METHODS: Fourteen endoscopists were asked to review an atlas of reference clips and images of 5 criteria derived from expert consensus. They then proceeded to score 50 deidentified DSOC video clips based on the visualization of the following: (1) tortuous and dilated vessels, (2) irregular nodulations, (3) raised intraductal lesion, (4) irregular surface with or without ulcerations, and (5) friability. The endoscopists then diagnosed the clips as neoplastic or non-neoplastic. Intraclass correlation (ICC) analysis was done to evaluate interrater agreement for both criteria sets and final diagnosis.
RESULTS: Clips of 41 malignant lesions and 9 benign lesions were scored. Three out of 5 revised criteria had almost perfect agreement. ICC was almost perfect for presence of tortuous and dilated vessels (0.86), raised intraductal lesions (0.90), and presence of friability (0.83); substantial agreement for presence of irregular nodulations (0.71); and moderate agreement for presence of irregular surface with or without ulcerations (0.44). The diagnostic ICC was almost perfect for neoplastic diagnosis (0.90) and non-neoplastic (0.90). The overall diagnostic accuracy using revised criteria was 77%, ranging from 64% to 88%.
CONCLUSIONS: The IOA and accuracy rate of DSOC using the new Mendoza criteria shows a significant increase of 16% and 20% compared with previous criteria. The reference atlas helps with formal training and may improve diagnostic accuracy.