EUVIMEDEUVIMED entdecken
Leitlinie

Guideline-Optimized Large-Language Models for Colonoscopy and Pathology Interpretation: A Three-Model Comparative Study Revealing Distinct Performance Behaviors

A Kahwagi, P Wehbe, A Rita, B Boustani, R Haddad, M Njeim, E Mahfouz, J Amara, K Honein, R Slim, C Yaghi

Endoscopy · 01.06.2026 · 58 (S 03 S 03) · S582-S583

DOI: 10.1055/s-0046-1821738 ↗

Kurz gefasst

Leitlinie in der Zeitschrift Endoscopy, erschienen am 01.06.2026. Verfasst von A Kahwagi und 10 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Aims

Large-language models (LLMs) are increasingly used to support colonoscopy report interpretation, but their diagnostic reliability and behavior before and after guideline optimization have not been systematically evaluated. This study compared the accuracy and completeness of ChatGPT, DeepSeek and Copilot when interpreting colonoscopy and corresponding anatomopathology reports, assessed the effect of structured evidence-based prompting, examined cost-related practical considerations and identified performance domains in which any model surpassed ChatGPT.

Methods

Ninety-one colonoscopy reports combined with their available anatomopathology results were processed by each LLM in two conditions: before and after receiving standardized guideline-based instructions. Each model was asked to interpret the full clinical sequence, integrating both endoscopic findings and pathology outcomes. The dataset was derived from real-world procedures performed at Hôtel-Dieu de France, Beirut, Lebanon, ensuring clinically authentic case representation. Two binary outcomes were recorded: accuracy, defined as correct interpretation of report and pathology correlation, and completeness, defined as inclusion of all essential reporting elements. Paired analyses used the McNemar test. Comparisons were conducted within each model pre- vs post-guidelines and between models at each timepoint.

Results

Across all three LLMs, the most frequent pre-guideline error was incorrect assignment of surveillance intervals, reflecting insufficient integration of evidence-based follow-up rules before structured prompting. ChatGPT demonstrated the largest improvement after guideline optimization, with accuracy increasing from 56.0% to 92.3% (p

Zitieren

APA

Kahwagi, A., Wehbe, P., Rita, A., Boustani, B., Haddad, R., Njeim, M., Mahfouz, E., Amara, J., Honein, K., Slim, R., & Yaghi, C. (2026). Guideline-Optimized Large-Language Models for Colonoscopy and Pathology Interpretation: A Three-Model Comparative Study Revealing Distinct Performance Behaviors. Endoscopy, 58(S 03 S 03), S582-S583. https://doi.org/10.1055/s-0046-1821738

Vancouver

Kahwagi A, Wehbe P, Rita A, Boustani B, Haddad R, Njeim M, et al. Guideline-Optimized Large-Language Models for Colonoscopy and Pathology Interpretation: A Three-Model Comparative Study Revealing Distinct Performance Behaviors. Endoscopy. 2026;58(S 03 S 03):S582-S583. doi:10.1055/s-0046-1821738

BibTeX
@article{Kahwagi2026,
  author = {Kahwagi, A and Wehbe, P and Rita, A and Boustani, B and Haddad, R and Njeim, M and Mahfouz, E and Amara, J and Honein, K and Slim, R and Yaghi, C},
  title = {Guideline-Optimized Large-Language Models for Colonoscopy and Pathology Interpretation: A Three-Model Comparative Study Revealing Distinct Performance Behaviors},
  journal = {Endoscopy},
  year = {2026},
  volume = {58},
  number = {S 03 S 03},
  pages = {S582--S583},
  doi = {10.1055/s-0046-1821738}
}
Kostenlos starten →