Identification of Sleep Apnea using Deep Learning and Machine Learning Techniques on ECG Data

Patients with sleep apnea (SA) frequently experience episodes where the airflow to their lungs stops or decreases for more than 10 seconds. Accurately recognising SA episodes is necessary to choose the best treatments and drugs. A variety of techniques, including Polysomnography (PSG), ECG (Electro - Cardiogram) signals, and others, can be used to identify SA. An ECG signal-based detection approach based on Machine Learning (ML) or Deep Learning (DL) would be more effective in terms of time and effort. This study employs two different approaches to forecast the user's SA risk. The first technique uses ML to forecast the likelihood that a person would develop SA based on a variety of parameters, including ESS, BQ, BMI, and others. The second technique uses ECG image data as its foundation. This study makes use of the PhysioNettECG SA V1.0.0 dataset, which has 70 recordings from 32 different subjects. The dataset has been pre-processed and divided into four classes: X (non apnoeic), B (modelate apnea), C (normal apnea), and A (severe apnea). With an F1-Score of 90.83%, the VGG16-LSTM model proved to be the most effective model for classification.