← Issue №20/ week of Nov 15, 2026/Endoscopy

An explainable multimodal model integrating post-ESD mucosal defect images and clinical features to predict delayed bleeding after colorectal endoscopic submucosal dissection.

From GI Signals issue №20: what this paper found, what it changes, and where it sits against the current standard of care, reviewed by Simon Mathews, MD.

Endoscopy prospective cohort · n=3,416 · Nov 10, 2026 · BMC Gastro · IF 2.5

An explainable multimodal model integrating post-ESD mucosal defect images and clinical features to predict delayed bleeding after colorectal endoscopic submucosal dissection.

New evidenceartificial intelligencecomputer-aided detectionESD
Clinical takeawayNo clinical action yet: a predictive model requiring external calibration and safety evaluation before guiding postoperative care.
What it foundPrismDB, a multimodal model integrating post-ESD mucosal defect images and clinical features, predicted delayed bleeding after colorectal ESD with AUC 0.804 (95% CI 0.769-0.837), sensitivity 64.73%, specificity 85.40%, and negative predictive value 96.50%.
ContextRefines current practice by incorporating mucosal defect images into delayed bleeding prediction, which existing tools omit.
No standard claimed for this paper

Every paper is compared to the standard governing its question. This one is not: either no standard in the corpus matches it, or the comparison did not hold up on review and was withdrawn rather than published unverified. Both are logged.

Zhao H … Chen J · BMC Gastroenterology · IF 2.5 · PubMed ↗Permalink
← Read the whole of issue №20 Every paper GI Signals surfaces gets a page like this one. All issues