Implementation and analysis of an oral health monitoring system using smartwatches and convolutional neural networks.

Abstract

Activity tracking provided by smartwatches offers a cost-effective widespread solution for the recognition of hand movements, which makes them an appealing tool to promote permanent health monitoring and, in particular, oral hygiene habits among the population. This study proposes the integration of wearable devices and artificial intelligence methods to identify manual movements associated with tooth brushing. The focus is on utilizing convolutional neural networks to recognize brushing gestures based on short samples of accelerometer data collected from wrist-worn devices. The architecture is systematically trained and validated using long-term datasets collected with different smartwatch models during the daily routines of a small group of experimental users. The results show the high effectiveness and capability of generalization of the detector under LOSO cross validation both when it is evaluated in an offline way and when it is trained and implemented on a smartwatch to operate in real time in a real-life scenario.