Advances in Machine Learning Algorithms in the Treatment of Obstructive Sleep Apnea

Abstract

Recent developments in machine learning (ML) and deep learning (DL) algorithms for the diagnosis, screening, and most importantly, treatment of obstructive sleep apnea (OSA) are thoroughly reviewed in this review article. The analysis highlights the empirical performance of different AI-powered models and identifies important research gaps by synthesizing findings from recent literature (2018–2025). With accuracy, sensitivity, and specificity frequently surpassing those of conventional tools, ML and DL models have shown notable advancements in OSA detection. For instance, using electrocardiogram (ECG) and oxygen saturation (SpO2) signals, a multimodal signal fusion multiscale Transformer model achieved 91.38% per-segment and 96.08% per-recording accuracy for OSA detection and severity assessment. The OSA event detection accuracy of a different one-dimensional convolutional neural network (1D-CNN) model for portable monitors was 84.3%. Artificial intelligence (AI) is being used more and more for treatment optimization in addition to diagnosis. This includes predicting and enhancing Continuous Positive Airway Pressure (CPAP) compliance, enabling remote monitoring and chronic disease management, and facilitating customized treatment regimens. The underrepresentation of diverse demographic groups in study cohorts, small sample sizes, and limited robust model validation are among the major research gaps that still exist despite these advancements. These factors work together to impede model generalizability and equitable application in healthcare. The future of AI in OSA management depends on addressing these constraints through thorough validation and inclusive data collection.