A new approach for the field detection of sleep bruxism based on inertial sensor data and machine learning classification.

Tracking of nightly oral parafunctional activities, such as grinding and clenching, is useful to monitor sleep bruxism. Although polysomnography remains the gold standard for overnight observations, home-based long-term recordings are wished, to ensure time-distributed data and better patient compliance. Technologies based on inertial measurement unit (IMU) sensors allow the production of compact, and easy-to-use sensing instruments. This study aimed to develop a proof of concept using IMUs to capture mandibular motion for sleep bruxism assessment and to evaluate the accuracy of machine learning algorithms in classifying motions. A mobile recording setup was developed, incorporating an IMU equipped with tri-axial accelerometer, gyroscope, and magnetometer. A set of 21 in-vivo recordings from three individuals containing grinding and opening/closing mandibular movements was collected for ML algorithm training. The data was manually labeled and divided into train and test data (80/20). Several models were trained to classify the mandibular motion. Overall, the trained models were capable of correctly classifying up to 96% of the test data. The best results were obtained when data from all three sensors were used simultaneously. The results indicate that IMU sensors are a valuable option to assess mandibular motion for a small-scale wearable device.