EUVIMEDEUVIMED entdecken
CrossrefInfokarte

The development and validation of artificial intelligence model to assess the degree of mucosal cleanliness of the upper gastrointestinal tract

Marcin Romańczyk, Sravanthi Parasa, Prateek Sharma, Krzysztof Budzyń, Magdalena Lesińska, Mikołaj Machaj, Tomasz Romańczyk, Steven Hicks, Michael Riegler

Diseases of the Esophagus · 2026-11-02 · 39 (6)

DOI: 10.1093/dote/doag130 ↗
AbstractBibliografische Angaben

Abstract

Abstract Mucosal cleanliness of the upper gastrointestinal (UGI) tract is one of the important quality indicators of esophagogastroduodenoscopy (EGD). The aim of the current study was to create and validate an artificial intelligence (AI)-based system to assess cleanliness during EGD. A total of 520 EGD video samples of the esophagus, stomach, and duodenum were collected between August 2021 and February 2025 at five endoscopic centers and annotated by two endoscopists using the previously validated PEACE scale (from 0—completely obscured, to 3—perfect cleanliness). The dataset was divided into training (332 videos), validation (84 videos), and test (104 videos) datasets. The RegNetY-3.2GF deep learning model architecture was used. The accuracy of videos testing for 12 classes ranges between 90.38% and 100%. The macro recall, precision, F1 score, and Matthews correlation coefficient (MCC) were 0.669, 0.697, 0.668, and 0.684 respectively. The precision, recall, and F1 score ranged between 0.341 and 0.880, 0.303 and 0.885, and 0.321 and 0.883, respectively. For distinguishing between inadequate (scores 0–1) and adequate (scores 2–3) cleanliness, the algorithm achieved a recall of 0.927, precision of 0.864, and a F1 score of 0.895, with a MCC of 0.77. All reported metrics were computed exclusively on the held-out test set. The current AI model demonstrated a high accuracy for both frames and video assessment of mucosal cleanliness along with appropriate identification of the UGI segments. This model allows for real-time use studies to assess this important EGD quality measure.

Wissen, das zu Ihrer Arbeit passt.

Vom kostenlosen Einstieg über die persönliche Recherche bis zur vollständigen Research Suite.

Tarife ansehen →

Sind alle Verlagsvolltexte enthalten?

Nein. Bibliografische Nachweise, Abstracts und Volltexte sind unterschiedliche Inhalte. Ob ein Volltext verfügbar ist, hängt von der Quelle, Open Access und Ihren vorhandenen Lizenzen ab. Ein EUVIMED-Tarif ersetzt keine zusätzlichen Verlagsrechte.

Tarife ansehen →