An Explainable AI Approach Towards Automatic Sleep Apnea Detection Based on ECG Signal

Sleep apnea is a serious abnormal sleep condition that deals with breathing pauses during a patient’s sleep, which may cause serious health concerns if not treated. This study gives an approach for diagnosing sleep apnea using explainable AI methods applied to single-lead ECG readings. The process includes extracting features from ECG readings and using models such as Multilayer Perceptron (MLP), Support Vector Machine (SVM), Gradient Boosting (GB), and Long Short-Term Memory (LSTM) networks to detect apnea events. We focus on improving the interpretability of our models by employing ante-hoc and post-hoc Explainable Artificial Intelligence (XAI) techniques such as Decision Trees and Local Interpretable Model-agnostic Explanations (LIME), which provide an improved understanding of the decision-making procedure. Experimental results demonstrate that our technique works effectively, with the suggested models recognizing apnea episodes using time windows with high accuracy on the Apnea ECG database. Incorporating LIME strengthens the prediction of the Machine Learning model, making it an impressive tool for medical decision support. This study illustrates the potential of explainability tools to design reliable, interpretable, and explainable machine learning models for detecting sleep apnea, with a chance of expanding these approaches to other medical applications. This approach provides insight into the critical features contributing to sleep apnea detection.