Computer-aided detection (CAD) systems have demonstrated their efficacy in aiding radiologists’ interpretations of diagnostic imaging exams. Mammography CAD, which was greeted with skepticism and suspicion by some radiologists at its introduction, has since enjoyed a groundswell of support. Lung CAD, for both x-ray and CT exams, is also gaining acceptance in the clinical mainstream. Colon CAD, developed to improve the sensitivity and consistency of polyp identification in CT colonography (CTC) exams, is showing promise of being the next application of the technology to find a place in radiological practice.
“To ensure the success of CAD in the clinical realm, it is important that CAD have similar sensitivities in varied patient populations,” wrote the authors of a multi-institution study testing colon CAD performance that was published in this month’s American Journal of Roentgenology.
Researchers from the diagnostic radiology department at the National Institutes of Health Clinical Center in Bethesda, Md., and the department of radiology at the University of Wisconsin in Madison, Wis., conducted an external validation trial of a CAD system (V3D Colon, Viatronix) on patients with positive CTC findings.
A group of 182 consecutively registered screening patients who had positive CTC findings of at least one polyp 6 mm in diameter or larger comprised the initial population for the external validation trial of the system. Of this cohort, 104 elected to undergo optical colonoscopy and polypectomy with pathologic analysis based on their CTC findings. This group comprised the study group for the trial, because findings in these patients were confirmed or refuted.
The study focused on the detection of adenomatous polyps due to their malignant potential. Each polyp 6 mm or larger seen with optical colonoscopy was located on 2D and 3D prone and supine virtual colonoscopic images by a radiologist using the V3D Colon software and formed the reference standard for the study.
“Matching of CAD of polyps with the reference standard was done in a completely automated manner without user interaction,” the authors wrote. “If any voxel within a polyp candidate matched the voxels within a traced reference standard polyp, the polyp candidate was labeled a true-positive finding; otherwise, it was labeled a false-positive finding.”
The CAD system had per-polyp sensitivities of 91.5 percent for adenomas 10 mm or larger and 82.1 percent for adenomas 6-9 mm, the researchers reported. The per-patient sensitivities were 97.6 percent for patients with adenomas 10 mm or larger and 82.4 percent for patients with adenomas 6-9 mm.
The scientists noted that the common reasons for CAD misses (false-negative findings) were the presence of adherent contrast medium, flat adenomas, and adenomas located on or adjacent to normal colonic folds. They observed that false-positive findings also were characterized by the same criteria; knowledge of which can help improve future iterations of CTC CAD systems as well as guide the use of the applications in a clinical setting.
“We found that the performance of a CAD system in a polyp-enriched cohort from a screening population at a medical center and geographic locale other than those used for development of the system was similar to (for ≥ 10 mm polyps) or significantly better than (for 6–9 mm polyps) the performance of the system during development of the software,” the authors stated.
They believe that the results of the external validation trial indicate that the performance of the colon CAD system is not idiosyncratic to the data used during the training and testing phase of software application development, and that it will probably perform well in broad clinical use in a screening population at average risk.