Recognition of Patient Groups with Sleep Related Disorders using Bio-signal Processing and Deep Learning.

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Polysomnography (PSG) is the standard for sleep disorder diagnosis but can be complex.
• Electrocardiography (ECG) and electromyography (EMG) signals contain valuable information for sleep disorder detection.
• Developing automated diagnostic tools is crucial for efficient clinical assessment.

Purpose of the Study:

• To develop and evaluate a deep learning framework for classifying sleep disorders using ECG and EMG signals.
• To identify breathing and movement-related sleep disorders, including obstructive sleep apnea (OSA) and restless legs syndrome (RLS).
• To compare the performance of the deep learning model against traditional diagnostic methods.

Main Methods:

• Bio-signal processing of EMG features using entropy and statistical moments.
• Iterative pulse peak detection algorithm with synchrosqueezed wavelet transform (SSWT) for ECG analysis.
• A deep learning framework integrating processed EMG and ECG features for classification.

Main Results:

• The deep learning framework successfully classified four groups: healthy subjects, OSA patients, RLS patients, and patients with both OSA and RLS.
• The model achieved a mean accuracy of 72% across all subjects.
• A weighted F1 score of 0.57 was obtained for the four-class classification problem.

Conclusions:

• The proposed deep learning framework demonstrates potential for non-invasive sleep disorder diagnosis using ECG and EMG.
• This approach offers a promising alternative or adjunct to traditional polysomnography.
• Further research and validation are needed to improve accuracy and clinical applicability.