Ensemble of Deep Learning Models for Sleep Apnea Detection: An Experimental Study

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Sleep Apnea is a prevalent sleep disorder, particularly affecting older adults.
• Timely diagnosis of Sleep Apnea is crucial and can be facilitated by advanced health monitoring systems.
• Obstructive Sleep Apnea (OSA) detection from Electrocardiogram (ECG) signals presents a non-invasive diagnostic avenue.

Purpose of the Study:

• To investigate the efficacy of various ensemble techniques for Obstructive Sleep Apnea (OSA) detection.
• To evaluate the performance of deep learning models, including CNN and CNN-LSTM architectures, when combined with ensemble methods.
• To compare the proposed ensemble approaches against existing state-of-the-art methods for OSA detection using ECG data.

Main Methods:

• Utilized the PhysioNet Apnea-ECG Database for experimental analysis.
• Applied four ensemble techniques: majority voting, sum rule, Choquet integral-based fuzzy fusion, and trainable Multi-Layer Perceptron (MLP) ensemble.
• Integrated ensemble methods with three deep learning models: two Convolutional Neural Network (CNN) variants and one CNN-LSTM hybrid model.

Main Results:

• Achieved a maximum OSA detection accuracy of 85.58% using the MLP-based ensemble approach.
• Demonstrated that ensemble techniques significantly improve the performance of deep learning models for OSA detection.
• The best-performing model surpassed the accuracy of several established state-of-the-art methods in the field.

Conclusions:

• Ensemble learning, particularly with MLP, offers a powerful strategy for enhancing Obstructive Sleep Apnea detection from ECG signals.
• Deep learning models combined with ensemble methods show significant promise for accurate and non-invasive sleep apnea diagnosis.
• The developed approach provides a competitive and effective solution for improving sleep apnea monitoring systems.