•  
  •  
 

Abstract

Obstructive Sleep Apnoea (OSA) is a prevalent sleep disorder characterised by repeated episodes of partial or complete upper airway obstruction during sleep, primarily due to the relaxation and collapse of soft tissues in the throat. These interruptions lead to disrupted sleep patterns and reduced oxygen saturation, increasing the risk of cardiovascular complications. Although Polysomnography (PSG) is considered the gold standard for diagnosing OSA, it is often uncomfortable for patients due to the extensive use of sensors and prolonged monitoring duration. As a result, there is a growing need for alternative diagnostic methods that are more efficient, comfortable, and cost-effective. This article presents a novel approach for detecting OSA using Electrocardiogram (ECG) signals, which are more accessible and less intrusive than full PSG. The proposed method integrates Kernel Principal Component Analysis (KPCA) for dimensionality reduction with a one-dimensional Convolutional Neural Network (CNN) model for classification. The process is tested using clinical data from the Second Sleep Heart Health Study (SHHS-II), which provides a large and reliable dataset for validation. The KPCA-CNN framework extracts critical features from the ECG signals and accurately classifies OSA events. Experimental results indicate that the model successfully identifies OSA patterns. The highest performance is achieved when the CNN is configured with a stride of 3, a kernel of 6, and a filter of 64, respectively. This configuration results in performance with an F1-score of 86.92%. These findings highlight the potential of the proposed method as a non-invasive, efficient screening tool for OSA. It may enhance early diagnosis and improve clinical outcomes.

Share

COinS