Proceedings zyxwvutsrqponm of the zyxwvutsrqpo 25* Annual Intemational Conference of the IEEE EMBS Cancun, Mexico September 17-2 1,2003 Dynamic Time Warping in the study of ERPs in dyslexic children S. Casarotto’, S. Cerutti’, A. M. Bianchi’, G. A. Chiarenza’ ‘Department of Biomedical Engineering, Polytechnic University, Milano, Italy *Department of Child and Adolescence Neuropsychiatry, Az. Osp. zyxwvu G. Salvini, Rho Hospital, Rho, Milano, Italy Abstract-Aim of this paper is to compute a normal ERP pattern (template) and to quantify the morphological characteristics of ERPs. ERPs were recorded from normal and dyslexic children in a passive and in an active condition. Dynamic Time Warping @TW) was used to align the averaged zyxwvu ERPs of normal subjects. After the computation of the templates, individual averaged ERPs were aligned with the corresponding template, in order to automates the identification of the relevant ERPs components. The latencies of these components in normal and dyslexic subjects were compared in the two different conditions. ERPs components of dyslexic children were always delayed in respect to normal children. Statistically significant latency differences were noticed both in the short-latency waves, related to attention mechanisms, and in the long-latency waves, presumably related to memory processes. P1 latency in T4 differed in the two groups of children in both tasks, with pXO.05. Morphological differences between the ERPs of normal and dyslexic children were also noticed in the right hemisphere: N2 latency in T4 differed in the two groups of children with p<O.O5. This result suggests that dyslexia is associated with more general disruptions of the cerebral functions than that confined to the classical linguistic areas. Keywords-Dynamic Time Warping, ERPs, dyslexia, reading-related potentials. I. INTRODUCTION Developmental dyslexia is a neurological disorder characterized primarily by reading difficulties despite average intelligence, adequate education and normal sensory acuity. The aetiology of this condition is still unknown: the most recent theories hypothesize a genetic disruption of the cerebral structure, that would produce compromised phonological awareness and visual/auditory perception [I]. In order to investigate the reading processes, event-related potentials (ERPs) were recorded in normal children and in children with developmental dyslexia. This approach is particularly efficacious for our purposes because it allows to investigate the functional aspects of cerebral activities with an high temporal resolution, in a non-invasive way. Comparing ERPs from different subjects is difficult for the high inter-individual variability of the morphology: dealing with children performing cognitive tasks this variability greatly increases. As a consequence, it is difficult to define a normal “pattern”. Usually, a grand-average is evaluated from a group of normal subjects, but this is heavily affected by inter-individual variability. In this work we propose an approach based on Dynamic Time Warping (DTW) technique for the automatic alignment of the waves and quantification of the morphological characteristics of ERPs. The method allows the calculation of a reliable template for normal subjects, which can be compared with pathological signals. 11. METHODOLOGY Dynamic Time Warping (DTW) is a non-linear alignment algorithm that reduces the temporal differences between signals with similar morphologies through local compressions and extensions of their temporal axes [2,3]. The DTW procedure consists in performing two different steps on paired signals: i) calculation of a Warping Function (WF) representing the best alignment between the signals; ii) computation of a new waveform (template) by averaging the original signals according to the WF. Given two signals x={xj zyxwvut 1<i<l} zyx (1) y={yj l<j<J} zyx I , .... ” ......... _.____I._I_ ~ ...... 5: ’I H’ ,__---’ . . zyx f,‘ 100 ms (b) Fig. 1. a) Plot of the Warping Function computed on the sampled signals x and y. b) Thin lines represent the original signals x (top) and y (bottom). Thin segments represent the temporal correspondence between morphologically similar samples of the two signals. Thick line represents the template resulting from the alignment of x and y 0-7803-7789-3/03/$17.00 02003 IEEE 2311