ExFormer: Exploring Convolution and Transformer Based Model for Sleep Apnea Recognition.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Obstructive sleep apnea (OSA) poses significant health risks, including cardiovascular disease.
• Early detection of OSA via electrocardiogram (ECG) signals is crucial for effective management.
• Current diagnostic methods can be invasive or time-consuming.

Purpose of the Study:

• To introduce ExFormer, a novel CNN-Transformer model for detecting OSA using single-lead ECG data.
• To evaluate the performance of ExFormer in accurately classifying OSA.
• To compare ExFormer's efficacy against existing OSA detection methods.

Main Methods:

• Developed ExFormer, a hybrid CNN-Transformer model integrating a multiperspective convolutional network (MCN) and a Transformer module.
• Utilized MCN for efficient feature extraction from short ECG sequences.
• Employed the Transformer module for enhanced data-parallel processing and capturing signal dependencies.

Main Results:

• ExFormer achieved high performance metrics: 85.47% accuracy, 79.28% sensitivity, 89.15% specificity, and 92.87% AUC.
• Demonstrated a 0.37% accuracy improvement over the best existing alternative model.
• ExFormer showed robust precision and reliability in classifying OSA events.

Conclusions:

• ExFormer offers a superior, accurate, and non-invasive tool for clinical OSA detection using ECG.
• The model's ability to capture local and long-range dependencies enhances OSA diagnosis.
• ExFormer facilitates prompt treatment, potentially improving patient outcomes and condition management.