Type of publication:
Oral presentation
Author(s):
Lei I.I.; Parisi I.; Bhandare A.; Perez F.P.; Lee T.; Shekhar C.; McStay M.; Anderson S.; Watson A.; Conlin A.; Badreldin R.; Malik K.; Jacob J.; Dixon A.; *Butterworth J.; Parsons N.; Robertson A.; Koulaouzidis A.; Arasaradnam R.
Citation:
Gut. Conference: BSG Annual Meeting, BSG LIVE 2025. Glasgow United Kingdom. 74(Supplement 1) (pp A48-A49), 2025. Date of Publication: 01 Jun 2025.
Abstract:
Background Artificial Intelligence (AI) assisted reading in Small Bowel Capsule Endoscopy (SBCE) has recently been shown to achieve comparable and potentially superior accuracy compared to standard clinician reading. In Colon Capsule Endoscopy (CCE), AI algorithms have also demonstrated some promising results.1 However, the extent of AI-assisted reading's advantage remains unclear, particularly regarding its performance across different polyp sizes, morphologies, locations, and non-polyp-related factors. Understanding this is essential for optimising AI performance and clinical integration. Objective(s) This CESCAIL sub-analysis evaluates the per-polyp diagnostic accuracy of AI-assisted versus standard clinician reads (pathways) and identifies key factors influencing AIassisted accuracy using AiSPEEDTM. Methods A total of 1,803 polyps from 673 patients were analysed at the per-polyp level to assess diagnostic accuracy in terms of sensitivity and PPV, as well as the factors influencing the improved accuracy of AI-assisted readings compared to standard clinician readings. Factors examined included polyp size, morphology, location, patient demographics (age and sex), bowel preparation quality, capsule excretion rates, comorbidities, medications, reading time, and video duration. Statistical methods included, McNemar's test, superiority and noninferiority analyses, Generalised Estimating Equations, and generalized linear models with interaction terms, were employed to identify key predictors of enhanced diagnostic accuracy in both AI-assisted and standard readings. Results AI-assisted reading demonstrated significantly higher sensitivity with clear superiority for smaller polyps (<10 mm) compared to larger ones (>=10 mm) (OR 2.27 vs 0.88, p<0.0001). While there was no observed difference in diagnostic accuracy between pathways for polyps >=10 mm, noninferiority was established. AI accuracy remained consistent between polyps measuring 6-9 mm and <=5 mm (p=0.64). The most notable improvement was observed with hyperplastic polyps (OR 5.4, p<0.0001), particularly in the rectal region (OR 5.7, p<0.0001). No significant differences were identified for pedunculated, subpedunculated, LST, or SSL polyps. Furthermore, AI-assisted readings were significantly more accurate for left-sided polyps compared to right-sided ones (OR 2.36 vs 1.66, p<0.0001), although AI-assisted reads outperformed standard reads in both locations. Conclusion This study highlights the strengths of AI-assisted reading, particularly for detecting smaller adenomas and hyperplastic polyps, with notable accuracy in the left colon. Next-generation AI should focus on distinguishing significant from diminutive polyps and enhancing polyp characterisation, especially for right-sided lesions.
DOI: 10.1136/gutjnl-2025-BSG.72
Link to full-text [NHS OpenAthens account required]
