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

UEG Week Berlin 2025 Recordings Gut Guide online What's new on Gutflix

Trending on Gutflix

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.

PREDICTION OF THE HILL CLASSIFICATION FOR HIATAL HERNIA USING ARTIFICIAL INTELLIGENCE

Ioannis Kafetzis 1, Karl Hermann Fuchs 1, Philipp Sodmann 1, Wolfram G. Zoller 2, Alexander Meining 1, Alexander Hann 1

Affiliations

1 University Hospital Würzburg, Würzburg, Germany

2 Katharinenhospital, Stuttgart, Germany

Summary

AI Generated

A novel active learning pipeline enables training of a deep learning model to classify hiatal hernia using the Hill classification from gastroscopic images, requiring 30% less expert annotation time than random image selection.

The active learning model was trained on 21,970 gastroscopic images from 14,885 examinations using eight consecutive steps covering about 10% of unlabeled training data.
The model achieved accuracy of 0.878 for classifying the four Hill types and 0.828 for diagnosing the presence versus absence of hiatal hernia on external test data.
Median annotation time per image was 3.62 seconds, and the active learning approach required at least 30% fewer annotations than random image selection to achieve similar performance.
The open-source active learning pipeline allows examiners to develop gastrointestinal endoscopy models without requiring programming skills.

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

Hiatal hernia (HH) is a common finding in endoscopic examinations that can cause significant symptoms, depending on its severity. The Hill classification assesses the degree of HH (type II to IV) by considering the width of the cardia and the hiatus on visual inspection during gastroscopy.

Aims & Methods

Our goal was to train a deep learning (DL) model that determines the Hill classification using a novel active learning (AL) pipeline. A total of 21,970 gastroscopic images representing 14,885 examinations from two endoscopy centers were retrospectively identified as the training dataset. A novel AL pipeline was developed to train the DL model. For comparison reasons, a second model was trained with randomly selected images from the same training dataset. An external, publicly available collection of HH images was annotated using the Hill classification and used as test data. Outcomes are the ability of the models to diagnose and classify HH and the time spent by the expert for annotation. The proposed AL pipeline is provided as Open-Source.

Results

The AL-based model was generated in eight consecutive steps, covering about 10% of the unlabeled training data. The model achieved mean accuracy, sensitivity, specificity, and F1-score of 0.878, 0.692, 0.896, and 0.67 respectively, when classifying the four different types of Hill. In diagnosing the existence of a HH (type I vs II-IV), the model achieved 0.828, 0.749, 0.902, 0.808 for the same metrics. The median annotation time per image was 3.62s (Q1-Q3; 3.14-4.69). Obtaining a model via the random image selection that performs similarly, required at least 30% more annotations and thus expert time commitment.

Conclusion

In this work we present a novel AL pipeline, enabling examiners to develop new models for gastrointestinal endoscopy without requiring programming skills. Using the proposed pipeline resulted in a well performing model in less time spent by the expert annotator, compared to traditional model training.

Disclosure

The authors have no conflicts of interest to disclose.

Conference

UEG Week Copenhagen 2023

Topics

Oesophagus

Submission format

Abstract

Session

PP 01 Oesophageal, gastric and duodenal disorders (Posters)

Citation

United European Gastroenterology Journal 2023; 11 (Supplement 8)

Published

2023

More Like This:

DISSECTING THE HINDRANCE OF HEDGEHOG PATHWAY IN INTRAHEPATIC CHOLANGIOCARCINOMA: A STEP TOWARDS TARGETED THERAPY

Guido Carpino

ENDOSCOPIC RETROGRADE CHOLANGIOPANCREATOGRAPHY BY CONVENTIONAL DUODENOSCOPE WITH COLONOSCOPY-ASSISTED PRE-PLACEMENT OF A LONG GUIDE TUBE IN PATIENTS WITH COMPLEX SURGICALLY ALTERED ANATOMY

Xiaofeng Zhang

LUMEN-APPOSING METAL STENTS WITH OR WITHOUT PIGTAIL STENT IN ENDOSCOPIC ULTRASOUND-GUIDED BILIARY DRAINAGE FOR MALIGNANT OBSTRUCTION: AN OPEN-LABEL MULTICENTRE RANDOMISED TRIAL (BAMPI TRIAL)

Cristian Tebé

Management of early-invasive (T1) colorectal cancer

James Hardwick

INFLAMMATORY BOWEL DISEASE IMPACT ON PATIENT’S SEXUAL DESIRE

María Dolores Martín-Arranz

FIRST-LINE PRESCRIPTION PATTERNS AND ERADICATION RATES IN IRELAND OVER A 10-YEAR PERIOD: DATA FROM THE EUROPEAN REGISTRY ON HELICOBACTER PYLORI MANAGEMENT (HP-EUREG)

Javier P. Gisbert

UEG Presentation

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

DISSECTING THE HINDRANCE OF HEDGEHOG PATHWAY IN INTRAHEPATIC CHOLANGIOCARCINOMA: A STEP TOWARDS TARGETED THERAPY

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.