BACKGROUND AND STUDY AIMS

In vivo optical diagnosis of small colorectal polyps has potential clinical and cost advantages, but requires accuracy and high interobserver agreement for clinically acceptability. We aimed to assess interobserver variability and diagnostic performance of endoscopic imaging modalities in characterizing small colonic polyps.

METHODS

High quality still images of 80 polyps < 1 cm were recorded using white-light endoscopy (WLE), autofluorescence imaging (AFI) and narrow-band imaging with and without magnification (NBI and NBImag). All images were assessed for quality, prediction of polyp histology, and vascular pattern intensity (with NBI) by nine experienced colonoscopists (four experts in advanced imaging) from five UK centers. Interobserver agreement (kappa statistic), sensitivity, specificity, and accuracy were calculated compared with histopathological findings.

RESULTS

Interobserver agreement for predicting polyp histology using NBImag was significantly better for experts (κ = 0.63, substantial) compared with nonexperts (κ = 0.30, fair; P < 0.001), and was moderate for all colonoscopists with WLE, AFI and NBI. Interobserver agreement for vascular pattern intensity using NBI was 0.69 (substantial) for experts and 0.57 (good) for nonexperts. NBImag had higher sensitivity than WLE (experts, 0.93 vs. 0.68, P < 0.001; nonexperts, 0.90 vs. 0.52, P < 0.001) and higher overall accuracy (experts, 0.76 vs. 0.64, P = 0.003; nonexperts 0.61 vs. 0.40, P < 0.001). AFI had worse accuracy than WLE for both expert colonoscopists (0.53 vs. 0.64, P = 0.02) and nonexperts (0.32 vs. 0.40, P = 0.04).

CONCLUSIONS

Of the imaging modalities tested, NBImag appeared to have the best overall accuracy and interobserver agreement, although not adequate for in vivo diagnosis. NBI and AFI did not have better sensitivity, specificity, or accuracy compared with WLE.