Machine learning-based automatic sleep apnoea and severity level classification using ECG and SpO₂ signals

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Sleep Medicine

Background:

• Sleep apnoea is a serious sleep disorder characterized by breathing interruptions.
• Traditional diagnosis via polysomnography (PSG) is complex, time-consuming, and relies on expert interpretation.
• There is a need for objective, efficient, and accurate methods for sleep apnoea diagnosis.

Purpose of the Study:

• To develop an automatic sleep apnoea and severity classification system.
• Utilize machine learning algorithms with electrocardiograph (ECG) and oxygen saturation (SpO2) signals.
• Improve diagnostic efficacy, reduce complexity, and enhance accuracy.

Main Methods:

• Extracted time and frequency domain features from ECG and SpO2 signals.
• Trained machine learning algorithms, including Support Vector Machine (SVM).
• Evaluated classification performance for both apnoea presence and severity.

Main Results:

• Achieved 99.1% accuracy, 98.1% specificity, and 100% sensitivity for sleep apnoea classification using SVM with combined ECG and SpO2 features.
• Obtained 88.9% accuracy, 90.9% specificity, and 85.7% sensitivity for severity classification.
• Combined features demonstrated superior accuracy, offering robustness even with single-channel signal degradation.

Conclusions:

• An automatic classification system using ECG and SpO2 signals is effective for diagnosing sleep apnoea and its severity.
• Machine learning, particularly SVM with combined features, provides a highly accurate and efficient diagnostic approach.
• This method offers a promising alternative to traditional PSG, enhancing diagnostic accessibility and reliability.