Artificial Intelligence-Based Classification and Prognostic Prediction in Small Bowel Crohn's Disease

Overview

This retrospective observational study aims to develop an artificial intelligence-based system for the precise classification and prognostic prediction of small bowel Crohn's disease. The study includes 437 patients with Crohn's disease who were hospitalized at Shanghai Tenth People's Hospital between January 1, 2020, and January 31, 2025. Clinical information, laboratory results, endoscopic findings, computed tomography enterography or magnetic resonance enterography images, and available pathological and molecular data will be collected from existing medical records. Artificial intelligence-based image segmentation and multimodal analysis will be used to identify and quantify intestinal lesions, strictures, mesenteric changes, fistulas, abscesses, and other disease characteristics. The study will examine whether these features can classify patients more accurately and predict clinical outcomes, including response to medical treatment, treatment failure or switching, and the need for surgery. The resulting system may support individualized assessment and clinical decision-making for patients with small bowel Crohn's disease.

Small bowel involvement is common in Crohn's disease and is associated with an increased risk of strictures, penetrating complications, and surgery. Because conventional ileocolonoscopy cannot fully assess most small bowel segments or transmural and extraintestinal abnormalities, computed tomography enterography (CTE) and magnetic resonance enterography (MRE) play important roles in evaluating small bowel Crohn's disease. However, interpretation of these images may vary among observers, and conventional imaging assessment may not fully quantify the complex intestinal and mesenteric features associated with treatment response and disease progression. This study will use an interactive artificial intelligence-based image segmentation method to identify and quantify small bowel lesions on existing CTE or MRE images. The imaging features of interest include the number and length of affected bowel segments, bowel wall thickness and enhancement, luminal narrowing, prestenotic dilatation, inflammatory or fibrotic characteristics of strictures, creeping fat, comb sign, internal fistulas, and intra-abdominal abscesses. Where available, pathological features from endoscopic biopsy or surgical specimens and molecular features, including tissue RNA expression and cytokine measurements, will also be analyzed. The imaging features will be integrated with clinical information, including demographic characteristics, disease duration, disease location and behavior, perianal disease, previous Crohn's disease-related surgery, clinical and endoscopic disease activity, nutritional status, laboratory findings, and treatment history. Multimodal data analysis will be used to establish a classification system for small bowel Crohn's disease and to develop a model for predicting subsequent clinical outcomes. Patients will be categorized according to their clinical course as having an effective response to medical treatment, treatment failure or recurrence requiring a treatment switch, or requiring Crohn's disease-related surgery. Univariable and multivariable analyses will be conducted to identify factors associated with these outcomes. The predictive performance of the resulting model will be evaluated using the area under the receiver operating characteristic curve, sensitivity, and specificity. The study is expected to provide an objective tool for disease classification, risk assessment, and individualized clinical decision-making in patients with small bowel Crohn's disease.