Single Inertial Sensor-Based Neural Networks to Estimate COM-COP Inclination Angle During Walking

Area of Science:

• Biomechanics
• Biomedical Engineering
• Machine Learning

Background:

• Gait stability is crucial for reducing fall risk, particularly in vulnerable populations.
• The center of mass-center of pressure inclination angle (COM-COP IA) is a key parameter for assessing postural control and balance recovery.
• Current methods for measuring COM-COP IA can be complex and expensive.

Purpose of the Study:

• To develop and evaluate artificial neural network (ANN) models for estimating COM-COP IA using inertial sensor data.
• To compare the performance of different ANN architectures, specifically feed-forward ANNs and long-short term memory (LSTM) networks.
• To assess the impact of low-pass filter cutoff frequencies on the accuracy of COM-COP IA estimation.

Main Methods:

• An inertial measurement unit (IMU) prototype was developed, incorporating accelerometer, gyroscope, and magnetometer sensors.
• The COM-COP IA was accurately measured using a 3D motion analysis system with force plates as ground truth.
• Feed-forward ANN and LSTM network models were trained to predict COM-COP IA from IMU signals.

Main Results:

• The LSTM network achieved a significantly lower relative root-mean-square error (rRMSE) of 9% compared to the feed-forward ANN's 15% rRMSE.
• The LSTM model demonstrated robust accuracy, remaining stable across various cutoff frequencies of the low-pass filter applied to input signals.
• The study successfully validated the feasibility of estimating COM-COP IA using a cost-effective inertial sensor system.

Conclusions:

• Artificial neural networks, particularly LSTM, can accurately estimate COM-COP IA from low-cost inertial sensors, providing valuable insights into gait stability.
• This technology holds potential for developing portable systems to monitor the balancing ability of elderly individuals and patients with impaired balance.
• The findings support the integration of machine learning and wearable sensors for accessible and effective balance assessment.