CT colonography CAD can boost sensitivity for hard-to-find polyps

CT colonography computer-aided detection (CAD) increased reader sensitivity for 6 mm to 9 mm polyps, while adding 1.6 minutes to image review time, according to a study published online Nov. 14 in Radiology. The findings may quiet concerns about the utility of CT colonography CAD related to lesions in this range.

Six mm to 9 mm polyps have proven problematic for CT colonography reviewers. Although dedicated centers have achieved sensitivity in the 85 percent range, rates in multicenter trials have hovered between 59 and 78 percent. The U.S. Preventive Services Task Force cited the procedure’s varying sensitivity for 6mm to 9 mm polyps as a source of uncertainty regarding the health impact of CT colonography.

Daniele Regge, MD, from the radiology unit at the Institute for Cancer Research and Treatment in Candiolo, Italy, and colleagues, designed a multicenter prospective trial to evaluate the sensitivity and specificity of CAD as a second reader in identifying patients with 6-9mm polyps.

The researchers enrolled 618 patients at ten academic centers from July 2007 through May 2009.

Patients first underwent a CT colonography exam, which was interpreted without CAD by one of 17 radiologists. Median unassisted reporting time was 6.5 minutes. After completing an unassisted review and locking the results, the radiologists reviewed the images a second time with the assistance of the CAD algorithm. Each lesion was noted as CAD false-negative or a new lesion. Evaluation of the CAD-detected lesions required an additional 1.6 minutes of reading time.

Colonoscopy was performed at least three hours after CT colonography. Two radiologists who were aware of colonoscopy results reconciled undetected lesions by CT colonography CAD.

A total of 52 participants had 6 mm to 9 mm lesions. CT colonography sensitivity for these lesions was 65.4 percent without CAD, and 76.9 percent with CAD. Specificity for these lesions was 91.8 percent without CAD, and 90.9 percent with CAD, according to Regge et al. The researchers attributed the lack of impact on specificity to the ability of experienced readers to reject CAD false-positive marks.

The researchers noted that CAD assisted reading allowed the detection of 10 additional lesions, including nine adenomas in the 6 mm to 9 mm range. More than half of these were advanced adenomas, according to Regge and colleagues. “This is relevant in proposing CT colonography for the diagnosis of clinically important lesions, considering that the prevalence of advanced histologic features in intermediate-size lesions is not marginal,” they wrote.

Regge and colleagues observed that readers mistakenly dismissed a large number of CAD true-positive marks as false positive. They speculated a lack of confidence with the CAD software might have caused this response, and suggested potential to further improve CAD-assisted performance. If researchers can drive additional improvements in CAD performance sensitivity might reach the 90 percent mark, according to Regge et al.