AFSleepNet: Attention-Based Multi-View Feature Fusion Framework for Pediatric Sleep Staging

Area of Science:

• Pediatric Sleep Medicine
• Artificial Intelligence in Healthcare
• Biomedical Signal Processing

Background:

• Pediatric sleep disorders are common, necessitating accurate sleep staging for diagnosis and treatment.
• Current sleep staging methods using single-view data (1D or 2D) miss crucial details, limiting precision medicine for children.
• A specialized network is needed to address the unique challenges of pediatric sleep analysis.

Purpose of the Study:

• To introduce AFSleepNet, a novel attention-based multi-view feature fusion network for pediatric sleep analysis.
• To enhance the accuracy and robustness of automatic sleep staging in children.
• To improve the reliability of diagnosing pediatric sleep disorders.

Main Methods:

• Utilized multimodal data including electroencephalogram (EEG), electrooculogram (EOG), and electromyogram (EMG).
• Employed a hybrid deep learning architecture combining 1D convolutional neural networks (CNNs) and bidirectional-long-short-term memory (BiLSTM).
• Integrated short-time Fourier transform (STFT) for spectral map generation and a self-attention mechanism for feature fusion, alongside a pre-training strategy.

Main Results:

• AFSleepNet achieved high performance on CHAT and clinical datasets, with mean accuracies of 87.5% and 88.1%, respectively.
• The multi-view fusion approach enhanced model robustness and prevented overfitting.
• Demonstrated superior accuracy and reliability compared to existing pediatric sleep staging methods.

Conclusions:

• AFSleepNet offers an efficient and accurate solution for automatic pediatric sleep stage analysis.
• The attention-based multi-view feature fusion network effectively addresses limitations of single-view methods.
• This advancement holds significant potential for improving the diagnosis and treatment of pediatric sleep disorders.