Efficiency of deep neural networks for joint angle modeling in digital gait assessment

Abstract Reliability and user compliance of the applied sensor system are two key issues digital healthcare biomedical informatics. For gait assessment applications, accurate joint angle measurements important. Inertial measurement units (IMUs) have been used in a variety applications can also provide significant information on kinematics. However, nonlinear mechanism human locomotion results moderate estimation accuracy kinematics thus angles. To develop “digital twins” as counterpart body lower limb angles, three-dimensional kinematic data were collected. This work investigates different neural networks modeling angles sagittal plane using records single IMU attached to foot. The evaluation based root mean square error (RMSE) show that long short-term memory (LSTM) deliver superior performance compared other machine learning (ML) approaches. Accordingly, deep LSTM architecture is promising approach IMU, reduce required physical IMUs subject improve practical application system.