Prof. Zhou Yi’s Team Achieves New Progress in Sleep Staging Research

Sleep staging is of paramount importance for analyzing sleep quality and diagnosing sleep-related brain disorders. With the in-depth implementation of the "China Brain Project" and the preliminary strategic layout for brain science and brain-inspired research in the upcoming national "15th Five-Year Plan," precise, non-invasive, and home-based assessment of brain functional states has become a major proposition at the intersection of neuroscience and artificial intelligence.Polysomnography (PSG) remains the standard method for sleep staging, monitoring sleep states by recording multiple physiological signals, including electroencephalogram (EEG), electrocardiogram (ECG), electrooculogram (EOG), and electromyogram (EMG). However, the analysis of these signals is heavily influenced by experience and subjective states, often leading to inconsistencies in interpretation. Given that manual analysis of PSG signals is both time-consuming and labor-intensive, many recent studies have utilized deep learning techniques for sleep staging. However, these methods often show limited adaptability to downstream tasks, rely heavily on complex multi-channel or EEG data, and require equipment that can cause discomfort to subjects. In contrast, single-lead signals offer a more convenient and comfortable alternative for sleep staging classification.

Recently, Prof. Zhou Yi’s team from our school published a research paper titled "SleepECGFusion: A Cross-Modal Deep Learning Framework for Automatic Sleep Stage Classification using Single-Lead ECG" online in Knowledge-Based Systems, a CAS Q1 Top Journal. The study proposes an innovative cross-modal deep learning framework named SleepECGFusion, which successfully integrates information from two complementary domains. Furthermore, the study verifies the hypothesis that increasing the duration of input signals consistently improves classification accuracy across all sleep stages. SleepECGFusion achieved superior performance in ECG-based sleep staging tasks compared to previous studies and demonstrated excellent transferability. Most importantly, the framework provides robust and comparable classification results for both healthy individuals and patients with sleep apnea. Future research aims to integrate more physiological signals to enhance robustness in highly disturbed sleep environments, explore time-contrastive learning methods to better handle temporal variability in long-term monitoring, and evaluate its application in other sleep-related neurological disorders.

Prof. Zhou Yi is the corresponding author of this paper, and Xuanhao Qi, a doctoral student (Class of 2024), is the first author. Prof. Zhou Yi’s team has long been engaged in research related to health and medical informatics, big data, and medical artificial intelligence. Postdoctoral fellows and graduate students interested in these research directions are warmly welcomed to join the team.