Case-Based Training Using Similarity and Categorization from a Multiple Correspondence Analysis

Jeffery, N, Teather D, & Teather B,A

Abstract The research presented in this paper relates to the development and evaluation of visual cues for accessing and comparing medical images of patients with a particular disease (pathology). The methodology utilises a scatter plot constructed from Multiple Correspondence Analysis to provide an overview of the variability in appearance of cases of a pathology. We investigate the extent to which the MCA scatter plot / map can serve as a model of psychological similarity and typicality. It is postulated that psychological similarity is captured in the spatial relations of items in the plot. The MCA scatter plot aims to provide an intuitive way of conceptualising similarity relations in that two stimuli are similar psychologically if they appear close together in the similarity space. It is also posited that the psychological notion of the typicality of cases within a disease may be visualised as the distance of any case from the centre of this map. It should also be possible to provide information using these scatter plots relating to the relative positions of cases in overlapping pathologies to allow for easy identification of problem cases (i.e. ones that lie in the region of overlap) and to assist in the categorisation of new additions to the archive (i.e. if a case is more similar to a prototype of one disease than the other then it may well be categorised as being a member of the first disease). The approach adopted for developing this plot and the derivation of the associated measures of typicality and similarity are described. A series of evaluations designed to test the accuracy and validity of the information provided in the plot and to assess whether such a plot could provide a useful vehicle for assisting computer based training are also described.