Keywords: core body temperature, heat injury
Heat injuries, which present a persistent threat to U.S. Armed Forces, could potentially be prevented by detecting a rise in core body temperature (Tc)—an early indicator of an impending heat injury. However, field assessment of Tc requires (impractical) invasive technologies. To address this problem, we developed a real-time, individualized early warning artificial intelligence (AI) system, which uses an individual’s physiological measurements (3-axis accelerometer, heart rate, and skin temperature) and environmental measurements (air temperature and humidity) to continually learn the individual’s heat-stress response and provide tailored Tc estimates in real time. We simulated real-time operation by using the system to learn the physiological response of 166 subjects from three distinct studies, with each subject exposed to multiple exertional and environmental conditions. When we compared the AI-estimated and the invasively measured TC, we observed an average error of 0.33?C [standard deviation (SD) = 0.18?C] across the 166 subjects over 530 hours. Furthermore, for the 22 subjects whose Tc exceeded 38.5?C, the error was only 0.25?C (SD = 0.20?C). Importantly, these results remained robust in the presence of simulated real-world operational conditions, yielding modestly larger errors (3 to 16%) when measurements were missing (40%) or laden with added noise. In addition, we demonstrated the capability of the AI system to continually learn an individual’s response and accurately capture the variations in Tc for the same individual under multiple exertional and environmental conditions. Collectively, our results demonstrate that the AI system provides accurate, real-time, individualized Tc estimates that can replace invasive Tc measurements, even when using corrupted non-invasive measurements.