| dc.description.abstract | Shortly after birth, healthy infants exhibit so-called fidgety movements, while infants who later develop cerebral palsy (CP) lack these movements. General Movement Assessment (GMA) which is a clinical method, has proven its accuracy in detecting the absence (or presence) of fidgety movements, but for practical reasons, this method has not been adopted widely in the clinics. In order to create a similar but objective computer-based approach, Berg (2008) and Meinecke (2006) have studied discriminative features based on movement data collected from electromagnetic sensors and video. In this thesis, in addition to evaluation and comparison of previously introduced features, different classification methods have been applied to a suboptimal subset of these features. The results from linear and nonlinear separability analyses of features, confirm that dynamic features have better descriptive capabilities compared to statistically characterized features. Furthermore, it turns out that fidgety movements in the head (neck) and the arms show significant potential in distinguishing normal and abnormal infants, compared to signals from the trunk and the feet. The achieved results show 86% sensitivity and 90% specificity, which are highly acceptable, but this study needs further attendance before having any clinical usability. This study contains the first step of a typical medical research, meaning that the global (generalized) validity of the implemented methods are yet to be investigated, suppose that a representative selection (data) is available. | nb_NO |
