Obstructive sleep apnea detection using discrete wavelet transform-based statistical features.

Area of Science:

• Cardiology
• Biomedical Engineering
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) affects nearly 10% of middle-aged adults, significantly impacting cardiovascular health and increasing mortality risk.
• Untreated OSA is linked to hypertension, diabetes, stroke, and other cardiovascular diseases, highlighting the need for early diagnosis.

Purpose of the Study:

• To develop and validate an automated diagnostic approach for obstructive sleep apnea (OSA) utilizing single-lead electrocardiogram (ECG) data.
• To identify significant ECG features for OSA detection and evaluate the performance of machine learning classifiers.

Main Methods:

• Extraction of three feature sets (PSD moments, waveform complexity, higher-order moments) from ECG subbands using Discrete Wavelet Transform (DWT).
• Application of correlation-based feature selection with Particle Swarm Optimization (PSO) to select 18 optimal features from an initial 32.
• Classification of ECG segments using Linear Discriminant Analysis, Nearest Neighbors, Support Vector Machine, and Random Forest classifiers.

Main Results:

• The Random Forest classifier demonstrated effective discrimination between normal and OSA ECG signals, achieving 89% accuracy with 50-50 hold-out and 90% with 10-fold cross-validation.
• The proposed method achieved an Area Under the ROC Curve (AUC) of 96% in both validation scenarios.
• Performance results surpassed those of most existing methodologies for OSA detection from ECG.

Conclusions:

• The automated ECG-based approach offers a reliable and accurate method for diagnosing obstructive sleep apnea.
• This technique holds potential for non-invasive, early detection of OSA, facilitating timely intervention and management.
• The feature selection and classification strategy effectively identifies OSA, outperforming existing methods.