ECGAN-Assisted ResT-Net Based on Fuzziness for OSA Detection

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Electrocardiogram (ECG) analysis is gaining traction for obstructive sleep apnea (OSA) detection.
• Challenges in ECG-based OSA detection include limited data, poor data quality, and incomplete labeling, hindering deep learning model generalization.

Purpose of the Study:

• To propose the ResT-ECGAN framework to enhance deep learning-based OSA detection using ECG data.
• To address data scarcity and quality issues in ECG datasets for OSA detection.

Main Methods:

• Developed a one-dimensional generative adversarial network (ECGAN) for generating synthetic ECG samples with improved data quality.
• Integrated ECGAN with ResT-Net, a deep learning model utilizing multi-head attention for efficient feature extraction.
• Employed fuzziness in ECGAN to filter generated signals, increasing high-quality data for training.

Main Results:

• The ResT-Net model achieved an accuracy of 0.885 on the Apnea-ECG database and 0.837 on a private database.
• Incorporating ECGAN-generated data augmentation improved ResT-Net accuracy to 0.893 and 0.848 on the respective databases.
• The proposed ResT-ECGAN framework demonstrated superior performance compared to state-of-the-art deep learning methods.

Conclusions:

• The ResT-ECGAN framework effectively enhances OSA detection performance, particularly in scenarios with limited labeled data.
• This study offers a novel approach to improve the accuracy and generalization of deep learning models for ECG-based OSA detection.