An Informatics System for Breath-by-Breath Analysis of Large-Scale Multi-Modal Time-Series Data in Sleep Research
Sleep medicine involves handling large volumes of multi-modal time-series data that capture diverse biological signals. Analyzing these signals on a breath-by-breath basis is crucial for understanding intricate respiratory patterns, which yield valuable scientific insights and inform clinical decision-making. However, manual analysis of such data is labor-intensive and prone to error, and there's a shortage of easy-to-use analytical tool for processing data at scale. To address these challenges, we have developed a comprehensive informatics system that automates breathing cycle segmentation, feature engineering, recognition of specific respiratory patterns, and visualization of findings, using standard physiological signals from sleep studies. Our pipeline includes a deep learning model for identifying flow limitation - the definitive indicator of airway collapse with significant analytic and clinical implications. We evaluated this system using real-world patient data from 41 individuals undergoing drug-induced sleep endoscopy (DISE) procedures. The system has been deployed as a web-based platform with a graphical user interface (GUI). This intuitive application is anticipated to enhance the efficiency of breath-level sleep data analysis and expand its accessibility to a broader scientific community.
