Estimation of core body temperature from skin temperature, heat flux, and heart rate using a Kalman filter

• Published in: Computers in biology and medicine Vol. 99; pp. 1 - 6
• Main Authors: Welles, Alexander P., Xu, Xiaojiang, Santee, William R., Looney, David P., Buller, Mark J., Potter, Adam W., Hoyt, Reed W.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.08.2018; Elsevier Limited
• Subjects: Armed forces; Bias; Body fat; Body mass; Body temperature; Computational physiology; Consent; Covariance; Data collection; Environmental health; Fitness equipment; Forehead; Heart rate; Heat; Heat flux; Heat transfer; Human subjects; Humidity; Kalman filters; Medicine; Metabolism; Physiological modeling; Physiological status monitoring; Real-time estimation; Relative humidity; Research centers; Root-mean-square errors; Scapula; Sensors; Skin; Skin temperature; Sternum; Temperature effects; Thigh; Volunteers; United States > US; Physiological modeling; Physiological status monitoring; Computational physiology; Real-time estimation
• ISSN: 0010-4825; 1879-0534
• DOI: 10.1016/j.compbiomed.2018.05.021

Core body temperature (TC) is a key physiological metric of thermal heat-strain yet it remains difficult to measure non-invasively in the field. This work used combinations of observations of skin temperature (TS), heat flux (HF), and heart rate (HR) to accurately estimate TC using a Kalman Filter (KF). Data were collected from eight volunteers (age 22 ± 4 yr, height 1.75 ± 0.10 m, body mass 76.4 ± 10.7 kg, and body fat 23.4 ± 5.8%, mean ± standard deviation) while walking at two different metabolic rates (∼350 and ∼550 W) under three conditions (warm: 25 °C, 50% relative humidity (RH); hot-humid: 35 °C, 70% RH; and hot-dry: 40 °C, 20% RH). Skin temperature and HF data were collected from six locations: pectoralis, inner thigh, scapula, sternum, rib cage, and forehead. Kalman filter variables were learned via linear regression and covariance calculations between TC and TS, HF, and HR. Root mean square error (RMSE) and bias were calculated to identify the best performing models. The pectoralis (RMSE 0.18 ± 0.04 °C; bias −0.01 ± 0.09 °C), rib (RMSE 0.18 ± 0.09 °C; bias −0.03 ± 0.09 °C), and sternum (RMSE 0.20 ± 0.10 °C; bias −0.04 ± 0.13 °C) were found to have the lowest error values when using TS, HF, and HR but, using only two of these measures provided similar accuracy. •Core temperature is estimated from heart rate, skin temperature, and heat flux.•Accuracy statistics of all combinations of location and sensor type are reported.•The pectoralis and sternum were found to have the lowest error values.•Lowest root mean square error and bias were 0.18 ± 0.04 and −0.01 ± 0.09 °C.•Method can be used when direct measurement of core body temperature is impractical.