Artificial Intelligence for the Detection of Small Bowel Lesions and Neoplasia: A Scoping Review

Artificial intelligence (AI) has emerged as a promising tool for improving the detection and characterization of small bowel (SB) lesions through endoscopic imaging. This scoping review mapped and synthesized the available evidence on the diagnostic performance of AI systems applied to capsule endoscopy (CE) and device-assisted enteroscopy for identifying SB lesions with potential malignant or premalignant relevance. The review was conducted according to the methodological frameworks of Arksey and O'Malley, Levac, and the Joanna Briggs Institute, and was reported in accordance with Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) recommendations. Searches were performed in PubMed, Scopus, and Embase, including original studies that evaluated AI systems on SB endoscopic images or videos and reported quantitative diagnostic metrics. A total of 13 studies were included, most of them retrospective in design. The majority originated from China and Japan, and capsule endoscopy was the most commonly used imaging modality. Automated lesion detection represented the principal application of AI, often combined with diagnostic classification tasks. Most models were based on convolutional neural networks, including architectures such as ResNet50, Single Shot Multibox Detector, YOLOv5, and nnU-Net. Across studies, diagnostic performance was consistently high, with sensitivities ranging from 81.2% to 98.6%, specificities from 88.6% to 99.8%, and area under the ROC curve values close to 1.0 in several reports. AI appears to improve the detection of SB lesions and significantly reduce endoscopic reading times, particularly in capsule endoscopy. However, the current evidence remains insufficient to support the use of AI as a screening tool for SB cancer, as most available studies are retrospective and lack robust prospective multicenter validation.