Core Temperature Estimation Using Wearable Earbud‐Type Thermometer and Machine Learning During Light‐Intensity Cycling Under Varying Indoor Ambient Temperatures

Continuous monitoring of core temperature is crucial for optimizing exercise performance, promoting health, and ensuring safety. Here, we evaluated the validity of a machine learning-based earbud-type core temperature sensor during exercise. At ambient temperatures of 10°C, 20°C, and 30°C, participants rested and then cycled for 25 min (90 W for males and 60 W for females), with an artificial wind of ~3.0 m/s directed at the face during the final 10 min of exercise. Throughout the experiment, two thermistors embedded in the earbud-type device monitored both ambient temperature and internal ear temperature. Additionally, an infrared temperature sensor integrated into the earbud-type thermistor measured internal ear temperature. Core temperature was measured through a machine learning-based approach, and esophageal temperature served as the reference. We created mixed-effects Bland-Altman plots for the relationship between estimated and esophageal temperature, and found that the bias was -0.003°C with a limit of agreement of -0.52°C-0.51°C. Root mean squared error and Pearson's r for the two-temperature relationship were 0.26°C and 0.81, and 76.9% of the data exhibited temperature differences within a margin of ≤ 0.3°C between the two measurements. Similar results were also observed with artificial wind. Including infrared temperature measurements did not enhance the aforementioned variables. We show that a machine learning-based wearable earbud-type thermometer can validly estimate core temperature in exercising individuals across 10°C-30°C indoor ambient temperatures with and without exposure to a wind speed of ~3.0 m/s. However, further improvement of the algorithm is needed to enhance estimation accuracy.