Physiological Features for Classification of Different Types of Peripheral Neuropathy Using Multimodal Wearable Sensors

Conventional screening methods for diagnosing peripheral neuropathy (PN) are often complex, time-consuming, and require costly clinical procedures. Moreover, PN presents as a heterogeneous group of clinical syndromes with overlapping symptoms, posing additional challenges for accurate identification. This study proposes a non-invasive, wireless, and wearable-sensor-based system for the classification of commonly occurring types of PN, specifically diabetic peripheral neuropathy (DPN) in the lower limbs and cardiac autonomic neuropathy (CAN) in clinical settings. Given that both conditions can alter gait dynamics, the proposed method analyzes human gait and plantar pressure during motion and correlates post-exercise cardiac activity by extracting physiological markers. Data were collected using two inertial measurement units (IMUs) and a foot plantar pressure sensor with eight sensing points to capture spatiotemporal and kinematic gait features, alongside an electrocardiogram (ECG) sensor to acquire heart rate variability (HRV) parameters at 100 Hz. A 5-meter walk test was conducted with 20 healthy individuals, 32 participants with DPN, and 13 with CAN. Extracted features were statistically evaluated using Chi-Square and Cramer’s V tests, demonstrating strong associations between gait, pressure, HRV parameters, and PN classifications. Various machine learning classifiers were trained and evaluated, with the Gradient Boosting classifier achieving the highest cross-validation accuracy of 98.27% and holdout accuracy of 94.54%. To enhance model transparency, SHAP (SHapley Additive exPlanations) analysis was employed to interpret the contributions of features to classification decisions. The proposed system offers a clinically viable, efficient, and accurate diagnostic tool for the early detection and classification of peripheral neuropathies.