Related Experiment Video
Updated: Jun 15, 2025
05:49
Author Spotlight: Advancing CBCT and Digital Dental Image Integration with AI-Assisted Digitization
Published on: February 23, 2024
794
Advancing artificial intelligence applicability in endoscopy through source-agnostic camera signal extraction from
Ioannis Kafetzis1, Philipp Sodmann1, Robert Hüneburg2,3
1Department of Internal Medicine II, Interventional and Experimental Endoscopy (InExEn), University Hospital Würzburg, Würzburg, Germany.
Plos One
|June 11, 2025
Summary
We developed an AI method to extract camera signals from endoscopic images, improving AI model generalizability across diverse sources. A new dataset, EPIC, was also created to support this advancement in medical AI.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Computer Vision
Background:
- Artificial intelligence (AI) in endoscopy faces challenges due to variations in image appearance from different equipment.
- These variations hinder the generalizability of AI models, limiting their real-world application.
- Effective image processing is crucial for successful AI deployment in endoscopic procedures.
Purpose of the Study:
- To develop a source-agnostic AI method for extracting camera signals from raw endoscopic images.
- To create a diverse and standardized dataset of endoscopic images (EPIC) to address variations in image sources.
- To improve the generalizability of AI models in endoscopy by enabling them to process images from various devices.
Main Methods:
- An AI-based method was developed to extract camera signals from endoscopic images irrespective of the image source.
- A comprehensive dataset, Endoscopic Processor Image Collection (EPIC), was curated from 4 endoscopy centers, including data from 9 processors, 45 endoscopes, and 4 capsule endoscopy devices.
- The camera signal extraction method was evaluated on public datasets and the EPIC dataset, comparing performance metrics like Intersection over Union (IoU) and Hausdorff distance (HD) against a baseline.
Main Results:
- The AI method achieved a mean IoU of 0.97 and a significantly lower mean HD (21 pixels) compared to the baseline on public datasets.
- On the EPIC dataset, the method showed no significant difference in IoU but a significantly lower HD, demonstrating its effectiveness on diverse, standardized images.
- Both the AI method and the EPIC dataset have been made publicly available.
Conclusions:
- A novel AI-based method effectively segments endoscope camera signals in a source-agnostic manner, crucial for medical AI.
- This method serves as a preprocessing step, allowing AI models to utilize endoscopic images from any source without performance compromise.
- The publicly available EPIC dataset and the developed AI method contribute to advancing AI applications in endoscopy.
