The diagnosis of visuo-spatial neglect through the computer-based analysis of hand-executed drawing tasks

This thesis documents the investigation of a technique for the computer-based assessment of visuo-spatial neglect for use within a population of stroke patients. Analysing the hand-drawn responses from a battery of neuropsychological tasks, a series of automated feature extraction routines have been implemented to accurately and consistently assess performance in a novel way, leading to a diagnostic indication of neglect severity. An investigation into the reliability of existing neglect assessment methods highlights the ambiguity in interpretation of marking criteria and the inaccuracy introduced due to human error in score calculation. The implemented feature extraction routines overcome these problems by algorithmically applying identical criteria to all test responses. The results of a clinically-based trial using the developed system show that significant performance differences can be identified both using conventional static features (the outcome of the test response) and novel dynamic time-based constructional features which establish previously unmeasured performance characteristics of neglect-based response while increasing the sensitivity of the detection of neglect. The correlation between the static features and existing assessments of neglect verify the ability of the computer-based battery to detect neglect. A feasibility study into the automated classification of feature measurements indicates the sensitivity of the individual tasks to detect neglect performance and shows that it is possible to classify responses by the analysis of the principal features extracted from test responses.