An experimental approach to automatically classify children with cerebral palsy

This experimental study is aimed at automatically classify the large indistinct group of children affected by diplegic Cerebral Palsy (CP), into four main clinical forms characterized by homogeneous walking patterns as clinically proposed by Ferrari et al. [1]. The classification is based on the use of data gathered on 104 patients at LAMBDA (Laboratorio per l’Analisi del Movimento del Bambino DisAbile, S.M.Nuova Hospital, Reggio Emilia, Italy) by means of an 8 cameras optoelectronic system (Vicon, UK) and the Total3Dgait biomechanical protocol [2]. This paper deals with the use of expert systems, such as artificial neural networks (ANN), that are able to learn by examples. ANN has been widely used in mechanics for the resolution of pattern recognition and diagnostic problems [3,4]. Our aims is to extend the use of these expert systems to classify the four diplegic forms through the analysis of the rotation angles of the lower limb joints along multiple gait cycles. On each walking trial of each patient a set of synthetic statistical parameters representative of joint angles, has been used as input to train a supervised ANN in order to divide patients into the four forms a priori clinically determined. The first attempt consists in a Feed-Forward network trained considering data coming from all the four forms. The effectiveness of the resulting ANN has been proved on a dataset acquired by new patiens. In this case, the performace on the form recognition was encouraging but still far from an implementation in clinical routine. In order to improve the efficiency we simplified the recognition problem by creating a network which was asked first to distinguish between two macro-categories, consisting of two forms each. On the basis of the two splitted dataset returned by this first ANN layer, two further network were trained seperately. To these network was finally asked to provide the affinity with one of the original four forms. In this “system of networks” case, the overall performance increased significantly becoming clinically meaningful.