ARTIFICIAL INTELLIGENCE TO QUANTIFY VILLOUS ATROPHY DURING CAPSULE ENTEROSCOPY

Aims To develop an image processing algorithm based on capsule enteroscopy (CE) CapsoCam (CapsoVision, Cupertino, CA, USA) lateral panoramic view to quantify villous atrophy.

Methods Frames from CapsoCam CEs were retrospectively obtained from celiac disease (CeD) patients and Controls (Co) referring to our Gastroenterology department from 2018 to 2020. Histology was assumed as reference standard in case of atrophy or not. Three 1 mm regions of interest (ROI) per frame were selected blindly by an expert gastroenterologist (LE) and then analysed for morphometric analysis using NIH Image J image-processing software and transformed in a numerical scale. Each ROI was then studied by using 3D a surface plot macro and isolines plots were obtained to identify the density of intestinal villi. All readers were blind to the histologic results.

Results A total of 306 ROI were acquired from 57 CE from CeD patients and 45 from Co. On a numerical scale, atrophic vs non-atrophic mucosa were represented by 7.03±1.54 vs 15.99±1.42. A sensitivity of 77% and a specificity of 79% in discriminating between atrophic (Marsh-Oberhuber > 3a) or non-atrophic (Marsh- Oberhuber < 2) mucosa with a cut-off of 14.10 (Youden Index) and an overall AUC of 0.805 (CI 95% 0.712-0.897) were obtained (Fig. 1).

Fig. 1

Conclusions Our process is able to discriminate the presence of villous atrophy in a definite set of patients. Further studies are needed to establish its validity in an external cohort.

Publication History

Article published online:
14 April 2022