The learning curve, accuracy, and interobserver agreement of endoscope-based confocal laser endomicroscopy for the differentiation of colorectal lesions

Abstract

Background: The endoscope-based confocal laser endomicroscopy (eCLE) system allows in vivo imaging of colorectal epithelium. Little is known about the learning curve for accurate interpretation of confocal images acquired with eCLE.

Objective: To determine the learning curve of eCLE, its diagnostic accuracy, and the intra- and interobserver agreement for the differentiation of colorectal lesions.

Design: Post hoc assessment of selected eCLE images.

Setting: Academic centers.

Patients: This study involved colonoscopic images from 47 patients.

Main outcome measurements: Learning curve of eCLE, accuracy, and intraobserver and interobserver agreement.

Methods: Three endoscopists received a short introduction to eCLE before evaluating 90 images. Observers assessed all eCLE images by using the Mainz classification. After each set of 30 images, the accuracy of each observer was assessed. The same procedure was repeated 6 months later by using the same set of images.

Limitations: Post hoc assessment.

Results: There were no significant changes between the first set of 30 images and the 2 consecutive sets (P = .08 and P = .180, respectively). The overall accuracy was 85.6%, 95.6%, and 92.2% for each observer. The κ values of the intraobserver agreement were 0.68, 0.84, and 0.77 for each observer. The κ value for interobserver agreement was 0.73 during the first and 0.72 during the second assessment.

Conclusions: Accurate post hoc interpretation of eCLE confocal images can be learned quickly. High diagnostic accuracy was achieved by all 3 observers during the initial stage of the assessment, which remained high thereafter. Intra- and interobserver agreement was substantial for all 3 observers. Future studies should focus on the real-time assessment of eCLE images.