Enhancing OSA Assessment with Explainable AI

Explainable Artificial Intelligence (xAI) is a rapidly growing field that focuses on making deep learning models interpretable and understandable to human decisionmakers.
In this study, we introduce xAAEnet, a novel xAI model applied to the assessment of Obstructive Sleep Apnea (OSA) severity. OSA is a prevalent sleep disorder that can lead to numerous medical conditions and is currently assessed using the Apnea-Hypopnea Index (AHI). However, AHI has been criticized for its inability to accurately estimate the effect of OSAs on related medical conditions. To address this issue, we propose a human-centric xAI approach that emphasizes similarity between apneic events as a whole and reduces subjectivity in diagnosis by examining how the model makes its decisions. Our model was trained and tested on a dataset of 60 patients’ Polysomnographic (PSG) recordings. Our results demonstrate that the proposed model, xAAEnet, outperforms models with traditional architectures such as convolutional regressor, autoencoder (AE), and variational autoencoder (VAE) . This study highlights the potential of xAI in providing an objective OSA severity scoring method.