Statistical Support for Uncertainty in Radiological Diagnosis

Radiological interpretation and diagnosis involves the comparison and classification of complex medical images and is typical of the categorisation tasks that have been the subject of observational studies in Cognitive Science. This paper considers the affinity between statistical modelling and theories of categorisation for naturally occurring categories. Statistical based measures of similarity and typicality with a probabilistic interpretation are derived. The utilisation of these measures in the support of diagnosis under uncertainty via interactive overview plots is described. The application of the methodology to magnetic resonance imaging of the head is considered. The methods detailed have application to other fields involving archiving and retrieving of image data.

Statistical Modelling and Structured Image Description for Intelligent Tutoring in MR Imaging of the Head

A collaborative research project between the Medical Systems Research Group, De Montfort University Leicester, and the School of Cognitive and Computing Sciences, University of Sussex, is examining the feasibility of utilising statistical-based principles and a structured image description language for tutoring about radiological image interpretation and diagnosis. This paper describes the construction of a sophisticated intelligent tutoring system (ITS) for the interpretation of magnetic resonance images in the diagnosis of cerebral disease.

The outcome of the project will be a tutoring system and diagnostic aid for MR images which can be evaluated in a realistic educational setting. An evaluation of the system will show whether computer-assisted tutoring can be effective in supplementing conventional teaching, by teaching a systematic approach to image interpretation and by providing exposure to a large archive of annotated images.

Structured computer-based training in the interpretation of neuroradiological images

Computer-based systems may be able to address a recognised need throughout the medical profession for a more structured approach to training. We describe a combined training system for neuroradiology, the MR Tutor that differs from previous approaches to computer-assisted training in radiology in that it provides case-based tuition whereby the system and user communicate in terms of a well-founded Image Description Language. The system implements a novel method of visualisation and interaction with a library of fully described cases utilising statistical models of similarity, typicality and disease categorisation of cases. We describe the rationale, knowledge representation and design of the system, and provide a formative evaluation of its usability and effectiveness.

Statistical support for uncertainty in radiological diagnosis

Radiological interpretation and diagnosis involves the comparison and classification of complex medical images and is typical of the categorisation tasks that have been the subject of observational studies in Cognitive Science. This paper considers the affinity between statistical modelling and theories of categorisation for naturally occurring categories. Statistical based measures of similarity and typicality with a probabilistic interpretation are derived. The utilisation of these measures in the support of diagnosis under uncertainty via interactive overview plots is described. The application of the methodology to magnetic resonance imaging of the head is considered. The methods detailed have application to other fields involving archiving and retrieving of image data.

The use of a computerized brain atlas to support knowledge-based training in radiology

Trainers of radiologists face the particular challenges of teaching normal and abnormal appearance for a variety of imaging modalities, providing access to a large appropriately-indexed case library, and teaching a consistent approach to the reporting of cases. The computer has the potential to address these issues, to supplement conventional teaching of radiology by providing case-based tutoring and diagnostic support based on a large library of images of normal and abnormal anatomy, described in a consistent terminology. The paper presents a new approach to computer-based training in radiology that combines a knowledge-based tutor with an on-line medical atlas. It describes two existing computer systems, the MR Tutor and ATLAS, and discusses the medical, computational, epistemic, and pedagogic issues involved in developing a combined Atlas-Tutor. Integrating an atlas with a training system could significantly improve the teaching and support offered, but practical difficulties include the need to merge knowledge representations and to incorporate techniques for registering atlas plates on images that exhibit abnormalities. The paper addresses these problems, and concludes by indicating how the Atlas-Tutor might be employed in practical radiology training.