A protocol that employs initial CAD review of CT colonography data followed by a second read by the radiologist delivers diagnostic performance similar to unassisted interpretation by a single radiologist, while boosting efficiency, according to a study published online April 29 in Radiology.
Dissemination of CT colonography has been hampered by several obstacles, including high interobserver variability, a time-consuming image review process and lack of cost-effectiveness. Researchers have suggested using CAD as a first reader to overcome some of these barriers. However, a first-reader CAD model followed by a 2D review may offer improved detection, according to Gabriella Iussich, MD, from the Institute for Cancer Research and Treatment in Turin, Italy, and colleagues.
To test this hypothesis, Iussich and colleagues devised a study to compare diagnostic performance and time efficiency of double-reading first-reader CAD followed by radiologist review with an unassisted image review.
The dataset included 98 negative CT colonography studies and 57 cases with positive results. Three radiologists read the studies either in the unassisted read mode or double-reading CAD mode, searching for polyps 6 mm or larger. Six weeks later, radiologists were presented with the same data in the opposite reading mode.
The mean rate of false positive results per patient was 19. CAD delivered sensitivity of 90 percent for lesions 6 mm or larger. Radiologists’ sensitivity and specificity at the 6 mm threshold were 74 percent and 93 percent, respectively, in the unassisted review. These rates were 77 percent and 90 percent for the double-reading CAD mode.
Image review times were similar for the unassisted and CAD reviewing times at 243 and 239 seconds, respectively. Mean review of the CAD prompts was 186 seconds; 2D review added 53 seconds. CAD review was faster when the number of CAD marks was 20 or fewer.
A total of 23 false-positive findings occurred in 21 patients in unassisted interpretation, and 31 false positives occurred in 22 exams in the double-reading CAD mode.
“According to our study, per-patient sensitivity of unassisted interpretation and that of [double reading CAD] interpretation were similar, and all 10 masses were recognized by using both modalities,” wrote Iussich et al.
In addition, the three readers missed “a substantial number of polyps” at both reading models, according to the researchers. More than 70 percent of these were prompted by CAD.
Given the similar diagnostic performance of both reading models and the improved efficiency of double-reading CAD, particularly with fewer CAD marks, the researchers concluded, “It may be an attractive reading strategy in a screening setting where the prevalence of disease is expected to be low and cost-effectiveness is an issue.”