Lower Extremity Muscle Activation Pattern Grouping Techniques by the Unsupervised Machine Learning-Driven Integrated Wearable System: A Proof-of-Concept Study

Le résumé fourni par la source

Accurate classification of muscle activation patterns is crucial for advancing rehabilitation, as it enables objective and physiologically informed patient grouping. Nevertheless, this approach remains underutilized in lower extremity rehabilitation due to the lack of standardized evaluation methods and robust implementation frameworks. To address this gap, we developed and clinically validated an interpretable, multi-granularity unsupervised grouping methodology using an integrated smart compression stocking system for objective monitoring of lower extremity muscle activation in healthy young adults. Twelve subjects performed maximal voluntary isometric contractions (MVIC) of the ankle plantar flexor muscles, generating a dataset of alpha values, each calculated as the ratio of normalized muscle torque measured by the Humac NORM equipment to the normalized, synchronized readout from the smart system, and indicating lower extremity muscle strength, density, and quality. The dataset used here was publicized in our previous study. Dimensionality reduction methods, including Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE), were applied to visualize the data, revealing implicit distinctions in muscle activation patterns between sex groups. Clustering methods, including K-Means, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and Agglomerative Hierarchical Clustering (AHC), consistently identified two optimal clusters, achieving a 0.2890 silhouette score and a 4.9997 Calinski–Harabasz (CH) index. The results demonstrated 75% accuracy and an 80% F1 score in distinguishing masculine and feminine activation groups, substantially outperforming the baseline dummy model. Sex emerged as the primary implicit factor influencing muscle activation patterns, with body mass index (BMI) as a secondary implicit factor. These findings highlight the potential of integrating smart wearable systems with unsupervised machine learning to support more personalized and accessible rehabilitation interventions.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.