Benchmark Algorithm for Surface Electromyography-Based Periodic Limb Movement Detection

Periodic limb movement disorder (PLMD) is a sleep disorder characterized by repetitive involuntary leg movements (LMs) during sleep called periodic limb movements (PLMs). In clinical settings, PLMs are detected through a two-stage procedure comprising LM detection and PLM classification based on surface electromyography (sEMG) recordings from the bilateral tibialis anterior. Although there is a scoring manual for LM detection and PLM classification, the strict interpretation of this may result in the underestimation of LMs due to the lack of intrinsic sEMG characteristics such as rapid amplitude fluctuations. Thus, current LM detection relies on manual labeling with subjectivity, and no unified LM labeling algorithms exist as a benchmark. As the first step toward establishing an automatic PLM labeling benchmark algorithm that is compatible with the scoring manual used in the clinical settings, we propose an LM detection algorithm that evaluates sEMG by the 0.5 s segment unit using a percentage parameter. In the evaluation using PSG-measured sEMG from 20 individuals in clinical settings, our algorithm achieved the highest F1 score of 0.751 when the percentage parameter was set as 40%, with a recall of 0.895 and a precision of 0.604. Since we confirmed strict interpretation of the scoring manual led to an underestimation of LMs (i.e., recall and precision of 0), the proposed algorithm is considered to have performed well.Clinical Relevance-This study aims to establish a standardized benchmark algorithm to reduce subjectivity in manual labeling and improve the reproducibility of LM detection while maintaining compatibility with the scoring manual used in clinical settings.