An Automatic Sleep Stage Classification Approach Based on Multi-Spike Supervised Learning

摘要

Sleep stage classification can provide important reference factors in enhancing diagnosis and treatment of sleep disorders and sleep-related diseases. However, such classification tasks based on polysomnography are more complicated. This article tried a novel approach to deal with the polysomnography data by using spiking neural networks, including a coding scheme for complex spatio-temporal data and a multi-spike supervised algorithm for training our neural network to implement the learning goal of spike trains. This automatic sleep stage recognition approach can be applied to solve polysomnography data multi-classification problem and avoid manual handling whenever possible. In the experiments, the effects of the different network parameters on the sleep stage classification accuracy are analyzed, and the final results indicate that the proposed approach can be used to assist sleep diagnosis.