Real-Time Obstructive Sleep Apnea Detection from Raw ECG and SpO₂ Signal Using Convolutional Neural Network

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) is a serious sleep disorder with significant health risks.
• Current diagnosis via polysomnography is resource-intensive and requires expert analysis.
• Real-time AI detection of OSA is desirable but often hindered by complex preprocessing.

Purpose of the Study:

• To develop an efficient, single convolutional neural network (CNN) architecture for real-time OSA detection.
• To automatically extract spatial features from physiological signals for apnea identification.
• To evaluate the performance of the CNN using electrocardiogram (ECG) and blood-oxygen saturation (SpO2) data.

Main Methods:

• A single CNN architecture was designed for simultaneous feature extraction and apnea detection.
• The model processed 10-second segments of ECG and SpO2 signals.
• Apnea detection was performed directly from the raw signals, minimizing preprocessing steps.

Main Results:

• The CNN achieved high classification accuracy: 94.2% for ECG and 96% for SpO2.
• Both models demonstrated excellent performance with an Area Under the Curve (AUC) score of 0.99.
• The proposed method enables efficient real-time apnea detection.

Conclusions:

• A single CNN model can effectively detect obstructive sleep apnea from ECG and SpO2 signals.
• This approach offers an efficient and accurate alternative to traditional diagnostic methods.
• The AI model shows promise for real-time, accessible sleep apnea screening.