Computer-generated surface and tone enhancements to distinguish neoplastic from non-neoplastic colon polyps less than 1 cm in diameter

Abstract

Computer-generated enhancements, which can highlight the surface and color of a colonic lesion, may be helpful to predict the histology; however, it remains unclear whether this technology can distinguish neoplastic from non-neoplastic colon polyps when the polyps are <1 cm without magnification. Images of colorectal polyps less than 1 cm in diameter were obtained from 54 patients who underwent non-magnified colonoscopy with surface enhancement (SE) and tone enhancement (TE). We calculated the sensitivity, specificity, and accuracy in the prediction of histology. Inter- and intra-observer consistency was evaluated by inviting four endoscopists to rate 45 static images. Overall sensitivity, specificity, and accuracy following the sequence of SE, TE colon, and TE pit pattern modes were 87.7% (95% confidence interval 81.3-94.1%), 84.1% (76.9-91.3%), and 86.1% (79.4-92.8%), respectively. For each modality, the results were 75.0% (68.7-81.3%), 82.7% (77.2-88.2%), and 77.2% (71.1-83.3%) for SE; 71.1% (64.5-77.7%), 78.8 (72.8-84.8), and 73.3% (66.8-79.8%) for TE colon mode; and 75.0% (68.7-81.3%), 80.8% (75.0-86.8%), and 76.7% (70.5-82.9%) for TE pit pattern mode. Their inter- and intra-observer agreements were all fair (κ range 0.522-0.568) and good (0.605-0.694), respectively. When the same rater evaluated the same lesion under different modalities, eight of 45 (18%) polyps yielded discordant interpretations, and the possibility of incorrect diagnoses was the highest with the TE colon mode. Computer-generated enhancements are satisfactory in predicting the histology of small colon polyps without the need for magnification. This advantage is mostly related to the pit pattern enhancement.