Ankle Kinematics Estimation Using Artificial Neural Network and Multimodal IMU Data.

Area of Science:

• Biomechanics and Biomedical Engineering
• Artificial Intelligence in Healthcare
• Wearable Sensor Technology

Background:

• Inertial measurement units (IMUs) offer portable joint kinematics monitoring but suffer from accuracy limitations, real-time processing challenges, and complex calibration needs.
• Existing methods often require precise sensor-to-segment calibration, hindering widespread clinical and daily application of IMU-based motion analysis.

Purpose of the Study:

• To introduce KEEN (KinEmatics Estimation Network), an innovative framework utilizing lightweight artificial neural networks (ANNs) for real-time, calibration-free multi-plane ankle kinematics prediction.
• To evaluate the efficacy of minimal IMU configurations and various ANN models for accurate ankle motion tracking.

Main Methods:

• Developed and evaluated five ANN algorithms, including Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) models, using 42 inputs from four IMUs.
• Assessed model performance in both intra-subject and inter-subject tasks to determine generalization capabilities.
• Investigated the feasibility of deploying a CNN model on a microcontroller for real-time kinematic estimation using a single heel-mounted IMU.

Main Results:

• A single heel-mounted IMU, when processed by the CNN model, provided clinically acceptable ankle kinematics estimations (RMSE: 4.13° ±0.55°).
• The LSTM network excelled in intra-subject tasks (RMSE: 1.88° ±0.02°), while CNN and CNN-LSTM models demonstrated superior inter-subject generalization.
• Real-time deployment of the CNN on a microcontroller with a single IMU yielded promising results (RMSE: 3.34° ±0.48°), demonstrating practical applicability.

Conclusions:

• KEEN framework effectively leverages ANNs and minimal IMUs for accurate, real-time ankle kinematics, overcoming calibration barriers.
• Minimal IMU configurations, particularly a single heel IMU with CNN, show significant potential for cost-effective and practical clinical applications.
• This approach offers a viable solution for early prevention and rehabilitation of ankle injuries, enhancing accessibility and usability.