Tactile Sensor-Based Body Center of Pressure Estimation System Using Supervised Deep Learning Models.

Area of Science:

• Biomechanics and Biomedical Engineering
• Sensor Technology
• Machine Learning

Background:

• Center of pressure (CoP) is crucial for balance and fall risk assessment.
• Traditional force plates are expensive and impractical; current low-cost alternatives have limitations.
• Previous machine learning models for CoP estimation using sparse sensors showed significant mediolateral/anteroposterior (ML/AP) NRMSE differences (3.2-4.7%).

Purpose of the Study:

• To develop and evaluate a cost-effective tactile sensor-based system for estimating CoP using deep learning.
• To improve the accuracy and reduce directional imbalances in CoP estimation compared to prior methods.
• To assess the system's performance across different balance protocols.

Main Methods:

• Proposed a tactile sensor system employing deep learning models (CNN/ResNet encoders with Bi-LSTM) to analyze 2D pressure distribution images and temporal patterns.
• Collected data from 23 healthy adults performing four balance protocols.
• Compared ResNet-Bi-LSTM and CNN-Bi-LSTM models against baseline CNN-LSTM and Bi-LSTM using leave-one-out cross-validation (LOOCV), evaluating with RMSE, NRMSE, and R².

Main Results:

• The ResNet-Bi-LSTM model with angular features achieved the best performance, yielding RMSE values of 18.63 ± 4.57 mm (ML) and 17.65 ± 3.48 mm (AP).
• Significantly reduced the ML/AP NRMSE difference to 1.3%, a substantial improvement over the 3.2-4.7% reported previously.
• Demonstrated superior performance with the lowest RMSE across models, particularly under dynamic balance protocols.

Conclusions:

• Tactile sensor-based systems integrated with advanced deep learning offer a promising, cost-effective alternative to force plates for CoP measurement.
• The proposed ResNet-Bi-LSTM model significantly enhances CoP estimation accuracy and reduces directional bias.
• Potential applications include gait analysis, real-time balance monitoring, and fall risk assessment, with future validation in patient populations planned.