Quantitative assessment of mucosal architecture using computer-based analysis of confocal laser endomicroscopy in inflammatory bowel diseases

Background and Aims: Confocal laser endomicroscopy (CLE) might discriminate mucosal lesions between Crohn's disease (CD) and ulcerative colitis (UC). However, the analysis of CLE images requires timeconsuming methods, a long training time, and potential impediments, such as significant interobserver variability. Therefore, we developed a computer-based method to analyze mucosal architecture from CLE images and discriminate between healthy subjects and patients with inflammatory bowel disease (IBD) as well as between UC and CD patients. Methods: We retrospectively screened patients who had undergone CLE either for an evaluation of IBD in remission or for colorectal cancer screening (control subjects) between 2009 and 2016. We assessed 14 morphologic and functional parameters in each CLE recording from 23 CD patients, 27 UC patients, and 9 control patients. Next, we constructed 2 scores, 1 for the IBD diagnosis and 1 for the differential diagnosis between UC and CD. Results: In IBD patients, the mean intercrypt distance, wall thickness, and fluorescein leakage through the colonic mucosa were significantly increased compared with control patients by 155%, 188%, and 297%, respectively (P<.05). In UC patients, the same parameters were significantly increased by 109%, 117%, and 174%, respectively (P <.05), compared with CD patients. IBD diagnosis had 100% (95% CI, 93%; 100%) sensitivity and 100% (95% CI, 66%; 100%) specificity. IBD differential diagnosis provided discrimination of UC from CD patients with 92% (95% CI, 75%; 99%) sensitivity and 91% (95% CI, 72%; 99%) specificity. Conclusions: Confirming these results using prospective validation cohorts can substantiate that computer-based analysis of CLE images may provide new biomarkers for the diagnosis and characterization of IBD.