A Deep-Learning-Based Approach for Delirium Monitoring in ICU Patients Using Thermograms

Patients in the ICU frequently suffer from delirium, which can delay their recovery and may cause significant distress. Despite standardized scoring systems, its diagnosis and classification however, remain largely subjective and are subject to intra-observer variability. Using infrared thermography images, so-called thermograms, for delirium analysis increases objectiveness and also allows for unobtrusive and continuous monitoring. We analyzed the conveyable information from movement and temperature information and designed a pipeline of deep neural networks which determine a patient's agitation with an accuracy of 66.76 %.