MCTSleepNet: transformer-based sleep classification with multimodal fusion

Automatic sleep stage classification plays a crucial role in assessing sleep quality and diagnosing sleep disorders. While many automated methods exist for sleep stage classification, most rely solely on single-channel electroencephalogram (EEG) signals. In contrast, multi-channel polysomnography (PSG), which includes EEG, electrooculogram (EOG), electromyogram (EMG), and electrocardiogram (ECG) signals, provides more comprehensive and accurate sleep staging information. These signals not only reveal complex physiological changes during sleep but also aid in understanding the characteristics and variations of different sleep stages. This capability offers valuable support for both sleep research and clinical practice. Therefore, we propose MCTSleepNet, a Transformer-based model for multi-channel sleep staging. It takes signal images transformed via time-frequency features as input. The core architecture uses a Transformer encoder to capture joint features across multiple channels, complemented by residual connections to retain original input information. The experimental results demonstrate that our model outperforms other state-of-the-art techniques on the publicly available ISRUC-S3 dataset. MCTSleepNet is an effective multi-channel sleep staging method that significantly aids in clinical sleep staging tasks.