UEG's educational platform for gastroenterology education bringing recordings, courses, guidelines, and more together in one place.

UEG Week Berlin 2025 Recordings Gut Guide online

Rate for better recommendations!

Your feedback helps us show you more of what you like and less of what you dislike!

More of this!

We will show you more similar content.

Not for me!

We won't suggest this again.

Bookmark!

Save this content for later!

New to myUEG? Create an account! Log In with myUEG account

UEG Poster

Standard Poster

Share via Email Share on Facebook Share on X Share on LinkedIn Share on Bluesky

This content is part of Gutflix. Log in with your myUEG account, or create one free, to watch it.

Not sure what you can access? Learn more about account types.

EVALUATION OF THE DIAGNOSTIC ACCURACY OF CHATGPT-4 IN RECOGNITION AND CLASSIFICATION OF ESOPHAGEAL VARICES: A MULTICENTER STUDY

Giulia Navazzotti 1, ANDREA PASTA 1, Francesco Calabrese 1, Marcello Fabio Maida 2, Giovanni Marasco 3, Gianluca Esposito 4, Arnaldo Amato 5, Andrea Anderloni 6, Irene Maria Bambina Bergna 7, Elisabetta Dell'Unto 4, Alessandro Vitello 2, Tommaso Pessarelli 7, Sebastiano Ziola 8, DAVIDE SCALVINI 9, Elisa Marabotto 10, Giorgia Bodini 11, Edoardo Giovanni Giannini 1, Manuele Furnari 12

Affiliations

1 University of Genoa, Genoa, Italy

2 Department of Medicine and Surgery, University of Enna ‘Kore’, Caltanissetta, Italy

3 University of Bologna, Bologna, Italy

4 Sant’Andrea Hospital, Sapienza University of Rome, Italy, Rome, Italy

5 Gastroenterology Unit, Ospedale Valduce, Como, Italy, Como, Italy

6 Humanitas Research Hospital, Milan, Italy

7 Università degli studi di Milano, Milano, Italy

8 University of Genoa, IRCCS Ospedale Policlinico San Martino, Genoa, Italy, Genoa, Italy

9 University of Pavia, Pavia, Italy

10 OSP Policlinico San Martino, Genova, Italy

11 Policlinico San Martino, Genova, Italy

12 University of Genoa., Genoa, Italy

Summary

AI Generated

ChatGPT-4 achieved diagnostic accuracy comparable to junior and trainee endoscopists in detecting and classifying esophageal varices from endoscopic images, but was less effective at differentiating small from large varices and identifying red signs.

In a study of 279 good-quality esophagogastroduodenoscopy images from six Italian centers, ChatGPT-4 without prior task-specific training was compared to experienced, junior, and trainee endoscopists using experienced endoscopists' evaluations as ground truth.
ChatGPT-4 detected esophageal varices with AUC 0.77 (sensitivity 86.2%, specificity 68.4%), comparable to most endoscopists but lower than one junior endoscopist.
ChatGPT-4 distinguished small (F1) from large (F2-F3) varices with AUC 0.75 (sensitivity 67.6%, specificity 81.7%), performing worse than 3/4 junior and 3/5 trainee endoscopists.
ChatGPT-4 classified F3 varices with AUC 0.71 (sensitivity 52.8%, specificity 89.7%), comparable to all endoscopists and superior to one trainee, but identified red signs poorly (AUC 0.58, sensitivity 20.5%).
The authors conclude the model shows potential as a supportive tool for training and preliminary triage but cannot replace experienced clinical judgment and requires further refinement for routine clinical use.

Download PDF

Was this helpful?

Thanks for your feedback.

This summary was generated by an AI large language model based on the content transcript. It is for informational purposes only and should not be considered a substitute for clinical judgment. Always rely on your professional expertise and the full clinical context when making clinical decisions.

Abstract

Introduction

Large language models, such as ChatGPT-4, have significantly enhanced access to Artificial Intelligence (AI) tools. However, their role in clinical practice of gastroenterology remains uncertain¹. Accurate classification of esophageal varices is pivotal in managing patients with portal hypertension as it directly influences prognosis and treatment².

Aims & Methods

This study aimed to assess the potential of ChatGPT-4 as a diagnostic tool to accurately classify esophageal varices, and to compare its diagnostic accuracy and reliability to experienced and trainee endoscopists.
We collected photos taken during esophagogastroduodenoscopy performed at the Endoscopic Units of six tertiary Italian centers. The images were input into blank sheets of ChatGPT-4, which was asked to classify the varices by size, indicate the presence of red markings, and evaluate image quality in terms of insufflation, clarity and lightning. The same set of images and evaluation criteria was shown to six experienced endoscopists (E1, E2, E3, E4, E5, E6), to four junior endoscopists (J1, J2, J3, J4) and to five trainee endoscopists (T1, T2, T3, T4, T5). The evaluations from the experienced endoscopists were used as a Ground Truth. ChatGPT-4 had not undergone any prior task‐specific training for esophageal varices detection and classification.

