A Novel Myo-Based Hybrid Neural Network for Tooth Brushing Monitoring

Twice-a-day tooth brushing is the most cost-effective means of maintaining oral health. Standard BASS brushing requires no less than 3 minutes to cover all tooth areas. However, the process of brushing teeth is relatively boring, especially the inner area of the mouth is difficult to clean, and it is difficult to ensure the comprehensive coverage and effective time of all tooth areas for each brushing without supervision. This paper proposed a novel tooth brushing evaluation system based on the wearable myoelectricity signals collector-Myo. A hybrid neural network model that consists of residual convolution and LSTM blocks to process the built-in myoelectricity and IMU signals is designed to detect the brushing area as a timer during tooth brushing. Experiments on data collected from 10 volunteers show that the final model can achieve an average accuracy of 95.66% on the 17 predefined tooth brushing behaviors. The system can remind the user of the insufficiently cleaned area by recording the cleaning time of each tooth area during the user’s brushing process, thereby prompting the user to further clean the corresponding area.