Document resource
Introduction Capsule endoscopy (CE) is an established non-invasive tool for evaluating small bowel (SB) pathology, but prolonged reading times can lead to reader fatigue. The development of artificial intelligence within reading software has shown promise to overcome this. This study assessed the diagnostic performance of TOP100, a software selecting 100 images most likely to contain pathology, against standard human reading (SR) across clinical indications.Methods A retrospective single-centre cohort study was conducted at a tertiary referral centre, including all patients who underwent PillCam™SB3 CE between January 2024 to August 2025. Data collected included baseline demographics, comorbidities, previous investigations, CE indications and adequacy of bowel preparation. CE findings from SR were compared with those from a second reader, blinded to the original report, who reviewed only the TOP100 images. All readers had read >500 CE in their lifetime. Diagnostic performance of TOP100 was compared with SR for each CE finding, with a sub-analysis based on CE indication.Results A total of 1382 CE were included; 8 were excluded from analysis (4 unable to swallow the capsule, 4 lacking Top100 readings), leaving 1374 CE for analysis (56.6% female, median age 52 years, IQR 33-65). The most common indications were suspected or known inflammatory bowel disease (IBD) (51.5%) and iron-deficiency anaemia (25.0%). Bowel preparation was adequate in 89.4%, examinations were complete in 90.8%. Median SB transit time was 3h42m (IQR 2h29m-4h39m). Diagnostic yield was significantly higher with SR than TOP100 (36.0%vs27.5%, p<0.001). TOP100 showed moderate sensitivity but high specificity; 64.0% (61.5%-66.5%) and 99.0% (98.5%-99.5%) for active bleeding, 58.2% (55.6%-60.8%) and 97.5% (96.6%-98.3%) for angiodysplasia, respectively. In overt bleeding, TOP100 detected 51.9% of angiodysplasias and 72.7% of active bleeding reported by SR, with 61.0% (51.1%-70.8%) sensitivity and 94.3% (89.7%-99.0%) specificity for P2 lesions. 1 Similarly, overall sensitivity and specificity were 55.3% (52.6%-57.9%) and 95.9% (94.9%-97.0%) for ulcers, 43.1% (40.5%-45.7%) and 89.0% (87.4%-90.7%) for erosions, respectively. In suspected or known IBD, diagnostic performance of TOP100 for ulcers was similar, while sensitivity for erosions improved by 9.4%.Conclusions SR remains the reference standard for reading and reporting CE, however TOP100 may be a useful adjunct for rapid video preview to support CE interpretation and case prioritisation, especially in high volume centres. The diagnostic performance of TOP100 is indication dependent, with improved performance noted in bleeding and IBD.Reference Saurin JC, Delvaux M, Gaudin JL, et al. Diagnostic value of endoscopic capsule in patients with obscure digestive bleeding: blinded comparison with video push-enteroscopy. Endoscopy 2003;35:576–84.