EndoStyle: Artificial Intelligence Image Transformation Tool for Colonoscopy (EndoStyle)

EndoStyle: Survey of Physicians on Endoscopic Image Style Transfer.

The study addresses the limitations of current AI systems in gastrointestinal endoscopy, which are tipically trained with data from a single type of endoscopy processor and have limited expert-annotated images. The investigators aim to develop and validate EndoStyle, an AI system that can generate images in the style of various processors from a single reference image. EndoStyle will be tested by showing endoscopists colonoscopy sequences with different image types to determine if they can distinguish AI-transformed images. Success would enhance AI training for diverse clinical setups.

Study Overview

• Status: Completed
• Conditions: Colon Cancer; Colon Rectal Cancer
• Intervention / Treatment: Device: EndoStyle

Detailed Description

The use of artificial intelligence (AI) in gastrointestinal endoscopy has become widespread. However, these systems are often only trained with data from a single type of endoscopy processor, which limits their applicability. In addition, the availability of images annotated by experts is limited, which affects data variability and thus the performance of AI systems.

The aim of this study is to develop a new artificial intelligence (AI) based system (EndoStyle) and validate its authenticity by means of a survey among physicians, which is able to generate multiple images in the style of different processor types (including Olympus, Pentax and Storz) from a single endoscopy reference image.

The investigators hypothesis is that the AI system is able to successfully change the image style of video processors, with the differences being imperceptible to the endoscopist's eye.

The methodology consists of showing to multiple endoscopists 28 colonoscopy sequences of 10 seconds duration each. In each one of them 3 images will be shown that can be all the possible combinations of images belonging to positive control, negative control, and Endostyle (intervention group). By performing a statistical comparison of the percentages of selected images for each group the investigators will be able to establish whether the participants are able to distinguish the images transformed by the AI.

If the results corroborate our hypothesis, our system could generate images that would allow a more customized training of AI systems for each clinical setup.

• Study Type: Observational
• Enrollment (Actual): 40

Contacts and Locations

Study Locations

• Germany
  • Bavaria
    • Würzburg, Bavaria, Germany, 97080
      • University Clinic of Würzburg

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: Yes

• Sampling Method: Non-Probability Sample

Study Population

The participants to the survey will be physicians with experience in colonoscopy.

Description

Inclusion Criteria:

• Physicians with experience in colonoscopy.

Study Plan

How is the study designed?

Number of groups / cohorts

3

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Positive control; The image shown to the participant belongs to the same colonoscopy shown in the 10 second video-sequence.
Negative control; The image shown to the participant belongs to the same colonoscopy shown in the 10 second video-sequence.
EndoStyle (intervention group); The image shown to the participant does not belong to the 10 second colonoscopy video-sequence but has been transformed with AI to simulate the style of the video.	Device: EndoStyle; The EndoStyle system is able to transform the style of the different video-processor images.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Perceptual Indistinguishability of AI-Transformed Endoscopic Images	Comparison of the accuracy for each study group.	5 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Time Taken for Image Identification	Time taken to assess the different tasks according to each study group.	5 months
Influence of regularly used processor	Influence of regularly used endscopy processor during clinical work for the perceptual indistinguishability of AI-transformed endoscopic images with defined processor classes	5 months
Influence of endoscopy experience	Influence of experience in endoscopy measured in lifetime performed colonscopies on perceptual indistinguishability of AI-transformed endoscopic images	5 months

Collaborators and Investigators

• Sponsor: Wuerzburg University Hospital
• Investigators: Principal Investigator: Alexander Hann, MD, University Hospital of Würzburg

Study record dates

Study Major Dates

• Study Start (Actual): August 15, 2024
• Primary Completion (Actual): August 22, 2025
• Study Completion (Actual): August 22, 2025

Study Registration Dates

• First Submitted: August 8, 2024
• First Submitted That Met QC Criteria: August 13, 2024
• First Posted (Actual): August 14, 2024

Study Record Updates

• Last Update Posted (Estimated): September 2, 2025
• Last Update Submitted That Met QC Criteria: August 25, 2025
• Last Verified: August 1, 2025

More Information

Terms related to this study

• Keywords: colonoscopy; Artificial intelligence
• Additional Relevant MeSH Terms: Neoplasms by Site; Neoplasms; Intestinal Diseases; Gastrointestinal Neoplasms; Digestive System Neoplasms; Digestive System Diseases; Gastrointestinal Diseases; Intestinal Neoplasms; Rectal Diseases; Colonic Diseases; Colorectal Neoplasms; Colonic Neoplasms
• Other Study ID Numbers: AI05

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: UNDECIDED

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No