Classification and transfer learning of sleep spindles based on convolutional neural networks

Area of Science:

• Neuroscience
• Computational Biology
• Medical Informatics

Background:

• Sleep is crucial for health, and electroencephalography (EEG) is vital for sleep disorder diagnosis.
• Sleep spindles, a key EEG phenomenon, are important in sleep science and diagnostics.
• Accurate identification of sleep spindles aids in understanding sleep patterns and diagnosing disorders.

Purpose of the Study:

• To propose a novel convolutional neural network (CNN) model for classifying sleep spindles.
• To investigate the effectiveness of transfer learning for applying models trained on healthy subjects to insomnia patients.
• To analyze the impact of transferring different numbers of convolutional layers on classification performance.

Main Methods:

• Developed a novel CNN model for sleep spindle classification.
• Employed transfer learning to adapt a model trained on healthy subjects for classifying spindles from insomnia patients.
• Evaluated classification performance with partial and full transfer of convolutional layers.

Main Results:

• The CNN model achieved high classification accuracy for both healthy (93.68%) and insomnia (92.77%) subjects' sleep spindles.
• Transfer learning, particularly transferring the first four convolutional layers, yielded strong results for insomnia subjects (92.80% accuracy).
• Demonstrated that features learned from healthy subjects' data are effectively transferable to insomnia subjects' data.

Conclusions:

• The proposed CNN model is effective for classifying sleep spindles.
• Transfer learning is a viable strategy for improving sleep spindle classification in insomnia patients.
• The model shows potential for rapid and effective diagnosis and treatment of sleep disorders.