Related Experiment Video
Updated: Aug 14, 2026
04:54
IntelliSleepScorer, a Software Package with a Graphic User Interface for Mice Automated Sleep Stage Scoring
Published on: November 8, 2024
MAF-SleepNet: A Multimodal Attention-Enhanced Fusion Network for Automatic Multi-Class Sleep Disorder Classification
Suleyman Yaman1,2, Hasan Guler3, Abdul Hafeez-Baig4
1Department of Biomedical, Vocational School of Technical Sciences, Firat University, 23119 Elazig, Türkiye.
Diagnostics (Basel, Switzerland)
|August 13, 2026
Summary
A new AI model, MAF-SleepNet, accurately classifies multiple sleep disorders using multimodal data. This deep learning approach offers a more reliable clinical assessment than previous methods by considering the episodic nature of sleep pathophysiology.
Area of Science:
- Artificial Intelligence in Medicine
- Computational Neuroscience
- Sleep Medicine
Background:
- Sleep disorders are complex, requiring polysomnography (PSG) for diagnosis.
- Deep learning (DL) shows promise for integrating heterogeneous sleep data.
- Existing DL models often lack clinical applicability due to single-disorder focus or simplified labeling.
Purpose of the Study:
- To develop a novel multimodal attention-enhanced fusion network (MAF-SleepNet) for automatic multi-class sleep disorder classification.
- To address limitations of existing DL approaches by incorporating multimodal data and advanced fusion techniques.
- To classify sleep disorders based on the International Classification of Sleep Disorders.
Main Methods:
- Developed MAF-SleepNet, a deep learning model integrating electroencephalography (EEG), electrooculography (EOG), and leg electromyography (EMG) signals.
- Employed modality-specific feature extraction and adaptive attention mechanisms for intra- and inter-modality dependency capture.
- Evaluated the model on a combined dataset of 141 recordings from three public databases, including five disorders and a healthy class.
Main Results:
- Achieved 86.07% accuracy and 82.67% macro-F1 under subject-independent cross-validation.
- Reached 98.96% accuracy and 98.93% macro-F1 under subject-dependent cross-validation.
- Demonstrated superior performance compared to studies using subject-dependent evaluation or epoch-level labeling.
Conclusions:
- MAF-SleepNet offers a reliable and clinically meaningful approach for sleep disorder classification.
- Adaptive multimodal fusion is effective for robust and clinically relevant sleep disorder assessment.
- Future research should explore additional modalities and multi-center validation.
