Classification System And Interobserver Agreement

Background and aimsConfocal laser endomicroscopy(CLE), an emerging tool for in vivo imaging, has been validated in differentiating benign and neoplastic colorectal lesions. Three different classification systems for the differentiation between benign and neoplastic colorectal lesions using CLE all have a high sensitivity and specificity. This study aims to:(1) compare the accuracy of these3classification systems in diagnosing neoplastic colorectal lesions;(2) to analyze the interobserver agreement;(3) to compare the diagnostic accuracy between experienced CLE investigators and nonexperienced investigators in identifying neoplastic colorectal lesions,MethodsPatients with hyperplastic polyp or adenoma in the colorectum diagnosed by histopathology were recruited from Qilu Hospital of Shandong University. CLE images of50colorectal lesionswere obtained from39patients in vivo and stored in50electronic files separately. According to the same standard, four confocal images (two superficial images and2deeper images) and1conventional white-light endoscopic images were selected from each of the50lesions in this study by an experienced endomicroscopist. Selected images were evaluated by3experienced CLE investigators and three nonexperienced ones respectively. The50electronic files containing CLE images and white-light endoscopic images were assessed according to each of the3diagnostic system at a2-week intervalin a randomized order. Before evaluation, an educational set including basic principles of CLE, detailed description of the CLE classification system and20educational cases (not included in the evaluation) was carefully studied by all the participating investigators.Results The diagnostic accuracy was84%,81%and86.67%respectively using the Maiz, Sanduleanu and Qilu classifications; the sensitivity was85.12%、78.57%and85.12%, the specificity was82.58%,84.09%and88.64%. The diagnostic accuracy was not influenced by the expertise in CLE. Interobserver agreement was substantial for each classification system(Maiz, Sanduleanu and Qilu, κ=0.676,0.619,0.732respectively).ConclusionsThe3classification systems for the prediction of colorectal hyperplastic polyp or adenoma have a high accuracy, sensitivity and specificity. In addition, there is a short learning curve for nonexperienced CLE investigators identified in this study, which indicates a broader clinical application of these CLE diagnostic system for colorectal polyps.