Real-time sleep apnea detection by classifier combination

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Sleep Medicine

Background:

• Polysomnography (PSG) is the gold standard for diagnosing sleep apnea and hypopnea syndrome (SAHS) but is costly and inconvenient.
• There is a need for efficient and valid alternatives for real-time SAHS detection.
• Electrocardiograph (ECG) and saturation of peripheral oxygen (SpO2) signals are non-invasive and readily available physiological measures.

Purpose of the Study:

• To investigate the efficacy of real-time SAHS detection using ECG and SpO2 signals, individually and in combination.
• To compare the diagnostic ability of ECG and SpO2 features for SAHS detection.
• To enhance SAHS detection performance through classifier combination.

Main Methods:

• Ten machine learning algorithms were employed for classification experiments.
• Features derived from ECG and SpO2 signals were extracted and analyzed.
• Classifier combination techniques, including AdaBoost, Bagging with REPTree, kNN, and Decision Table, were utilized.
• The study analyzed full overnight recordings from 25 sleep-disordered-breathing suspects for minute-based detection.

Main Results:

• SpO2 features demonstrated superior diagnostic ability compared to ECG features for SAHS detection.
• Classifier combination significantly enhanced classification performance by leveraging complementary information.
• The best classifier combinations achieved sensitivity, specificity, and accuracy around 82% for real-time SAHS detection.

Conclusions:

• Real-time SAHS detection using combined ECG and SpO2 signals with machine learning offers a promising alternative to PSG.
• SpO2 signal analysis is particularly effective for SAHS detection.
• Classifier combination strategies can substantially improve the accuracy and reliability of automated sleep apnea diagnosis.