A Multi-Scale CNN-LSTM Network with Squeeze and Excitation Block for Leg Movement Detection during Sleep

Detection of leg movements during sleep is crucial for diagnosing periodic limb movement syndrome and restless leg syndrome. However, there are significant differences in the manual scoring of leg movement in nocturnal polysomnography studies. Therefore, we propose a CNN-LSTM network specifically designed for classifying sleep leg movement events. By leveraging the strengths of convolutional neural networks (CNN) and long short-term memory (LSTM) networks, our approach effectively captures both local and global features for enhanced feature representation. Additionally, a multi-scale convolutional (MC) module and the squeeze and excitation (SE) module are introduced with the purpose of extracting features at different scales and frequencies and weakening redundant features for better performance. Experimental results demonstrate that our network outperforms previous baseline methods in detecting sleep leg movement events on the pubulic dataset.