Explainable AI-driven scalogram analysis and optimized transfer learning for sleep apnea detection with single-lead...

Area of Science:

• Cardiology
• Medical Technology
• Artificial Intelligence

Background:

• Sleep apnea is a serious sleep disorder characterized by repetitive respiratory cessation, necessitating prompt intervention due to associated neuropsychological issues.
• Current diagnostic methods like polysomnography, while accurate, are often limited by the need for complex multichannel electrocardiogram (ECG) recordings and advanced feature extraction.
• Deep learning presents a promising avenue for developing more accessible sleep apnea detection tools.

Purpose of the Study:

• To develop and validate a novel deep learning-based method for detecting sleep apnea using single-lead ECG signals.
• To evaluate the performance of the proposed method across multiple public datasets.
• To enhance the interpretability of the deep learning model using explainable AI techniques.

Main Methods:

• Utilized continuous wavelet transform to convert single-lead ECG signals into time-frequency representations (scalograms).
• Employed an optimized pre-trained GoogLeNet architecture for transfer learning to classify sleep apnea.
• Validated the model on the PhysioNet Apnea ECG, UCDDB, and MIT-BIH polysomnographic datasets for per-segment classification.

Main Results:

• Achieved high performance on the Apnea ECG dataset with 93.85% accuracy, 93.42% sensitivity, 94.30% specificity, and 93.83% F1-score.
• Demonstrated robust performance on the UCDDB dataset (87.20% accuracy) and MIT-BIH dataset (88.58% accuracy).
• LIME (Local Interpretable Model-agnostic Explanations) was used to provide insights into the model's predictions.

Conclusions:

• The proposed deep learning method effectively detects sleep apnea using single-lead ECG signals, offering a potentially more accessible diagnostic approach.
• The model's consistent performance across diverse datasets highlights its robustness and potential for wider clinical application.
• Explainable AI further supports the clinical utility by providing transparency in the model's decision-making process.