AI-enabled Endoscopic Prediction of Post-operative Recurrence in Crohn's Disease

Background:

Up to 70% of Crohn's disease (CD) patients will undergo a surgical resection in their lifetime. However, surgery is non-curative since 50% of patients have a recurrence, and about one-third need repeat surgery. The tools currently used to assess CD recurrences, such as faecal calprotectin (FCP), cross-sectional imaging (small bowel ultrasound, MRI scan) and conventional endoscopy, have a limited role in predicting early Post-Operative CD recurrence (POCr). Distinguishing inflammatory disease recurrence from post-surgical ischemic or suture-related alterations poses a significant challenge. Endoscopic Enhanced imaging (EEI) techniques like virtual electronic chromoendoscopy (VCE) and biopsy-like probe-based confocal laser endomicroscopy (pCLE) combined with artificial intelligence, can improve the detection of mucosal/vascular changes before major alterations become evident. VCE is available simply by switching a button. The pCLE probe will be passed through the endoscope channel like a biopsy forceps, enabling real-time, histology-like images of the intestine's lining and the gut barrier.

Study summary:

This is a multicentre prospective international observational study. This study aims to introduce a novel multidimensional approach to precision imaging, enabling the identification and stratification of high-risk patients who can potentially benefit from early treatments to halt the progression of CD.

The investigators will develop a novel endoscopic assessment system using EEI to evaluate early post-surgical changes and predict POCr. By integrating with immune marker profiling, clinical data, and AI assessment of EEI and histology, the investigators further plan to improve risk stratification and reduce interobserver variability. A detailed exploratory analysis will only be done in a cohort of patients in Ireland. The correlation between the new scoring system and established endoscopic and histologic scores, cross-sectional imaging, and non-invasive markers of inflammation will be evaluated. A multimodal machine learning model will be developed on EEI videos, histology, clinical data and immune molecular analysis to stratify patients' risk of early recurrence and long-term outcomes. The study will be divided into three phases:

• In the first phase, descriptor criteria for the assessment of post-operative Crohn's Disease will be defined. Gastroenterologists experienced in IBD endoscopy will review images and videos from an existing library showing the different grade of inflammation of the modified Rutgeerts score. These will be used for a stepwise discussion. A round table discussion using modified Delphi method will be conducted to ensure equal participation and identify the best component descriptors of endoscopic recurrence of CD. The components that achieved 100% consensus will be selected and the most important endoscopy predictive variables will be confirmed by using a machine learning technique. Finally, a new endoscopic score will be generated. Further, the investigators will first validate the new endoscopic score using the first 30 consecutive VCE and pCLE videos of POCr patients recruited in the multicenter PROSPER study. A structured consensus will be conducted with experts in Inflammatory Bowel Disease, endoscopy and histology to define the endoscopic findings of mucosal, vascular and intestinal barrier function. Subsequently, the investigators will prospectively validate the score in a large cohort of POCr patients enrolled in the PROSPER study and assess the diagnostic accuracy of the new scoring system in predicting post-surgical recurrence. Clinical information, blood, saliva, stool, and bowel specimens will be taken. Cross-sectional imaging (magnetic resonance imaging -MRI-, intestinal ultrasound -IUS-), endoscopy VCE and pCLE (in equipped centres) will be performed according to stool calprotectin 3 months after surgery. Patients will be followed up for 24 months and the results of the follow-up colonoscopy performed, as standard of care, within 18 months from the index colonoscopy, will be collected.

• In the second phase, the investigators will externally validate and reproduce the new scoring system by gastroenterologists using a computerized training module.

• In the third phase, an advanced computer-aided quantitative analysis of videos, images from VCE and pCLE, and digital histology will be developed and validated to enhance the prediction of POCr. Additionally, further machine learning models will be developed, utilizing comprehensive data from blood, stool, cross-sectional imaging, endoscopy, histology, immune markers, and OMICs to predict POCr and long-term outcomes.