Research will be undertaken with the long-term objective of improving the means by which the performance of imaging technologies is assessed. The increased complexity and the dynamic nature of the medical imaging field--in particular, the proliferation of CT scans and the emergence of magnetic resonance imaging--make especially apparent now the need for improved measures of assessment. Such improved measures can be expected to lead to more appropriate use of those technologies. The proposed project has two overall aims: first, the develop improved methods of assessing diagnostic accuracy; and second, to test and demonstrate these methods with three imaging technologies, as used with a single organ and disease--i.e. mammography, thermography, and diaphanography in the diagnosis of breast cancer. A methodology will be developed for defining the perceptual features of an image or set or images that are of diagnostic importance, and for determining the optimal combination of these features in a diagnostic judgment. This methodology will further include techniques for training image interpreters to use the features according to the optimal prescription as well as computer-based decision aids to reinforce the optimal behavior in routine practice. A fundamental idea is that measures of accuracy, which underlie measures of utility, should reflect the best, practically attainable, performance of the technology. The proposed methodology elicits information about perceptual features from experts in a technology and diagnostic problem by means of structured interviews and group discussions. It corroborates and refines the main features by mathematical techniques of multidimensional scaling and hierarchical clustering as based on similarity judgments of pairs of images. Discriminant analysis is used to establish a final set of features and their optimal weights. A computer classifier is developed to accept interpreters' ratings of the features and to issue a probability estimate for the diagnosis under consideration. Preliminary studies, based on mammographic images, indicate that this methodology can produce a very substantial gain in diagnostic accuracy. The product of this project will be a publication clearly describing the new methodology, so that it can be generally applied by clinical investigators in technology assessment. Further, the empirical comparison of the accuracies of three technologies, and the determination of their accuracy in various combinations, will be of immediate import for the medical community.
| Swets, J A (1992) The science of choosing the right decision threshold in high-stakes diagnostics. Am Psychol 47:522-32 |
| Seltzer, S E; Judy, P F (1987) Gaining insight: improved images enhance understanding. Radiographics 7:1221-40 |
