Defining Intraoperative Turbulence Using a Deep Learning Approach

Project Description Narrative:

Real-time intraoperative data provides a unique resource for understanding the physiological reaction to surgical stress. During every surgery, a substantial amount of data is collected from noninvasive and invasive monitoring systems embedded within the operating room. This data is then fed into an anesthesia information system which can be integrated with electronic health records. In Wisconsin, there are approximately one-million surgical procedures performed annually; gaining additional insight into the moment-by-moment conduct of an operation creates an opportunity to transform perioperative care statewide and beyond. A significant gap in the current use of intraoperative data is related to limitations in how the data are modeled and applied.

This project aims to overcome such limitations with a new paradigmatic approach to analyzing and understanding intraoperative data. Researchers will create a data system to facilitate the development of a modeling approach to characterize intraoperative turbulence. This will be accomplished using a multivariable deep learning approach to detect anomalies in high-complexity time-series data. The frequency, magnitude, and duration of intraoperative anomalies will be parameterized to develop a composite measure for intraoperative turbulence.