Adaptive deformable model for colonic polyp segmentation and measurement on CT colonography

Area of Science:

• Medical Imaging
• Computational Pathology
• Gastroenterology

Background:

• Polyp size is a key biomarker for assessing colorectal polyp malignancy risk.
• Accurate segmentation and measurement of colonic polyps are crucial for clinical decision-making.
• CT colonography (CTC) offers a non-invasive method for polyp detection and characterization.

Purpose of the Study:

• To present an improved computational approach for colonic polyp segmentation and measurement using CT colonography images.
• To enhance the accuracy of polyp size measurements, especially for polyps located on haustral folds.
• To provide reliable linear and volumetric measurements for malignancy risk assessment.

Main Methods:

• A novel method combining knowledge-guided intensity adjustment, fuzzy clustering, and an adaptive deformable model for polyp segmentation.
• Development of a dual-distance algorithm with a counter-force mechanism to improve segmentation of polyps on haustral folds.
• Derivation of linear and volumetric measurements from automated segmentation and validation against optical colonoscopy and manual CTC measurements.

Main Results:

• High volumetric measurement correlation (R²=0.98) and good linear measurement correlation (R²=0.61) compared to manual methods.
• Mean Dice coefficient of 0.752 (SD 0.154) for automated versus manual volume overlap.
• No significant difference found via paired t-test when comparing automated and manual measurements.

Conclusions:

• The proposed method provides accurate and reliable colonic polyp segmentation and measurement on CT colonography.
• The dual-distance algorithm effectively addresses the challenge of segmenting polyps on haustral folds.
• This improved approach has the potential to enhance the non-invasive assessment of polyp malignancy risk.