A lot of the work into perceptual and observer performance issues in radiologic imaging (and medical imaging in general) began as a result of a series of studies [1, 2] done after WWII. These studies were designed to determine which of four roentgenographic and photofluorographic techniques was better for mass screening of chest images for tuberculosis. What seemed to be a rather easy and practical question to answer in a fairly straightforward investigation turned out to yield results that created more questions than answers. Inter-observer and intraobserver variabilities were so high that it could not be determined which imaging method yielded better diagnostic accuracy. Prior to this study, it was generally presumed that radiologists did not differ that much from each other in their diagnoses, and that if the same image was shown to the same radiologist at two points in time the diagnosis would not differ substantially. These studies suggested otherwise - radiologists were not as consistent as previously thought. One followup study  even tried to get radiologists to simply describe characteristics of radiographic shadows. Again there was wide inter-observer variation and moderately high intra-observer variation even on the seemingly straightforward description of lesion characteristics. This suggested that differences/errors in performance might lie in perceptual and cognitive factors rather than in technical factors such as bad technique or poor processing. Even today studies are being conducted that look at reader variability and ways to reduce it .
Online access to SPIE eBooks is limited to subscribing institutions.