A multimodal (FACILE) classification for optical diagnosis of inflammatory bowel disease associated neoplasia
Research output: Contribution to journal › Article
Colleges, School and Institutes
BACKGROUND: Characterization of colonic lesions in inflammatory bowel disease (IBD) remains challenging. We developed an endoscopic classification of visual characteristics to identify colitis-associated neoplasia using multimodal advanced endoscopic imaging (Frankfurt Advanced Chromoendoscopic IBD LEsions [FACILE] classification).
METHODS: The study was conducted in three phases: 1) development - an expert panel defined endoscopic signs and predictors of dysplasia in IBD and, using multivariable logistic regression created the FACILE classification; 2) validation - using 60 IBD lesions from an image library, two assessments of diagnostic accuracy for neoplasia were performed and interobserver agreement between experts using FACILE was determined; 3) reproducibility - the reproducibility of the FACILE classification was tested in gastroenterologists, trainees, and junior doctors after completion of a training module.
RESULTS: The experts initially selected criteria such as morphology, color, surface, vessel architecture, signs of inflammation, and lesion border. Multivariable logistic regression confirmed that nonpolypoid lesion, irregular vessel architecture, irregular surface pattern, and signs of inflammation within the lesion were predictors of dysplasia. Area under the curve of this logistic model using a bootstrapped estimate was 0.76 (0.73 - 0.78). The training module resulted in improved accuracy and kappa agreement in all nonexperts, though in trainees and junior doctors the kappa agreement was still moderate and poor, respectively.
CONCLUSION: We developed, validated, and demonstrated reproducibility of a new endoscopic classification (FACILE) for the diagnosis of dysplasia in IBD using all imaging modalities. Flat shape, irregular surface and vascular patterns, and signs of inflammation predicted dysplasia. The diagnostic performance of all nonexpert participants improved after a training module.
|Number of pages||9|
|Early online date||12 Dec 2018|
|Publication status||Published - 2019|