Automatic Classification of Barefoot and Shod Populations Based on the Foot Metrics and Plantar Pressure Patterns

Abstract

The human being’s locomotion under the barefoot condition enables normal foot function and lower limb biomechanical performance from a biological evolution perspective. No study has demonstrated the specific differences between habitually barefoot and shod cohorts based on foot morphology and dynamic plantar pressure during walking and running. The present study aimed to assess and classify foot metrics and dynamic plantar pressure patterns of barefoot and shod people via machine learning algorithms. One hundred and forty-six age-matched barefoot (n = 78) and shod (n = 68) participants were recruited for this study. Gaussian Naïve Bayes were selected to identify foot morphology differences between unshod and shod cohorts. The support vector machine (SVM) classifiers based on the principal component analysis (PCA) feature extraction and recursive feature elimination (RFE) feature selection methods were utilized to separate and classify the barefoot and shod populations via walking and running plantar pressure parameters. Peak pressure in the M1-M5 regions during running was significantly higher for the shod participants, increasing 34.8, 37.3, 29.2, 31.7, and 40.1%, respectively. The test accuracy of the Gaussian Naïve Bayes model achieved an accuracy of 93%. The mean 10-fold cross-validation scores were 0.98 and 0.96 for the RFE- and PCA-based SVM models, and both feature extract-based and feature select-based SVM models achieved an accuracy of 95%. The foot shape, especially the forefoot region, was shown to be a valuable classifier of shod and unshod groups. Dynamic pressure patterns during running contribute most to the identification of the two cohorts, especially the forefoot region.

Toward improved understanding of foot shape, foot posture, and foot biomechanics during running: A narrative review

“…Also, the midfoot shape metric (artificial intelligence) was associated with lateral forefoot pressures during walking in both habitually barefoot and shod populations. An improved predictive statistical model (support vector machine) also showed an increased prediction accuracy on the basis of the same data set ( Xiang et al, 2022a ). Rapid prediction of foot function from easily measured foot shape metrics may become the norm and extend into clinical diagnosis of foot pathologies and customized footwear development.…”

Rethinking running biomechanics: a critical review of ground reaction forces, tibial bone loading, and the role of wearable sensors

“…Understanding the role of external TA in both external impact loading and internal tibial bone loading, therefore, becomes crucial ( Matijevich et al, 2019 ). Enhancing the evaluation of these factors through machine learning not only presents an intriguing area of research but also holds substantial potential implications for future applications in sports medicine, injury prevention, and rehabilitation strategies ( Zadpoor and Nikooyan, 2011 ; Johnson C. D. et al, 2020 ; Xiang et al, 2022a ; Xiang et al, 2022b ; Gao et al, 2023 ; Lloyd et al, 2023 ; Uhlrich et al, 2023 ; Xiang et al, 2023 ).…”

Automated recognition of asymmetric gait and fatigue gait using ground reaction force data

“…Bilateral limb asymmetry may not be evident during the initial stages of running but may arise as muscle fatigue and/or exercise intensity changes ( Arampatzis et al, 1999 ; Xiang et al, 2022a ). Mizrahi et al (2000) demonstrated that running-induced lower limb muscle fatigue increases the tibia’s vertical acceleration during heel strike, suggesting that more load accumulates in the shank after running fatigued.…”