INFORMS Open Forum

Abstract: Sleep apnea is a prevalent but under-diagnosed disorder. Automating its detection faces three major hurdles: the prohibitive cost of manual fine-grained annotation, the uncertainty inherent in model predictions, and the pervasive corruption of physiological signals by artifacts. This presentation introduces an integrated machine learning pipeline that systematically addresses these challenges to enable accurate, real-time, and scalable apnea severity monitoring.

First, we present a weakly-supervised deep learning model that estimates fine-grained, per-instant apnea severity using only coarse-grained, session-level labels during training. The core innovation is a knowledge-enhanced dual-granularity consistency loss, which integrates clinical diagnostic rules to guide the model in learning accurate fine-grained predictions from weak supervision, dramatically reducing labeling costs. Second, to optimally refine the model with minimal expert effort, we introduce a dual-granularity Bayesian active learning framework. This framework quantifies both aleatoric and epistemic uncertainty and uses a novel acquisition function that prioritizes instances where model predictions are both highly uncertain and misaligned with clinical knowledge. It strategically queries the most informative fine-grained labels for human annotation, maximizing diagnostic accuracy while minimizing labeling burden. Finally, we present a novel unsupervised artifact removal framework, a dual-branch convolutional autoencoder that decomposes raw, multichannel physiological signals into clean and artifact components without needing labeled noise data. By learning to impute corrupted regions, it provides a robust, artifact-resilient signal foundation for downstream diagnosis.

Biographical Sketch: Dr. Xiaochen Xian is currently an assistant professor in H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology. Prior to joining Georgia Tech, she was an assistant professor in the Department of Industrial and Systems Engineering at the University of Florida. Her research interest mainly focuses on big data analytics and system informatics. Specifically, her research includes big data stream monitoring and sampling, engineering knowledge-enhanced complex process modeling and diagnosis, on-demand machine learning, and system informatics and spatiotemporal real-time prediction. Dr. Xian's research has been supported by federal and local agencies including NSF, NIH, the Florida Center for Cybersecurity, and the Florida Space Grant Consortium. She is the recipient of multiple awards, including NIH NIBIB Trailblazer Award, Cottmeyer Family Faculty Fellowships, multiple paper awards from INFORMS, IISE, and IEEE, and feature articles in IISE magazine, AIE, and YoungStats. Dr. Xian is an associate editor of IEEE Transactions on Automation Science and Engineering.