Early Prediction of Hemodynamic Shock in the Intensive Care Units with Deep Learning on Thermal Videos: A Retrospective Longitudinal Study (Preprint)

Abstract

Shock is one of the major killers in Intensive Care Units and early interventions can potentially reverse it. In this study, we advance a non-contact thermal imaging modality for continuous monitoring and prediction of hemodynamic shock in advance.

We aim to monitor and predict the advent of hemodynamic shock 6 hours in advance using an automated non-contact thermal imaging decision pipeline.

Thermal Videos were captured in a Pediatric ICU-setting along with vitals time-series data. Deep-learning-based body-part segmentation models were trained to extract the Center-to-Peripheral temperature value difference from the videos. Extracted time-series data along with heart rate was finally analyzed using Long-Short Term Memory models to predict the shock status up to the next 6 hours.

103,936 frames from 406 non-contact thermal videos were recorded longitudinally upon 22 patients. Our models were able to predict the shock well till 6 hours of lead time using thermal information and achieved the best area under the receiver operating characteristics curve of 0.81±0.06 and area under the precision-recall curve of 0.78±0.05 at 5 hours, providing sufficient time to stabilize the patient.

Our approach leverages thermal imaging as a non-invasive and non-contact modality to continuously monitor hemodynamic shock, and thus, provides a reliable shock prediction using an automated decision pipeline that can provide better care and save lives. 

None