A modelled classification approach to the automatic analysis of hand drawn neuropsychological tests

The evaluation of a patient’s drawing abilities is widely used for the diagnosis of neuropsychological conditions such as visuo-spatial neglect. Conventionally an analysis of simple constructional tasks such as pencil and paperbased drawing tests are analysed by a qualified medical specialist. However, variability in the assessment of these drawing tasks caused by the subjective nature of marking interpretation can result in disagreement between assessors. Assessment of these drawing tasks is typically only made on the drawing outcome, which fails to account for important constructional information encapsulated within the task performance. A novel computer-based diagnostic system based on a neural network solution has been developed which enables the objective automatic algorithmic analysis of task responses, including these novel constructional data, forming an effective diagnostic tool for assisting clinicians. In this paper we describe the test system and present a detailed statistical analysis of diagnostic results obtained, in particular concentrating on the development of a feature selection and modelled classification structure for validation against an existing diagnostic standard.