Related Experiment Video
Updated: Nov 4, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
735
Computer-aided detection of colorectal polyps using a newly generated deep convolutional neural network: from
Lukas Pfeifer1, Clemens Neufert1, Moritz Leppkes1
1Department of Internal Medicine 1, Division of Gastroenterology, Ludwig Demling Endoscopy Center of Excellence, Friedrich-Alexander-University, Erlangen-Nuernberg, Germany.
European Journal of Gastroenterology & Hepatology
|May 25, 2021
Summary
A new artificial intelligence system, a deep convolutional neural network (DCNN), significantly improved adenoma detection rate (ADR) in human colonoscopy trials. This AI tool enhances polyp detection, potentially reducing interoperator variability in colorectal cancer screening.
Area of Science:
- Gastroenterology
- Medical Imaging
- Artificial Intelligence in Medicine
Background:
- Artificial intelligence (AI) offers a promising approach to enhance endoscopist performance in detecting colorectal polyps.
- Current methods for polyp detection exhibit interoperator variability, highlighting the need for objective, automated solutions.
- Deep convolutional neural networks (DCNNs) are being explored for their potential in medical image analysis, including polyp identification.
Purpose of the Study:
- To evaluate a novel deep convolutional neural network (DCNN) for the automated detection of colorectal polyps.
- To assess the DCNN's performance in ex vivo testing and in a first-in-human clinical trial.
- To determine the impact of the DCNN on the adenoma detection rate (ADR) and interoperator variability.
Main Methods:
- A DCNN was trained using a large dataset of 116,529 colonoscopy images from 278 patients, with 788 polyps identified.
- Ex vivo performance was evaluated using 15,534 frames from 45 videos, with manual annotation serving as the gold standard.
- In vivo testing involved 42 patients undergoing routine colonoscopy, with the DCNN operating in real-time in a back-to-back approach.
Main Results:
- Ex vivo analysis showed the DCNN achieved 90% sensitivity and 80% specificity for polyp detection and localization, with an AUC of 0.92.
- In vivo, the DCNN significantly increased the polyp detection rate from 38% to 50% (P = 0.023) and ADR from 26% to 36% (P = 0.044).
- The DCNN identified 13 additional lesions, predominantly diminutive and flat polyps, including three sessile serrated adenomas.
Conclusions:
- The developed DCNN demonstrates high sensitivity for automated colorectal polyp detection in both ex vivo and in vivo settings.
- This AI tool has the potential to significantly improve polyp detection rates during colonoscopy procedures.
- The DCNN could serve as a valuable adjunct for endoscopists, enhancing the accuracy and consistency of colorectal cancer screening.
