Enhancing Neonatal Sleep Analysis with Multi-resolution CNN and Mamba Integration.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Accurate neonatal sleep stage classification is crucial for diagnosing sleep disorders.
• Manual annotation of EEG signals is laborious and time-consuming.
• Current deep learning methods face challenges in capturing bidirectional inter-stage dependencies.

Purpose of the Study:

• To develop an efficient deep learning framework for neonatal sleep staging.
• To improve the accuracy of sleep classification by addressing limitations in temporal modeling.
• To enhance the clinical applicability of automated sleep analysis in neonates.

Main Methods:

• Integration of a Multi-Resolution Convolutional Neural Network (MRCNN) for hierarchical feature extraction.
• Utilizing a Bidirectional Mamba Block for advanced temporal modeling of EEG signals.
• Evaluation on a clinical neonatal sleep dataset using 10-fold cross-validation for multiple sleep tasks.

Main Results:

• The proposed framework demonstrated superior performance compared to state-of-the-art methods.
• Achieved high accuracy (0.877) and macro F1-score (0.868) in Quiet Sleep (QS) detection.
• Showcased robustness and clinical applicability in sleep-wake classification and three-class sleep staging.

Conclusions:

• The developed framework offers an efficient and accurate solution for neonatal sleep staging.
• The method effectively captures both intra-stage features and inter-stage dependencies.
• Provides valuable tools for clinicians in monitoring neurodevelopment and guiding early interventions.