Deep Learning Model and Multi Modal Late Fusion For Predicting Adverse Events Following Cardiothoracic Surgery in the ICU Using STS Data and Time Series Intraoperative Data

Abstract

Abstract Accurate prediction of post-operative adverse events following cardiothoracic surgery is crucial for timely interventions, potentially improving patient outcomes and reducing healthcare costs. By leveraging advanced deep learning techniques, this study highlights the transformative potential of incorporating intraoperative variables into predictive analytics models to enhance postoperative care for cardiothoracic surgery patients in the ICU. We developed deep learning predictive models for anticipating adverse events in patients following cardiothoracic surgery using a dataset from the Society of Thoracic Surgeons’ database ( 4 ) and intraoperative data. Our models perform late fusion by integrating static patient data and intra-operative time-series data, utilizing Fully Connected Neural Networks (FCNN) and long short-term memory (LSTM) networks, respectively. The hybrid model was validated through five-fold cross-validation, demonstrating robust performance with a mean AUC of 0.93, Sensitivity of 0.83 and Specificity of 0.89. This work represents a significant step forward in the proactive management of cardio thoracic surgery patients in the ICU by effectively predicting potential adverse events associated with mortality in the post operative period.

Ähnliche Arbeiten