Summary

Objective: Evaluation of spontaneous infant movements is an important tool for the detection of neurological impairments. One important aspect of this evaluation is the observation of movements which exhibit certain complex properties. This article presents a method to automatically extract segments which contain such complex patterns in order to quantitatively assess them. Methods: Expert knowledge is represented in a principal component model which captures the term complexity as the multivariate interactions in the kinematic chains of the upper and the lower limb. A complexity score is introduced which is used to quantify the similarity of new movements to this model. It was applied to the recordings of 53 infants which were diagnosed by physicians as normal or pathologic. Results: Time segments marked as complex (from five infants) by physicians could be detected with a mean accuracy of 0.77 by the automated approach. The median of the best complexity scores of the pathologic group (n = 21) is significantly lower (p = 0.001) than the median of the normal group (n = 27). Conclusion: Using the complexity score we were able to quantify movement complexity in regard of the understanding of physicians. This could be useful for clinical applications.