Automated extraction of image segments from clinically diagnostic hand-drawn geometric shapes

Simple geometric shape drawing tasks are commonly used to diagnose and monitor patient performance for a range of clinical and neuropsychological conditions. Assessment relies upon observing the presence of components within a drawn image. The application of assessment criteria has been shown to vary amongst trained raters. An algorithm is presented to automatically extract the components from the static image of shape drawing responses. Specifically, images taken from a group of patients with visuo-spatial neglect and control subjects show the accurate identification of horizontal, vertical and diagonal components. Examples of performance metrics based on the features extracted from the component analysis show clear differences between neglect and control responses which are able to detect differences in performance more sensitive to the standard number of component assessment.