Feature Extraction with Multiple Receptive Fields Conducted on ECG Signals for Performance Enhancement in OSA Severity Classification

Obstructive sleep apnea (OSA) is a common sleep disorder, and polysomnography (PSG) is the gold standard for diagnosing OSA. However, patients often have to wait long before receiving the costly PSG test in a hospital. Hence, low-cost and easy-to-use portable screening tools were developed for predicting OSA. Based on our recent study, this paper presents a feature extraction method with multiple receptive fields applied to the input electrocardiography (ECG) signals of a model to improve its performance in OSA severity classification. This work also employs unsegmented ECG signals as input to keep all the advantages from our original approach. The proposed model achieves an overall accuracy of 59.49% for four-level OSA severity classification, giving an improvement of approximately 2% compared to our original work. The effectiveness of the proposed method is demonstrated.