NLP-based classification of bowel preparation quality and its association with adenoma detection rate
Oct 7, 2025·
Besim Fazil Agargun
Abstract
Late-breaking oral presentation. A natural-language-processing model (TF-IDF + logistic regression) classified bowel-preparation quality in 11,374 free-text colonoscopy reports with 92% cross-validated accuracy; the derived labels were then linked to adenoma detection in 5,420 screening/diagnostic colonoscopies (ADR 20.4%).
Date
Oct 7, 2025 —
Event
Location
Berlin, Germany
Authors
Gastroenterology Fellow
Gastroenterology and hepatology fellow at Istanbul University, Istanbul Faculty of Medicine.
My clinical work centres on inflammatory bowel disease and diagnostic and therapeutic endoscopy;
my research combines real-world cohorts with natural-language processing and AI-assisted analytics —
including an NLP programme over more than 11,000 colonoscopy reports and the ELASTIBD transient
elastography study of liver fibrosis and steatosis in IBD. I am also a PhD candidate in clinical trials.