From macroscopic screening to microscopic characterization: AI and CLE in small bowel disorders

Background Small bowel diseases have long posed a diagnostic challenge in gastroenterology due to their deep anatomical location, tortuous configuration, and nonspecific clinical presentation. Capsule endoscopy (CE) and device-assisted enteroscopy (DAE) have improved small bowel visualization, yet CE reading burden and DAE procedural complexity remain core limitations. Artificial intelligence (AI)-assisted endoscopy and confocal laser endomicroscopy (CLE) have emerged as promising solutions. Methods This review narratively synthesizes the current evidence on AI-assisted capsule endoscopy (AI-CE) and AI-assisted device-assisted enteroscopy (AI-DAE) in the diagnosis of small bowel disorders. A critical appraisal approach was adopted to evaluate methodological quality, lesion type coverage, and standardization across studies. A literature search of PubMed was performed up to September 1, 2026, and reference lists of key reviews were screened to identify additional relevant articles. Results AI-CE has accumulated substantial evidence for Tier 1–2 (lesion detection and macroscopic classification), yet its capability does not extend to Tier 3 (microscopic characterization). AI applications in DAE are in early stages, facing challenges of data scarcity and cross-device generalization. CLE enables real-time “optical biopsy” but is constrained by the pharmacological limitation that fluorescein sodium does not stain cell nuclei and the absence of small bowel-specific diagnostic criteria. While AI + CLE has achieved notable progress in other parts of the gastrointestinal tract, the small bowel remains nearly blank. Conclusion The current core bottleneck lies not in technical feasibility but in the uneven distribution of research effort. This review proposes a four-step integrated framework—CE macroscopic screening → AI-CE reading → DAE-guided CLE targeted characterization → AI-CLE image analysis—to bridge the gap between macroscopic screening and microscopic characterization. This framework remains conceptual and requires prospective validation.