A Smartwatch-based Approach for Oral Health Monitoring using Deep Learning

Monitoring toothbrushing is of great importance to promote the improvement of oral health habits. In this study wearable devices and artificial intelligence methods are combined to identify 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. As a benchmarking dataset, we train and test the model with a public database containing signals obtained from a wide set of experimental subjects performing a variety of different manual actions. The results show the high effectiveness of the detector, which achieves quality metrics (sensitivity and specificity) higher than 95% when it is employed to discriminate brushing actions from other 23 hand movements. This performance is improved if the architecture is refined through a simple low-pass filtering of the classifier's decisions, aimed at correcting the classification of anomalous and isolated mobility patterns.