Abstract
PURPOSE: Plantar soft tissue properties affect foot biomechanics during movement. This study aims to explore the relationship between plantar pressure features and soft tissue stiffness through interpretable neural network model. The findings could inform orthotic insole design. METHODS: A sample of 30 healthy young male subjects with normal feet were recruited (age 23.56 ± 3.28 years, height 1.76 ± 0.04 m, weight 72.21 ± 5.69 kg). Plantar pressure data were collected during 5 trials at the subjects' preferred walking speed (1.15 ± 0.04 m/s). Foot soft tissue stiffness was recorded using a MyotonPRO biological soft tissue stiffness meter before each walking trial. A backpropagation neural network, optimized by integrating particle swarm optimization and genetic algorithm, was constructed to predict foot soft tissue stiffness using plantar pressure data collected during walking. Mean impact value analysis was conducted in parallel to investigate the relative importance of different plantar pressure features. RESULTS: The predicted values for the training set are slightly higher than the actual values (MBE = 0.77N/m, RMSE = 11.89 N/m), with a maximum relative error of 7.82% and an average relative error of 1.98%, and the predicted values for the test set are slightly lower than the actual values (MBE = -4.43N/m, RMSE = 14.73 N/m), with a maximum relative error of 7.35% and an average relative error of 2.55%. Regions with highest contribution rates to foot soft tissue stiffness prediction were the third metatarsal (13.58%), fourth metatarsal (14.71%), midfoot (12.43%) and medial heel (12.58%) regions, which accounted for 53.3% of total contribution. CONCLUSION: The pressure features in the medial heel, midfoot area, and lateral mid-metatarsal regions during walking can better reflect plantar soft tissue stiffness. Future studies should ensure measurement stability of this region and refine insole designs to mitigate plantar soft tissue fatigue in the specified areas.