Advancing Post-Operative Patient Monitoring with AI - Clinical Assist and alert Algorithm
The CASSANDRA project is developing advanced multimodal machine learning (ML) systems to enhance post-operative patient monitoring through continuous risk stratification and real-time prediction of post-surgical complications. Its aim is to develop clinical decision support systems (CDSS) that enable the early detection of complications, a critical factor influencing perioperative outcomes.
While intensive care units provide close surveillance, patients often experience reduced monitoring after transfer to standard care wards. CASSANDRA addresses this gap by automating continuous assessments of patient health data, facilitating earlier identification of clinical deterioration and supporting improved surgical decision-making and post-operative care.
At its core, CASSANDRA processes continuous time-series data from telemedical vital sign monitoring, enabling real-time analysis of dynamic physiological parameters. Additionally, the system integrates vectorized clinical notes using Natural Language Processing (NLP) and structured clinical information to create a comprehensive multimodal patient representation. By dynamically combining these heterogeneous data streams, CASSANDRA delivers continuous risk assessments and supports precise complication risk stratification throughout the hospital stay.
To further improve patient outcomes, the CASSANDRA working group focuses on enhancing preoperative risk stratification. For this purpose, federated learning and transfer learning techniques are applied to enable secure, decentralized model training across institutions and to adapt pre-trained models to specific clinical environments.
Info
- Project lead: Dr. Axel Winter, Dr. Max Maurer
- Team: Robin van de Water, Bjarne Pfitzner, Daniela Zuluaga, Christoph Riepe, Wolf-Heinrich Hahn, Prof. Bert Arnrich, Prof. Igor Sauer
- Duration: 2021 - 2026
- Funding: Innovationsfonds des Gemeinsamen Bundesausschusses
Publications
- Winter A, van de Water RP, Pfitzner B, Ibach M, Riepe C, Ahlborn R, Faraj L, Krenzien F, Dobrindt EM, Raakow J, Sauer IM, Arnrich B, Beyer K, Denecke C, Pratschke J, Maurer MM. Enhancing Preoperative Outcome Prediction: A Comparative Retrospective Case-Control Study on Machine Learning versus the International Esodata Study Group Risk Model for Predicting 90-Day Mortality in Oncologic Esophagectomy. Cancers 2024. 10.3390/cancers16173000