Automatic classification of gait patterns in children with cerebral palsy using fuzzy clustering method

Background: Subjective classification of gait pattern in children with cerebral palsy depends on the assessor's experience, while mathematical methods produce virtual groups with no clinical interpretation. Methods: In a retrospective study, gait data from 66 children (132 limbs) with a mean age of 9.6 (SD 3.7) years with cerebral palsy and no history of surgery or botulinum toxin injection were reviewed. The gait pattern of each limb was classified in four groups according to Rodda using three methods: 1) a team of experts subjectively assigning a gait pattern, 2) using the plantarflexor-knee extension couple index introduced by Sangeux et al., and 3) employing a fuzzy algorithm to translate the experiences of experts into objective rules and execute a clustering tool. To define fuzzy repeated-measures, 75% of the members in each group were used, and the remaining were used for validation. Eight parameters were objectively extracted from kinematic data for each group and compared using repeated measure ANOVA and post-hoc analysis was performed. Finally, the results of the clustering of the latter two methods were compared to the subjective method. Findings: The plantarflexor-knee extension couple index achieved 86% accuracy while the fuzzy system yielded a 98% accuracy. The most substantial errors occurred between jump and apparent in both methods. Interpretation: The presented method is a fast, reliable, and objective fuzzy clustering system to classify gait patterns in cerebral palsy, which produces clinically-relevant results. It can provide a universal common language for researchers.