Introduction

In ulcerative colitis (UC) in clinical remission, Mayo Endoscopic Score (MES)1 results in clinical relapse compared to MES0. Recently, EndoBRAIN-UC, an Artificial Intelligence (AI)-assisted colonoscopy system with contact microscopy capability, has become commercially available. EndoBRAIN-UC predicts Geboes score (GS) ≥ 3.1 as an auxiliary diagnosis and determines the presence of histological inflammation (Active). We aim to evaluate the impact of the presence of histological inflammation (GS ≥ 3.1) on MES1 relapse in real clinical practice using EndoBRAIN-UC.

Aims & Methods

The study will include UC patients with colonoscopically confirmed MES1 during clinical remission (partial Mayo score [PMS] <2 and rectal bleeding score 0), who will undergo AI-assisted diagnosis of MES1 sites and will be compared in the AI-Active (GS≥3.1 equivalent) group and AI-Healing (GS<3.1 equivalent) group in terms of cumulative relapse rate over the following 2 years.
Biopsy was performed at the site of AI-assisted diagnostic decision and histological evaluation was also performed. Histologic remission was defined as GS <3.1, and clinical relapse was defined as PMS ≥2 or intensified treatment for UC (excluding topical treatments only).

Results

Fifty-five patients (29 male [52.7%], median disease duration 15.7 [1.5-40.8] years, median PMS 0 [0-0]) were enrolled: AI-Active group by AI judgment: 36 patients (65.4%), GS≥3.1 by biopsy pathology: 23 patients (41.8%). The AI-Active decision identified GS≥3.1 significantly superior to the AI-Healing decision, with a sensitivity of 82.6%, specificity of 46.9%, positive predictive value of 52.8%, and negative predictive value of 78.9% (p=0.04). No significant difference in cumulative relapse rate was observed in the interim analysis at 1 year between the AI-Active (3/33 patients) and AI-Healing (2/17 patients) groups (Log rank test: p=0.77). On the other hand, there was a trend toward cumulative relapse in GS≥3.1 (4/19) compared with GS<3.1 (1/31) (Log rank test: p=0.06).

Conclusion

The AI-Active decision was able to significantly determine GS≥3.1 and GS<3.1 with a sensitivity of 82.6%, but interim analysis showed that it was difficult to predict relapse in UC patients in clinical remission and with MES1. (UMIN000049858)