Automatic segmentation of polyps in colonoscopic narrow-band imaging data

Area of Science:

• Medical imaging
• Computational pathology
• Gastroenterology

Background:

• Colorectal cancer is a leading global cancer, preventable through early detection of adenomatous polyps during optical colonoscopy (OC).
• Current OC practices involve removing all polyps for histological analysis, including benign hyperplastic polyps, leading to unnecessary risks and costs.
• Accurate polyp segmentation is crucial for automated polyp classification systems but is often hindered by the need for manual segmentation.

Purpose of the Study:

• To develop an automated polyp segmentation algorithm for colonoscopic narrow-band imaging (NBI) data.
• To overcome the limitation of manual segmentation required by existing polyp classification algorithms.
• To improve the efficiency and accuracy of polyp identification during optical colonoscopy.

Main Methods:

• Development of Shape-UCM, an enhanced version of the gPb-OWT-UCM boundary detection and segmentation algorithm.
• Integration of prior knowledge about polyp shapes into the Shape-UCM algorithm to address scale selection issues.
• Automatic segmentation of polyps in colonoscopic NBI images.

Main Results:

• Shape-UCM achieved high performance in automatic polyp segmentation on a test set of 87 images.
• The algorithm demonstrated a specificity of 92%, a sensitivity of 71%, and an accuracy of 88%.
• Shape-UCM outperformed previous segmentation methods by incorporating polyp shape priors.

Conclusions:

• Shape-UCM represents a significant advancement in automated polyp segmentation for optical colonoscopy.
• This novel approach can facilitate the development of more effective optical biopsy applications.
• Automated segmentation can reduce unnecessary polyp removals and histological analysis, improving patient outcomes and healthcare efficiency.