Why do rads often dismiss polyps correctly identified by CAD during CT colonography?
 | |
| CT Colonoscopy |
According to the study's lead author Stuart A. Taylor, MD, from the department of specialist x-ray at University College Hospital, and colleagues, one of the challenges created through the use of CT colonography is ensuring that radiologists who interpret the results are properly trained and competent. The authors argued that CAD software can help reduce perceptual error and thereby improve radiologist performance. But the authors also noted that the sensitivity of standalone CAD is often greater than that of CAD assisted by a reader as radiologists sometimes will dismiss lesions correctly marked by CAD.
The purpose of the study was to retrospectively describe the characteristics of polyps incorrectly dismissed by radiologists despite appropriate CAD prompting during CT colonography.
A total of 111 polyps with a diameter of at least 6 mm and detected with CAD were collated from three previous studies, in which researchers investigated radiologist performance with and without CAD. Two new observers, each of whom had interpreted more than 300 validated CT colonography data sets, evaluated the 111 polyps according to predefined criteria, including polyp size, morphology, and location; data set quality; ease of visualization; tagging use and polyp coating; colonic curvature; CAD mark obscuration and number of false-positive findings.
The polyps detected with CAD were divided into two groups. The control group contained all 25 polyps detected by 100 percent of the readers in the original study prior to the use of CAD. The second was the group of missed polyps that contained the 86 polyps missed by at least one reader in the original study prior to the use of CAD. The missed polyps were further subdivided into two groups-- those with CAD gain (50 polyps for which use of CAD resulted in additional correct detection by at least one reader in the original study) and those (36) with no CAD gain.
Of the 111 polyps, 20 were in the rectum, 36 were in the sigmoid colon, 17 were in the descending colon, eight were in the transverse colon, 18 were in the ascending colon and 12 were in the cecum. Fifty-six polyps were 6–9 mm in diameter and 55 were 10 mm in diameter or larger.
According to the authors, polyps classified as easy to visualize were more numerous in the control group, and were significantly larger than the missed polyps. Polyps coated or submerged by tagged fluid were more likely to be missed than those that were uncovered.
In a finding that the authors described as “counterintuitive,” the odds of a missed polyp being correctly identified by at least one reader after CAD decreased with increasing polyp size. “The larger the polyp, the more likely it was to be misclassified by readers as a false–positive finding.”
The authors suggested that the most likely reason for this finding is that most radiologists detect large lesions without CAD and leave others that are difficult to characterize, such as those that have an irregular or flat shape. “These polyps remain unrecognized, despite a correct prompt by CAD,” they wrote.
The authors concluded that polyps 6 mm in diameter or larger detected with CAD, those of smaller size and those coated or submerged by tagged fluid are at increased risk of being missed by unassisted readers. They also concluded that even after correct CAD prompting, large lesions, particularly if they are irregular or flat, are at greater risk of being incorrectly dismissed.
The authors called for further studies to investigate the benefit of additional highlighting of these lesions with CAD.