Results

We collected 327 photos, of which 279 were considered to be of good quality by the experienced endoscopists. In detecting esophageal varices, ChatGPT-4 showed a lower AUC (0.77, 95% CI 0.66-0.88, Se 86.2%, Sp 68.4%) compared to one of the junior endoscopists (J2 AUC 0.92, 95% CI 0.87-0.98, Se 90.4%, Sp 94.4%; p<0.05) and comparable to other endoscopists (p>0.05). ChatGPT-4 showed a lower AUC (0.75, 95% CI 0.70-0.80, Se 67.6%, Sp 81.7%) than 3/4 junior endoscopist and 3/5 trainee endoscopists in distinguishing small (F1) from large (F2-F3) varices (p<0.05). For the classification of F3 varices, ChatGPT-4 achieved a higher AUC (0.71, 95% CI 0.63-0.80, Se 52.8%, Sp 89.7%) compared to one of the trainee endoscopists (T5 AUC 0.58, 95% CI 0.51-0.65; p<0.05) and comparable to that of all endoscopists (p>0.05). In identifying red signs, ChatGPT-4 (AUC 0.58, 95% CI 0.52-0.65, Se 20.5%, Sp 96.2%) performed less effectively than the endoscopists (p<0.05).

Conclusion

ChatGPT-4 achieved diagnostic accuracy comparable to junior and trainee endoscopists in detecting esophageal varices and classifying F3 varices, but it was overperformed in differentiating small (F1) from large varices (F2–F3) and in identifying red signs. While it cannot replace the clinical judgment of an experienced operator, the model shows potential as a supportive tool for training and preliminary triage of endoscopic images. Further refinement and dedicated training on endoscopic sign interpretation are needed to enhance its reliability and applicability in routine clinical practice.

References

1) Klang E, Sourosh A, Nadkarni GN, Sharif K, Lahat A. Evaluating the role of ChatGPT in gastroenterology: a comprehensive systematic review of applications, benefits, and limitations. Therap Adv Gastroenterol. 2023 Dec 25
2) Nguyen GC, Correia AJ, Thuluvath PJ. The impact of cirrhosis and portal hypertension on mortality following colorectal surgery: a nationwide, population- based study. Dis Colon Rectum. 2009 Aug

Conference

UEG Week Berlin 2025

Topics

Oesophagus

Submission format

Abstract

Session

AI in the upper GI tract (Posters)

Citation

United European Gastroenterology Journal 2025; 13 (Supplement 8)

Published

2025

More Like This:

EXOSOMAL CAGA INDUCES MACROPHAGE POLARIZATION AND FERROPTOSIS BY JAK1-2/STAT1 SIGNALING PATHWAY IN HELICOBACTER PYLORI-ASSOCIATED GASTRITIS

Jian Qi

MYELOID MAS REGULATES PKM2-MEDIATED SPI1 LACTYLATION TO PROGRAM THE METABOLIC-INFLAMMATORY SENESCENCE AXIS OF MONONUCLEAR PHAGOCYTE SYSTEM DURING THE METABOLIC DYSFUNCTION-ASSOCIATED STEATOTIC LIVER DISEASE (MASLD)

JIng Li

THE IMPLEMENTATION OF A NEW TYPE OF MERGING ANALYSIS REVEALS THE DYNAMIC INTERACTIONS BETWEEN BACTERIA AND FUNGI IN THE LOWER GASTROINTESTINAL TRACT

Alexander Link

ADVANCING INFLAMMATORY BOWEL DISEASE CARE IN AFRICA: OUTCOMES FROM A MULTIDISCIPLINARY EDUCATIONAL SYMPOSIUM

Akwi Asombang

Exploiting the immune response in GI cancer: Next chapter

Tamara Matysiak-Budnik

SYPHILITIC HEPATITIS: A RARE BUT REVERSIBLE CAUSE OF UNEXPLAINED HEPATITIS

Gagik Hakobyan

UEG Presentation

Share via Email Share on Facebook Share on X Share on LinkedIn Share on Bluesky

EXOSOMAL CAGA INDUCES MACROPHAGE POLARIZATION AND FERROPTOSIS BY JAK1-2/STAT1 SIGNALING PATHWAY IN HELICOBACTER PYLORI-ASSOCIATED GASTRITIS

This content is part of Gutflix. Log in with your myUEG account, or create one free, to watch it.

Not sure what you can access? Learn more about account types.