Polyp Segmentation in Colonoscopy Images Using Fully Convolutional Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colorectal cancer is a leading cause of cancer-related mortality, particularly in men.
• Colon polyps are primary precursors to colorectal cancer; early detection via colonoscopy is crucial for effective treatment.
• Accurate polyp segmentation in colonoscopy videos is challenging due to variations in polyp size and shape.

Purpose of the Study:

• To propose a novel polyp segmentation method utilizing convolutional neural networks for colonoscopy videos.
• To enhance the accuracy of polyp detection and segmentation in endoscopic imaging.
• To improve early diagnosis of colorectal cancer through advanced video analysis.

Main Methods:

• Development of a polyp segmentation method based on convolutional neural networks.
• Implementation of a novel image patch selection strategy during the network's training phase.
• Application of effective post-processing techniques on the network's output probability map during testing.

Main Results:

• The proposed method demonstrated enhanced performance in polyp segmentation within colonoscopy videos.
• Evaluation on the CVC-ColonDB database indicated superior accuracy compared to existing segmentation methods.
• The combined strategies of novel training and post-processing significantly improved segmentation results.

Conclusions:

• The developed convolutional neural network-based polyp segmentation method offers a more accurate approach for analyzing colonoscopy videos.
• This advancement can aid in the earlier and more reliable diagnosis of polyps, contributing to better patient outcomes in colorectal cancer.
• The proposed techniques represent a significant step forward in automated polyp detection for colonoscopy.