Validating Joint Acoustic Emissions Models as a Generalizable Predictor of Joint Health

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rheumatology

Background:

• Joint acoustic emissions (JAEs) offer a non-invasive method for assessing joint health.
• Advancements in sensing hardware enable at-home JAE monitoring.
• Existing JAE models require improvement for generalizability.

Purpose of the Study:

• To investigate the impact of recording setup, location, and participant population on JAE models.
• To enhance the generalizability of JAE models for predicting joint health conditions.
• To establish JAEs as a reliable measure for joint health assessment.

Main Methods:

• Evaluated a JAE model for predicting erythrocyte sedimentation rate (ESR) in rheumatoid arthritis (RA) patients using benchtop data.
• Trained feature-based and convolutional neural network (CNN) models with healthy and RA data to predict ESR.
• Tested models on data including healthy, pre-radiographic osteoarthritis (Pre-OA), and osteoarthritis (OA) populations.

Main Results:

• Benchtop model testing yielded an AUC of 0.79, sensitivity of 0.73, and specificity of 0.81.
• Feature-based model achieved AUCs of 0.69 (Pre-OA) and 0.94 (OA).
• CNN model demonstrated superior performance with AUCs of 0.85 (Pre-OA) and 0.99 (OA), and high sensitivity and specificity.

Conclusions:

• JAE models can be generalized across different recording setups, locations, and participant populations.
• Validated JAE models show potential for accurate detection of Pre-OA and OA.
• This work provides a foundation for using JAEs as a scalable, non-invasive joint health monitoring tool.