Epidermal Graphene Sensors and Machine Learning for Estimating Swallowed Volume

Abstract

Assessment of liquid intake is necessary to obtain a complete picture of an individual's hydration status. Measurements using state-of-the-art wearable devices have been demonstrated, but none of these devices have combined high sensitivity, unobtrusiveness, and automated estimation of volume, i.e., using machine learning. Such a capability would have immense value in a variety of medical contexts, such as monitoring patients with dysphagia or the performance of athletes. Here, an epidermal sensor platform is combined with machine learning to measure swallowed liquid volume based on signals obtained from the surface of the skin. The key component of the device is a composite piezoresistive sensor consisting of single-layer graphene decorated with metallic nanoislands and coated with a highly plasticized form of the conductive polymer poly(3,4-ethylenedioxythiophene):poly(styrenesuflonate) (PEDOT:PSS). Surface electromyography (sEMG) signals obtained with conventional electrodes are used in concert with the strain measurements. The use of strain and sEMG measurements together both (1) improve the accuracy of estimated volumes and (2) permit the differentiation of swallowing from motion artifacts. In a cohort consisting of 11 participants, the combined measurements of strain and sEMG - processed by the machine learning algorithm - were able to estimate unknown swallowed volumes cumulatively between 5 and 30 mL of water with greater than 92% accuracy. Ultimately, this system holds promise for numerous applications in sports medicine, rehabilitation, and the detection of nascent dysfunction in swallowing.