Evaluation of Lower-Limb Brunnstrom Recovery Stage via High-density Plantar Pressure and Global Fuzzy Granular Support Vector Machine.

Low interrater reliability and inefficiency are present in the subjective clinical Brunnstrom recovery stage (BRS-LL) assessment for stroke patients. Although wearable technology offers solutions, existing BRS-LL automatic assessment studies face a trade-off between accuracy and ease of use: multimodal systems are accurate, but complex, while single-modal methods are simpler but less accurate. To address the complexity of sensor deployment, we develop flexible high-density (HD) plantar pressure (PP) sensing insoles (48 units) that naturally integrate into regular shoes without external modules. PP data are collected from 52 stroke patients. The high-dimensional 297 PP features are extracted to enhance signal representation. A global fuzzy granular support vector machine (GFGSVM) algorithm is proposed to overcome the accuracy limitations of unimodal studies. The results show that the increased PP sensing density from 12 to 48 units enhances feature-BRS-LL correlations (69% improved by over 20%) and BRS-LL classification accuracy by 8.1%-11.6%, highlighting the advantages of HD PP sensor. Through leave-one-subject-out cross-validation, GFGSVM achieves an accuracy of 95.9% sample level and 98.1% individual patient level, surpassing five popular evaluation algorithms by 12.8%-26.2%. The system's accuracy exceeds single-modal (+9.1%) and multimodal studies (+1.71%) by utilizing only a pair of HD PP insoles with GFGSVM. Overall, this study provides an efficient BRS-LL evaluation scheme that combines both portability for clinical applications and high assessment accuracy, effectively resolving the trade-off and offering an effective tool for long-term monitoring and screening of stroke patients.