FeverPhone
Joseph Breda, Mastafa Springston, Alex Mariakakis, Shwetak Patel
Abstract
Smartphones contain thermistors that ordinarily monitor the temperature of the device's internal components; however, these sensors are also sensitive to warm entities in contact with the device, presenting opportunities for measuring human body temperature and detecting fevers. We present FeverPhone --- a smartphone app that estimates a person's core body temperature by having the user place the capacitive touchscreen of the phone against their forehead. During the assessment, the phone logs the temperature sensed by a thermistor and the raw capacitance sensed by the touchscreen to capture features describing the rate of heat transfer from the body to the device. These features are then used in a machine learning model to infer the user's core body temperature. We validate FeverPhone through both a lab simulation with a skin-like controllable heat source and a clinical study with real patients. We found that FeverPhone's temperature estimates are comparable to commercial off-the-shelf peripheral and tympanic thermometers. In a clinical study with 37 participants, FeverPhone readings achieved a mean absolute error of 0.229 °C, a limit of agreement of ±0.731 °C, and a Pearson's correlation coefficient of 0.763. Using these results for fever classification results in a sensitivity of 0.813 and a specificity of 0.904.