End of Apnea Event Prediction Leveraging EEG Signals and Interpretable Machine Learning

TL;DR

This study uses EEG signals and machine learning to predict when apnea events end, aiming to improve sleep apnea treatments.

Contribution

The study introduces interpretable machine learning to identify EEG markers that predict the termination of apnea events.

Findings

The Extra Trees model achieved high accuracy in predicting apnea event termination.

Frequency-band energy, Teager–Kaiser energy, and signal complexity were key EEG features for prediction.

Temporal analyses showed how these features change during apnea termination.

Abstract

Obstructive sleep apnea is a prevalent sleep disorder with serious health implications. While previous studies focused on detecting apnea events, little is known about the factors that determine whether an apnea episode continues or terminates. Understanding these mechanisms is crucial for optimizing treatment strategies. In this study, we analyzed 30-s brain activity segments during continuous and ending apnea events to identify neurophysiological markers of event termination, with particular emphasis on the most influential EEG features. Frequency-domain and complexity features were extracted, and several ensemble machine learning models were trained and evaluated. Our results show that the Extra Trees model achieved the highest performance, with an accuracy of 0.88, F1-score for ending apnea of 0.87, and an area under the receiver operating characteristic curve of 0.95. Feature importance analyses and SHAP visualizations highlighted frequency-band energy, Teager–Kaiser energy, and signal complexity as key contributors. Temporal analyses revealed how these features evolve during apnea termination. These findings suggest that cortical activation and transient arousal processes play a decisive role in ending apnea events and may facilitate the development of more advanced adaptive or closed-loop sleep apnea therapies.

Figures