On the Relationship between Physical Metrics and Numerical Observer Studies for the Evaluation of Image Reconstruction Algorithms
Physical metrics, such as the root mean squared distance, are simple to calculate, but we do not have a great confidence in their medical relevance. We have a greater confidence in numerical observer studies, but they are more complicated to carry out. Suppose that a claim is made that a particular physical metric yields similar results to a particular numerical observer study. How can we evaluate the validity of such a claim? In this paper we introduce the notion of “rank ordering similarity” to measure the relationship (closeness, similarity) between algorithm performance measures. We use this notion to study the relationship between a particular physical metric (relative clipped error) and a particular numerical observer method (ROC on threshold). We also investigate the role of contrast in this context.
Unable to display preview. Download preview PDF.
- Fukunaga, K. (1972). Introduction to statistical pattern recognition, Academic Press, New York.Google Scholar
- Green, D.M., and Swets, J.A. (1966). Signal detection theory and psychophysics, John Wiley & Sons, New York.Google Scholar
- Hanson, K.M. (1988). Method to evaluate image-recovery algorithms based on task performance, Proceedings of the Society of Photo-Optical Instrumentation Engineers 914, pp. 336–342.Google Scholar
- Herman, G.T. (1980). Image reconstruction from projections: The fundamentals of computerized tomography, Academic Press, New York.Google Scholar
- Yeung, K.T.D., and Herman, G.T. (1989). Objective measures to evaluate the performance of reconstruction algorithms, Medical Imaging III: Proceedings of the SPIE 1092, to appear.Google Scholar