Description

This project intends to establish a multi-center cohort comprising 200 confocal small-bowel endoscopy cases plus 400 gastrointestinal control cases, equally split into two arms: an exploratory arm and a validation arm. Controls will be selected according to the research objective: for neoplastic lesions, normal-appearing small bowel, colonic adenomas, and early gastric adenocarcinoma; for non-neoplastic lesions, normal-appearing small bowel, colonic segments involved by IBD, and gastric segments involved by IBD.

Main inclusion: patients with suspected small-bowel disease who require endoscopy for lesion characterization, including but not limited to clinically suspected active or quiescent small-bowel Crohn's disease with or without complications, unclassified IBD such as CMUSE, hereditary small-bowel neoplastic syndromes (e.g. Peutz-Jeghers), and small-bowel lymphomas that need endoscopy plus histology.

During single- or double-balloon enteroscopy, areas endoscopically suspected and adjacent normal mucosa will be examined with confocal laser endomicroscopy (CLE). Conventional enteroscopic and CLE findings will be summarized; CLE variables will include, but are not limited to, mucosal-barrier assessment (epithelial gap score, cell-shedding score, Watson score), inflammatory activity based on crypt architecture, ENHANCE mucosal-healing score, and will be augmented by AI image-recognition systems (Yingnuo™ \& Xiaohua-Tanying™).

Using the exploratory arm, parameters that show statistically significant contribution to diagnosis by integrating endoscopic and comprehensive clinical diagnosis will be identified to build a new confocal enteroscopy-based model for small-bowel IBD, which will then be tested in the validation arm for sensitivity, specificity, AUC, etc. The auxiliary role of AI in lesion recognition will also be evaluated.

In parallel, quality-control (QC) indicators will be developed from the clinical-use cohort. Key indicators include: procedural-success rate (proportion of completed confocal enteroscopies), diagnostic accuracy (concordance with pathology or gold standard, including AI-induced drift), adverse-event rate, and image-quality score.

Contributions to QC of the entire workflow-pre-procedure (scope model, CLE mode setting, patient risk/anesthesia assessment, demographics), intra-procedure (insertion/withdrawal time, time to lesion, scope posture, completeness of CLE parameters, bowel preparation), and post-procedure (specimen handling, recovery, observation time, prophylactic antibiotics)-will be quantified by logistic regression to identify high-impact factors and build a scientific QC system. This system will be applied to the validation arm for feasibility and accuracy, and regular internal audits will be conducted for continuous improvement.