emissions (KAEs) show potential as non-invasive OA biomarkers.
• Current KAE classifiers use conventional machine learning with limited generalizability and interpretability.

Purpose of the Study:

• To develop a robust and interpretable deep transfer learning framework for OA classification directly from raw KAE signals.
• To improve the accuracy and generalizability of KAE-based OA detection.
• To provide insights into the acoustic patterns driving OA classification using explainable AI (XAI).

Main Methods:

• A deep transfer learning framework was developed to classify OA from raw KAE signals.
• The method learns discriminative time-frequency representations.
• Explainable AI (XAI) was integrated to ensure predictions are based on physiologically plausible acoustic components.
• A diverse KAE dataset, including participants with high BMI, was used for systematic comparison with benchmark algorithms.

Main Results:

• The proposed deep transfer learning framework achieved an average accuracy of approximately 89% in distinguishing OA from healthy knees.
• The method demonstrated strong performance and stability across multiple dataset splits and initializations.
• XAI visualizations confirmed that model predictions were based on relevant time-frequency regions of the KAE signals.

Conclusions:

• Deep transfer learning offers a promising approach for accurate, interpretable, and scalable knee osteoarthritis assessment.
• The framework shows potential for non-invasive OA detection and future home monitoring applications.
• This study highlights the utility of analyzing acoustic emissions for understanding knee joint health.