Estimating dynamic plantar pressure distribution from wearable inertial sensors using a hybrid CNN-BiLSTM architecture

Purpose: Plantar pressure distribution is a crucial indicator in gait analysis, with significant value in clinical diagnoses and sports optimization. Traditional measurement methods, however, are often limited by expensive equipment and laboratory settings. This study aimed to develop an accurate, portable and cost-effective method using a deep learning model based on data from wearable Inertial Measurement Units (IMU) to predict comprehensive plantar pressure distributions.

Methods: We proposed a hybrid model combining a Convolutional Neural Network (CNN) and a Bidirectional Long Short-Term Memory (BiLSTM) network. The CNN extracts local features from IMU data; the BiLSTM captures temporal dependencies; a temporal attention mechanism optimizes the prediction of key time steps; and body weight information is integrated to accommodate individual differences.

Results: Experimental results show that in 10-fold cross-validation, the model achieves a Mean Squared Error of 0.98 and a Structural Similarity Index of 0.89, demonstrating excellent prediction accuracy and distribution similarity.

Conclusions: This study provides a cost-effective method for plantar pressure analysis, which is expected to be integrated into wearable devices for real -time gait monitoring, with applications in rehabilitation and sports optimization.