Project Details
Monitoring of Sedation for Endoscopy with Artificial Intelligence (SAFE-AI)
Applicants
Dr. Jakob Garbe; Professor Dr. Thomas Schmid
Subject Area
Gastroenterology
Anaesthesiology
Anaesthesiology
Term
since 2024
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 547230187
Endoscopic interventions are an important and indispensable component of modern medicine. Since endoscopic procedures subject patients to physical and psychological stress, they are performed in sedation. Furthermore, steady conditions for examination, which are essential for complex interventions, can only be achieved in sedation. Usually, a trained nurse who is guided by the subjective clinical impression of the patient state and few vital parameters such as pulse oximetry and blood pressure administers the sedation agents. The clinical impression includes grimacing, paralinguistic, sweating and others. Experience of the nurse is known to be strong predictor of sedation quality. Hitherto, an objective device to monitor sedation is not available. While the overall rate of severe adverse events related to sedation is low, patients with known risk factors have a significantly higher risk to experience severe adverse events. For these high risk intervention a better device to monitor and guide the sedation is desirable. With a trend towards minimally invasive endoscopic procedures reliant on steady and continuous conditions, the need for optimally guided sedation is supported even further. In three interconnected modules, this project aims to develop a prototypic prediction model of the state of consciousness and sedation depth. Both, classic machine learning and deep learning, as methods of artificial intelligence will be utilized in a competing fashion to create this prediction model. The data for the modelling effort will be generated in the first module. Here, real-time bio-signal data of endoscopic procedures from gastroenterology and pneumology will be recorded prospectively in multiple centers. These signals include EEG as well as classic vital parameters such as ECG and pulse oximetry and others. As a reference, state of consciousness and clinical sedation depth will be continuously documented in the stream of data. In the second module the EEG and other vital parameters will be processed to extract signal information for further use in classic machine learning models. These processed parameters are adapted from general anesthesia research and will be refined. Furthermore, new concepts for signal processing and information extraction will be explored. As a result of the third module and this project overall a valid, robust and reliant AI model for the prediction of state of consciousness and sedation depth will be available to be integrated into an reliable and clinically useful objective monitor for sedation during endoscopic procedures.
DFG Programme
Research Grants