Neural Network Estimation of Eardrum Temperature Using Multiple Sensors Integrated on a Wristwatch-Sized Device

Abstract

In this study, a novel system to help estimate eardrum temperature utilizing neural networks as well as data acquired from multiple sensors integrated on a wristwatch-type wearable device is successfully demonstrated. Conventional estimation methods, which use a heat balance model of the body, cannot be applied as it needs parameters, such as the thermal index of the body and the information on clothes, which cannot be measured by the wristwatch-type device. We introduced sensors that measure environmental quantities and vital signals to experimentally acquire sensing data from outdoor locales. To improve the estimation accuracy, time series data were input to a neural network, and the system was optimized by comparing the estimation accuracy while varying the time length and time interval of each sensor. By setting the time interval to 10 s and the time length to 30 as one dataset, the standard deviation of the error between the measured and the estimated eardrum temperatures is 0.046 °C, and the ratio of the data where the error between the measured and estimated values was within 0.1 °C was 97%.