Distributed Neural Network System for Multimodal Sleep Stage Detection

Existing automatic sleep stage detection methods predominantly use convolutional neural network classifiers (CNNs) trained on features extracted from single-modality signals such as electroencephalograms (EEG). On the other hand, multimodal approaches propose very complexly stacked structures with multiple CNN branches merged by a fully connected layer. It leads to high computational and data requirements. This study proposes replacing distributed system for detection. has relatively low training requirements while providing highly competitive results. The proposed classification decision-making (MM-DMS) method applies shallow network, arbitrating between outcomes given an assembly of independent networks (CNNs), each using different signal. Experiments conducted CAP Sleep Database data, including EEG-, ECG-, EMG modalities representing six stages sleep, show that MM-DMS significantly outperforms CNN. fully-connected arbitration included in traditional majority voting-, average probability-, maximum probability methods.