Development and Multicenter External Validation of a Real-Time Artificial Intelligence Diagnostic System for Diabetic...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Neurology

Background:

• Diabetic peripheral neuropathy (DPN) is a common diabetes complication with significant morbidity.
• Current screening methods are resource-intensive and have limitations.
• Early detection of DPN is crucial for effective management and intervention.

Purpose of the Study:

• To develop and validate STRIDE, an interpretable machine learning model for DPN classification using wearable sensor data.
• To assess the performance and interpretability of the STRIDE model.
• To evaluate the potential of STRIDE for scalable and resource-efficient DPN screening.

Main Methods:

• A multicenter diagnostic study enrolled 206 participants with varying DPN statuses.
• Gait and balance parameters were captured using LEG-Sys/BalanSens wearable sensors.
• An interpretable machine learning model (STRIDE) integrating random forest and logistic regression was developed and validated.

Main Results:

• The STRIDE model demonstrated strong classification performance (Random Forest AUROC=0.80).
• Key predictive features included stride length variability and double-support time.
• External validation showed 89.3% concordance with electromyography, with predictive value for future DPN development.

Conclusions:

• STRIDE offers a promising, interpretable, and scalable approach for DPN classification.
• The model leverages biomechanical data from wearable sensors for early detection.
• STRIDE has the potential to facilitate timely interventions in diverse settings.