JMIR Publications

ABSTRACT

Background:

Radiographic knee osteoarthritis (ROA), defined by radiographic degenerative changes in the knee joint, is associated with abnormal plantar loading and altered gait coordination during walking. Insole-based plantar-pressure sensing offers a practical wearable approach for ROA screening, but existing methods typically rely either on conventional biomechanical descriptors or on end-to-end temporal models, highlighting the need for an interpretable subject-level framework that jointly captures local biomechanical abnormalities and distributed coordination changes.

Objective:

This study aimed to develop an interpretable wearable plantar-pressure framework for subject-level screening of ROA by combining conventional biomechanical descriptors with a coordination-level representation, in order to capture inter-regional plantar coordination that may not be adequately characterized by conventional descriptors alone.

Methods:

Using a plantar-pressure dataset from a prior study by our group, we developed an interpretable subject-level screening framework that integrates conventional biomechanical phenotypes with a newly defined coordination-level Plantar Synergy Index (PSI). The framework was evaluated under a leakage-controlled subject-level split with an independent test set, together with additional analyses of feature stability, phase-resolved PSI patterns, robustness to perturbation and design variation, and comparisons with representative end-to-end temporal baselines.

Results:

The proposed framework achieved favorable subject-level performance on the independent test set (F1=0.842, AUC=0.818). Fold-wise stability analysis showed that stable retained descriptors were dominated by biomechanical features, whereas PSI contributed a smaller but reproducible set of coordination descriptors. Phase-resolved and lag-scale analyses suggested that ROA-related differences reflected broad changes in stance-phase coordination rather than isolated local abnormalities. The framework also remained robust under perturbation and design variation and compared favorably with representative end-to-end temporal baselines.

Conclusions:

We developed an interpretable subject-level framework for screening ROA from wearable plantar-pressure signals. By jointly modeling biomechanical phenotypes and coordination-level information, the framework achieved robust performance across multiple evaluations. These findings support plantar pressure as a potentially useful screening-oriented digital phenotype.