A statistical procedure to assist dysgraphia detection through dynamic modelling of handwriting

Abstract

Dysgraphia is a neurodevelopmental condition in which children encounter difficulties in handwriting. Dysgraphia is not a disorder per se, but is secondary to neurodevelopmental disorders, mainly dyslexia, Developmental Coordination Disorder (DCD, also known as dyspraxia) or Attention Deficit Hyperactivity Disorder (ADHD). Since the mastering of handwriting is central for the further acquisition of other skills such as orthograph or syntax, an early diagnosis and handling of dysgraphia is thus essential for the academic success of children. In this paper, we investigated a large handwriting database composed of 36 individual symbols (26 isolated letters of the Latin alphabet written in cursive and the 10 digits) written by 545 children from 6,5 to 16 years old, among which 66 displayed dysgraphia (around 12\%). To better understand the dynamics of handwriting, mathematical models of nonpathological handwriting have been proposed, assuming oscillatory and fluid generation of strokes (Parsimonious Oscillatory Model of Handwriting [Andr\'e, 2014]). The purpose of this work is to study how such models behave when applied to children dysgraphic handwriting, and whether a lack of fit may help in the diagnosis, using a two-layer classification procedure with different compositions of classification algorithms.