0705 Detection of Periodic Limb Movements During Sleep in RLS Using Tonic Motor Activation System and Machine Learning

Abstract Introduction Periodic limb movements during sleep (PLMS) are the most important objective correlate of restless legs syndrome (RLS) and, thus, could be beneficial for long-term monitoring and management. We aimed to develop a machine learning (ML) model to detect PLMS based on inertial measurement unit (IMU) movement sensors within an existing therapeutic device for refractory RLS – the tonic motor activation (TOMAC) system. Methods Twenty-two individuals with RLS completed a total of 32 nights of polysomnography (PSG) while wearing bilateral TOMAC units externally on the lower legs at the head of the fibula. Each TOMAC unit continuously recorded accelerometer and gyroscope data from a 6-axis IMU. Features including signal statistics, frequency domain characteristics, and positions were extracted from each IMU sensor in 2-second epochs throughout the night. Features were aggregated into 60-minute windows, input into a random forest ML model, and evaluated using leave-one-out cross-validation. Ground truth PLMS assessment was based on tibialis anterior EMG collected as part of PSG and technician-scored without excluding movements related to respiratory events. Classification metrics, Pearson correlations, and Bland-Altman plots were used to evaluate agreement, variability, and bias between ML model predictions and ground truth, based on ability to estimate periodic limb movement index (PLMI) for each hour and ability to screen for pathological PLMS based on the diagnostic cutoff PLMI >15/hour. Results For calculating PLMI for each hour (n= 215 hours), ML model predictions were highly correlated with ground truth (r= 0.890) and showed minimal bias (mean ± SD= 0.11 ± 21.05). For classifying potential presence/absence of pathological PLMS, ML model predictions had high accuracy (0.907), specificity (0.958), and area under the curve (AUC, 0.841). Conclusion The TOMAC IMU feature-based ML model showed strong agreement with PSG-based ground truth for calculating PLMI and screening for pathological PLMS. This suggests that the therapeutic TOMAC system might also offer the possibility for long-term, in-home PLMS monitoring. Support (if any) Sponsored by Noctrix Health and supported by NIH/NINDS R44NS117294.