In compressing a radiological image, the fundamental question is: Will the compressed image still be as diagnostically useful as the original? In the previous chapter, we presented several clinical studies, experimental protocols, and statistical analysis techniques. Taken together, these provide a methodology for answering this type of question, and they also provide the answer to the question in the context of particular images sets, compression algorithms, and diagnostic tasks. There remain, however, a number of questions that must be addressed in any study of this type. For example, was the experiment designed well enough? Is the answer different for different radiologists? How does diagnostic utility relate to other measures of image quality? In this chapter we present various statistical approaches to these broad questions. We first discuss statistical size and power, and learning effects, both of which speak to the question of whether the clinical experiment was well designed. Next, we present a comparison of judges, and we discuss how diagnostic utility can be related to other measures of image quality.

0 0

Post a comment