For this study, only high-quality images that show clearly crypt or vascular architecture were selected. Four confocal images (two for superficial crypt structures and two for deeper vascular structures) click here and one white-light colonoscopic image that were selected from each polyp were stored in a separate folder. In total, 50 folders of images for 50 polyps were collected for prospective evaluation. Corresponding histologies of the 50 polyps were 27 adenomas and 23 non-adenomas lesions. Three different DVDs were created, each containing an educational set and a prospective evaluation set. The educational set contains a description of the study, one of the three
diagnostic systems, basic principles of CLE, and the 20 educational images not from the polyps in this study. And the prospective evaluation set contain 50 folders of images collected in 50 colorectal polyps as stated above. The 50 image folders for prospective evaluations were arranged in randomized orders in each DVD to avoid bias from video recognition. The DVDs were sent to the observers at 2-week intervals. Before starting the evaluation set, each assessor had to study the educational set carefully. Six endoscopists who were not involved in
the performance of the procedure and also did not participate in the images selecting procedure selleck products were chosen to participate in this study. They were assigned into two groups. Group one included three experienced endomicroscopists who had performed more than 300 CLE procedures. Group two included three non-experienced endoscopists who were unfamiliar with CLE but had at least 5 years’ expertise in performing conventional MCE colonoscopy. All observers were blinded to the histological and clinical data. Subsequently, the six assessors studied the instruction of the study and the educational set. After 2 h, they predicted the histology of each of the 50 colorectal polyps according to the principles of the corresponding diagnostic system. Thereafter, the image description and correct histopathology diagnosis were displayed. The prospective evaluation set included 50 files display in a randomized order. The observers could run the images as many
times as necessary until they got the confirmed prediction. Contrary to the educational set, there was no histology and correct images description feedback. The process was repeated every 2 weeks. The sensitivity and specificity of CLE images for the prediction of adenomas were calculated. Global accuracy was estimated based on the true positive proportion and true negative proportion. The differences in accuracy between non-experienced and experienced assessors for the prediction of adenomas were tested with the chi-squared test. P < 0.05 was considered to be statistically significant. Cohen’s κ coefficient was used to measure the degree of interobserver agreement. The κ value was estimated as average agreement across all pairs of observers.