Deep learning meets small-bowel capsule endoscopy: A step toward faster and more consistent diagnosis of obscure gastrointestinal bleeding

Small-bowel capsule endoscopy (SBCE) is the first-line diagnostic tool for obscure gastrointestinal bleeding but is limited by labor-intensive review, reader dependency, and interobserver variability. In this issue, Kwon <i>et al</i> present a convolutional neural network-based system capable of both gastrointestinal segment localization and multi-lesion detection, erosions/ulcers, angiodysplasia, and bleeding, using full-length SBCE videos. The model achieved > 97% localization accuracy and approximately 99% lesion detection accuracy in internal testing, and in external validation reduced reading time from nearly one hour to nine minutes without compromising detection performance. Methodological strengths include a two-step detection-classification pipeline, temporal smoothing, and ensemble learning, closely simulating human reading workflow. Compared with previous artificial intelligence (AI)-SBCE studies, this integrated "localization + detection" approach represents a step toward clinically deployable AI assistance. However, generalizability remains limited by single-platform training, a focus on three lesion types, and modest external validation size. Wider lesion coverage, multi-platform validation, and real-time integration will be essential for broader adoption. This study underscores the potential of AI to improve efficiency and consistency in SBCE interpretation and marks meaningful progress toward its routine clinical use.