Optimized Clustering Techniques for Gait Profiling in Children with Cerebral Palsy for Rehabilitation

Abstract

Cerebral palsy (CP) is a neuro-development disease in children. It is quite an intricate task to categorize gait pattern into normal and CP based pathology. In this study, nature-inspired meta-heuristic algorithms are explored on a publicly available gait dataset of 156 subjects for automatic gait profiling of children with cerebral palsy. Five cases are considered to explore the feature selection criteria before applying clustering technique. Finding the optimal number of clusters is a challenging task in the unsupervised learning area. In this study, an optimal number of gait profiles in the datasets is identified based on voting from mean square error, silhouette coefficient and Dunn index. The study demonstrates that optimized based gait profile clusters could assist quantitatively in clinical rehabilitation evaluation for the children affected by CP.