SleepFC: Feature Pyramid and Cross-Scale Context Learning for Sleep Staging.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Automated sleep staging using electroencephalography (EEG) is crucial for diagnosing sleep disorders and assessing sleep quality.
• Current methods face challenges in extracting salient wave features, capturing sleep stage transition rules, and addressing class imbalance in EEG data.

Purpose of the Study:

• To propose a novel method, SleepFC, for improved automated sleep staging from single-channel EEG signals.
• To address the limitations of existing methods in feature learning, transition rule extraction, and class imbalance.

Main Methods:

• SleepFC utilizes a convolutional feature pyramid network (CFPN) for multi-scale feature extraction from salient EEG waves.
• Cross-scale temporal context learning (CSTCL) is employed to capture informative sleep stage transition rules.
• A class adaptive fine-tuning loss function (CAFTLF) based classification network addresses the class imbalance problem.

Main Results:

• SleepFC demonstrated superior performance compared to state-of-the-art approaches across three public benchmark datasets.
• The method showed a significant advantage in accurately recognizing the challenging N1 sleep stage.

Conclusions:

• SleepFC offers an effective solution for automated sleep staging by integrating multi-scale feature learning, temporal context, and class imbalance handling.
• The proposed method advances the accuracy and reliability of EEG-based sleep analysis, particularly for subtle sleep stages.