Answers are grounded in this paper and cite their sources.
Introduction Narcolepsy is a chronic neurological disorder characterized by dysregulated sleep–wake transitions, including abnormal rapid eye movement (REM) intrusions, sleep fragmentation, and irregular sleep-stage progression, which significantly impair daily functioning and quality of life. Accurate detection of narcoleptic sleep-stage irregularities remains clinically challenging due to the high temporal variability of electroencephalographic (EEG) signals and the subjective limitations of manual polysomnographic scoring. Existing automated methods rely primarily on raw EEG signal processing using sequential or spectral models and fail to exploit the rich spatial and temporal structure embedded in EEG visualization representations such as time-frequency spectrograms and topographic maps. To address this gap, this paper proposes a Hybrid Image–Sequence Model (HISM) that integrates convolutional neural network (CNN)-based spatial feature extraction from EEG visualization maps with bidirectional long short-term memory (BiLSTM)-based temporal sequence modeling to detect narcoleptic sleep-stage irregularities across multi-stage EEG recordings. Methods EEG signals are transformed into structured visual representations, including Short-Time Fourier Transform (STFT) spectrograms and Continuous Wavelet Transform (CWT) maps, enabling the model to simultaneously capture intra-epoch spatial patterns and inter-epoch temporal transitions characteristic of narcoleptic progression. The hybrid CNN–BiLSTM architecture extracts complementary spatial and sequential features, improving detection sensitivity for REM-onset anomalies and wake-intrusion events. The Sleep-EDF Expanded and SLEEPTIGHT polysomnographic datasets have been used for experimental evaluation. Results The proposed HISM framework achieved a SOREMP sensitivity of 0.887, a REM intrusion detection performance of 0.908, an AUC-ROC of 0.951, and an overall detection accuracy for narcoleptic irregularity of 92.9%, with a false alarm rate of 0.057. Discussion These results provide a comprehensive evaluation of sleep-stage irregularity detection and illustrate HISM’s potential for automated analysis of EEG-based sleep patterns in a patient with narcolepsy.
No takes yet. Share an insight, caveat, or question.
Alsanea et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: