Objectives: Endoscopic tri-modal imaging incorporates high-resolution white-light endoscopy (HR-WLE), narrow-band imaging (NBI), and autofluorescence imaging (AFI). Combining these advanced techniques may improve endoscopic differentiation between adenomas and non-neoplastic polyps. In this study, we aimed to assess the interobserver variability and accuracy of HR-WLE, NBI, and AFI for polyp differentiation and to evaluate the combined use of AFI and NBI.
Methods: First, still images of 50 polyps (22 adenomas; median 3 mm) were randomly displayed to three experienced and four non-experienced endoscopists. All HR-WLE and NBI images were scored for Kudo classification and AFI images for color. Second, the combined AFI and NBI images were assessed using a newly developed algorithm by six additional non-experienced endoscopists.
Results: The outcomes measured were interobserver agreement and diagnostic accuracy using histopathology as reference standard. Experienced endoscopists had better interobserver agreement for NBI (kappa=0.77) than for AFI (kappa=0.33), whereas non-experienced endoscopists had better agreement for AFI (kappa=0.58) than for NBI (kappa=0.33). The accuracies of HR-WLE, NBI, and AFI among experienced endoscopists were 65, 70, and 74, respectively. Figures among non-experienced endoscopists were 57, 63, and 77. The algorithm was associated with a significantly higher accuracy of 85% among all observers (P<0.023). These figures were confirmed in the second evaluation study.
Conclusions: Non-experienced endoscopists have better interobserver agreement and accuracy for AFI than for HR-WLE or NBI, indicating that AFI is easier to use for polyp differentiation in non-experienced setting. The newly developed algorithm, combining information of AFI and NBI together, had the highest accuracy and obtained equal results between experienced and non-experienced endoscopists.