Wearable IMU-Based Human Activity Recognition Algorithm for Clinical Balance Assessment Using 1D-CNN and GRU Ensemble...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Artificial Intelligence in Healthcare

Background:

• Balance assessment is crucial for patient rehabilitation.
• Traditional methods like the Berg Balance Scale (BBS) require manual scoring.
• Existing machine learning approaches for automated BBS scoring have limitations.

Purpose of the Study:

• To enhance automated balance assessment using a wearable inertial measurement unit (IMU) system.
• To develop and optimize a deep learning algorithm for BBS scoring.
• To improve the accuracy and efficiency of balance evaluation in clinical settings.

Main Methods:

• Development of an automatic scoring algorithm utilizing a wearable IMU system.
• Implementation of a deep learning model combining 1D Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU).
• Optimization through a stacking ensemble model, data preprocessing (down-sampling, sampling rate adjustment), and data augmentation.

Main Results:

• The optimal stacking ensemble model comprised two 1D-CNN heads and one GRU head with a simple meta-learner.
• Data preprocessing and augmentation significantly improved model performance and addressed data imbalance.
• The developed model achieved a maximum accuracy of 98.4% across 14 BBS tasks.

Conclusions:

• The proposed deep learning approach significantly outperforms previous methods for automated BBS scoring.
• Wearable IMU systems integrated with advanced AI offer a promising tool for objective and accurate balance assessment.
• This technology has the potential to revolutionize clinical balance evaluations, improving patient outcomes.