Measurement of Sweat Gland Activity by Sweat Sensing and Deep Learning

Semi-continuous monitoring of general ward patients is expected to reduce hospital mortality. Sweat is a biomarker-rich biofluid with potential for non-invasive, semi-continuous patient monitoring. Efforts to correlate the biomarkers in sweat to the condition of a patient have yielded only limited results. The only clinically approved use of sweat biomarkers is the diagnosis of cystic fibrosis. Knowledge of the sweat rate per gland and the number of active sweat glands from which the sweat is sampled is difficult to obtain and yet plays an important role to accurately estimate the concentration of biomarkers found in sweat. To estimate the sweat rate per sweat gland a discrete sweat-sensing device can be used. A model was created to generate synthetic signals of such a discrete sweat-sensing device. In this work, we adopt a deep-learning strategy to estimate the sweat rate per gland, and the number of active sweat glands, based on the simulated signals, with the aim of demonstrating the feasibility of this approach. Our approach demonstrated its capability to estimate the sweat rate within a 10% margin of error across all tested datasets. The number of glands could be estimated with a minimum accuracy of 70%. This shows that deep learning is a promising method to interpret the signals of a discrete sweat-sensing device.