Computer assisted tutoring in radiology

The computer-based tutoring of visual concepts such as radiological images is a largely unexplored area. A review of the literature highlights limited use of computers as a means of radiological education, with the majority of systems being little more than hypermedia prototypes, on-line teaching files, and passive CD-Roms. Very few of the systems discussed in the research literature draw on the work from cognitive science and educational psychology for tutoring visual concepts. This thesis details and evaluates a method for indexing and retrieving images from large image databases via a cognitively informed graphical representation of visual concepts. This representation is derived using Multiple Correspondence Analysis from a statistical analysis of features present in the images. It is intended that this representation be used to assist in the computer based teaching of any discipline that requires the tutoring of visual concepts. The domain used as a test bed for this research is Magnetic Resonance Imaging in neuroradiology. Tutoring the interpretation of MR scans of the head represents a real practical problem due to the inherent complexity of the brain and variations in tissue contrast within image sequence and disease/pathology. The literature of concept categorisation and representation is critically reviewed to inform the design of a graphical representation of a concept (a pathology in the application domain). A specification for the design of an explicit representation of visual concepts is drawn from this literature review, together with methods of tutoring based around this model. A methodology for producing this representation is described, and implemented in the development of an overview plot for the concept of a disease of the brain, with associated statistical measures derived and operationalised for typicality and similarity of cases within a disease. These measures are precursors to the development of computer based tutoring strategies for image interpretation. The overview plots provide the user with a visual display of the variation in presentation of cases for a given disease. An important aspect of the research is the evaluation of the accuracy of this representation and its usability in browsing cases in a tutoring environment. A prototype tutoring environment has been designed based around case based teaching, and an image browser which offers the majority of the functionality available in the tutoring environment was used in the evaluations. A series of evaluations was undertaken to examine the accuracy and usability of the overview plot and its associated measures of typicality and similarity. The overview plot is shown to provide a good first approximation to the experts similarity estimates of cases within a disease and as such provides a representation of a disease that closely matches the experts view of the similarity of cases. A second series of evaluations involving novice, intermediate and expert radiologists using the computer based image browser demonstrated significant differences in how the experts, novices and intermediates interact with the image archive via the overview plot. These findings are consistent with those found in previous novice-expert studies involving chest x-ray interpretation. There were significant differences between the experts and novice/intermediates in terms of their placement of cases within the overview plot with the experts placement of cases closer to the statistical placements. The users reaction to the overview plot were favourable with subjects commenting on its intuitiveness and ease of use. Hence, it is concluded that the overview space could serve as a useful tool in assisting the training of novice and intermediate radiologists providing a vehicle for moving the less experienced subjects towards the experts' conception of a disease. The research offers a methodology for teaching based on the integration of principles from cognitive science and statistics, and as such, adds to the body of knowledge concerning computer based training of visual concepts. (author)