Background and Objectives: Lower limb dysfunction is a prevalent complication following stroke. This study aimed to identify key gait features influencing balance function in stroke patients through multidimensional gait analysis and machine learning, thereby guiding clinical rehabilitation strategies. Methods: The study included 46 stroke patients (59.78±8.4 y old) and 75 healthy controls (62.49±5.03 y old). Balance function was assessed using the Berg Balance Scale (BBS). Gait parameters during natural walking were captured using inertial sensors and foot pressure systems, with participants categorized into hemiplegic-side, non-hemiplegic-side, and healthy control groups. Principal component analysis (PCA) was employed to reduce dimensionality, retaining 90% of the variance. Machine learning models, including gradient boosting decision tree, support vector machine, logistic regression, random forest, and K-nearest neighbors, were evaluated via stratified 5-fold cross-validation. Key features were identified using average absolute coefficient values (AACV) and validated through Spearman correlation with BBS. Group differences were also analyzed. Results: AACV revealed that the average toe-off angle (1.6314) and heel-strike angle (0.9362) were critical parameters. The logistic regression model achieved optimal performance with an accuracy of 88.2% and an F1-score of 0.879. Both angles significantly correlated with BBS scores (toe-off: r=0.295, P =0.0042; heel-strike: r=0.497, P <0.001). Group comparisons demonstrated impaired angles in hemiplegic limbs compared with non-hemiplegic limbs and healthy controls (all P <0.001), following a gradient of hemiplegic < non-hemiplegic < healthy. Conclusions: This study concludes that toe-off angle and heel-strike angle are key parameters influencing lower limb balance function in stroke patients. These findings suggest that targeted training for the ankle plantar flexion and dorsiflexion muscle groups is highly significant for the gait rehabilitation of stroke patients’ lower limbs.
Liu et al. (Tue,) studied this question.