Stand-alone colon computer-aided detection (CAD) delivered excellent performance for polyp detection in a large screening population with high sensitivity and an acceptable number of false-positive results, according to a study published in the September edition of Radiology.
Few studies have examined CT colonography (CTC) CAD in a screening population, a critical prerequisite “to allow confident generalization of results prior to potential widespread implementation of CTC screening,” wrote the study's lead author Edward M. Lawrence, BS, from the department of radiology at University of Wisconsin School of Medicine and Public Health in Madison.
The optimal CAD system identifies all relevant-sized polyps and keeps false-positives at a manageable minimum, the authors noted.
Lawrence and colleagues undertook the study “to evaluate the stand-alone CAD performance for detection of colorectal polyps measuring 6 mm or larger at CT colonography in a large asymptomatic screening cohort.” The study evaluated the use of CTC CAD with oral contrast tagging, which has become the preferred screening method.
The retrospective study population included 1,638 asymptomatic women (mean age, 56.7) and 1,408 asymptomatic men (mean age, 57.1) who underwent CTC between March 2006 and Dec. 2008.
Five abdominal radiologists with at least 500 cases of CTC experience read the studies using a primary 3D approach supplemented by secondary 2D detection. Three abdominal radiologists with two to seven years CTC experience served as the ground truth panel and at least one reviewed potential polyps considered for inclusion to confirm a true polyp designation, explained the authors.
Among the study cohort, 373 patients had at least one polyp measuring 6 mm or larger and 142 patients had at least one polyp measuring 10 mm or larger to be evaluated with CAD, reported Lawrence. Per-patient CAD sensitivity was 93.8 percent at the 6 mm threshold and 96.5 at the 10 mm threshold; and per-patient and per-lesion sensitivity was 100 percent for the 12 patients with 13 cancers.
CAD had greater sensitivity for polyps with nonflat morphologic features when compared with polyps with flat morphologic features at both 6 mm and 10 mm, wrote Lawrence. And its sensitivity for advanced neoplasia and cancer was 97 percent and 100 percent, respectively.
CAD missed lesions measuring 6 mm or larger in 60 cases on both views and missed 306 polyps on one view. Thirty-three percent of false negatives were submerged lesions.
The mean and median false-positive CAD rates were 4.7 and 3 marks per series. Nearly 20 percent of false-positive marks could be easily dismissed, according to researchers, and another 30 percent of the false positive marks indicated tagged stool.
The system also identified an additional 15 polyps 6 mm or larger missed on initial CTC interpretation by the expert readers. “Our study, in conjunction with prior studies of reader performance, supports the notion that CAD adds a layer of detection redundancy to help reduce the likelihood of missing relevant polyps when applied in the second-reader paradigm,” wrote Lawrence and colleagues.
The researchers identified the next frontier for CAD—moving CAD beyond detection into diagnosis--and suggested that the assignment of a priority score or ranking to each mark could limit reductions in specificity.
CTC CAD showed excellent polyp detection in a large screening population undergoing CTC with cathartic preparation and oral contrast tagging, and the technology provided data that may complement interpretation even among experienced readers, concluded Lawrence and colleagues.