Medical Imaging Techniques: A Review
Fundamental to the evaluation of a medical imaging technique is a reliable and valid measure of diagnostic accuracy. Ideally, an assessment of accuracy will be expressed in such terms that it can also serve as the basis for a valid description of that technique's utility, in terms of cost and benefit.
The relative operating characteristic (ROC) is the single analytic technique known to provide both the desired accuracy index and the desired basis for a description of utility. The reliability and validity of the ROC's accuracy index stem from the fact that this index reflects only the inherent discrimination factors of the diagnostic alternatives for an image reader. This index, specifically, is unaffected by decision factors, ie, by a reader's tendencies to favor or disfavor one diagnostic alternative, tendencies which will vary both between readers and within a reader. Further, the ROC analysis yields the particular balance or trade off among all image-based decision probabilities--including the probabilities of various correct and incorrect decisions--that will result from any particular decision factors that the investigator might want to specify. Therefore, in evaluating the utility of an imaging technique, by pursuing a decision flow diagram from the imaging stage to outcomes at some further stage, the investigator can begin with appropriate image based decision probabilities. The appropriate trade off among image-based decision probabilities will be the one that reflects the values, costs, and other event probabilities, that the investigator believes are inherent in that diagnostic/therapeutic technique, rather than the trade off that happens to be yielded by any test reader.
This report focuses on the use of the ROC in assessing technique accuracy. I begin with a brief review of the basic concepts of ROC analysis, emphasizing contrasts between the ROC's accuracy index and certain other indices commonly used in medical evaluation.