Sleep stage classification using convolutional neural networks and bidirectional long short-term memory

Abstract

Classification of sleep stage is very useful to detect the occurrence of sleep apnea. This classification requires mechanisms that automatically and efficiently process polysomnography data. However, the process requires a system to be able to extract the relevant features which are then used to classify the sleep stage. The best solution is sequence classification because it not only concerns the contents of each segment or the sequence of data. One of the best order-based identifiers today is Long Short-Term Memory (LSTM). The LSTM can only update for forwarding directions. To process the data in two directions, it implemented Bidirectional Longs Short Term Memory (Bi-STM). Also, the implementation also applies Convolutional Neural Networks (CNN) as a feature learning before using Bi-LSTM. The result shows that F-measure Bi-LSTM is better than LSTM but use CNN as a learning attribute for Bi-LSTM cause an F-measure decrease.