Detection and prediction of hypothermia in ICUs using thermal images

Abstract

Hypothermia is a critical condition in which the human body starts to lose heat faster than it can produce resulting in extremely low body temperatures. Specifically, when the core body temperature drops below 95 degrees, the patient is formally reported to be suffering from hypothermia. Symptoms of hypothermia include shivering, slow breathing, confusion, weak pulse, exhaustion or loss of coordination. Severe hypothermia can result in unconsciousness, multiple organ failure (due to permanent tissue damage), coma or even death. Hypothermia is a very common condition in intensive care units (ICU) and lack of timely care can result in deaths. Due to these reasons, it is extremely useful to detect and predict hypothermia in order to provide timely care, thus saving lives. Contactless monitoring is a promising and economical solution especially for infants and elderly in developing countries with low doctor-to-patient ratios and high risk of hospital-acquired infections. In this work, we aim to develop direct contactless classification methods to predict hypothermia in ICU set ups using thermal videos and deep learning techniques. In this work, hypothermia prediction is done 1 hour, and 2 hour in advance.