Scleral image analysis via artificial intelligence to evaluate inflammatory status in inflammatory bowel disease.

Abstract

Purpose: Among the extraintestinal ocular manifestations of patients with inflammatory bowel disease (IBD), episcleritis is associated with the inflammatory activity of IBD. The use of artificial intelligence to evaluate scleral images may help establish a non-invasive method for assessing the degree of inflammation. This study aims to provide a new approach for evaluating IBD inflammation by analyzing scleral images.

Methods: We collected scleral images from patients with IBD across different inflammatory states and from controls. Then, we analyzed the images using the MASK R-CNN and HRNet method to build models for assessing the degree of inflammation.

Results: A total of 942 scleral images were collected and divided into 5 groups: the control group, inflammatory Crohn's disease (ICD) group, CD in remission (RCD) group, inflammatory ulcerative colitis (IUC) group, and UC in remission (RUC) group. In the training set, the accuracy and F1-score for discriminating between IBD and non-IBD controls, inflammatory IBD and IBD in remission, ICD and RCD, IUC and RUC were 99.02%, 97.99%, and 98.44%, 98.34%, and 99.04%, 98.98%, and 100%, 100%, respectively. In the test set, the accuracy, F1-score and the Area Under the Curve (AUC) were 95.59%, 92.5%, 92.9%, and 85.1%, 81.2%, 85.8%, and 74.2%, 72.5%, 77.9%, and 92.9%, 91.1%, 91.3%, respectively. The inflammatory ulcerative colitis group exhibited the most prominent scleral congestion, which was correlated with disease severity.

Conclusion: In this small-sample exploratory study, we propose that the evaluation of ocular manifestations in patients with IBD may serve as a non-invasive method for assessing inflammatory activity, especially in patients with UC.