Deep Neural Networks for Gait Cycle Percentage Prediction in Frail Older Adults Using a Foot-Mounted IMU

Accurate gait phase prediction is crucial for effective gait training interventions, which are essential for preventing falls in the elderly. This study introduces a novel gait cycle percentage prediction method using end-to-end deep learning networks (DNNs) through a foot-mounted inertial measurement unit (IMU). The model is trained and tested on the GSTRIDE database, comprising data from 158 frail older adults with a total of 176,470 recorded strides. Utilizing 10-fold cross-validation, the model demonstrates a mean absolute error (MAE) of 0.04 and a root mean square error (RMSE) of 0.11. The proposed architecture enables real-time prediction, analyzing each stride in under 0.007 seconds. This demonstrates its potential for integration into wearable devices, allowing continuous monitoring and feedback for gait training.