Reduction of false positives by internal features for polyp detection in CT-based virtual colonoscopy

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Gastroenterology

Background:

• Surface-based measures in computed tomographic (CT) colonography for polyp detection generate a high rate of false positives (FPs).
• Existing methods often struggle to differentiate between colonic polyps and normal colon wall features based solely on external geometry.

Purpose of the Study:

• To develop and evaluate a novel computer-aided detection (CAD) method that utilizes internal features to reduce the FP rate in CT colonography.
• To improve the accuracy of polyp detection by analyzing the internal characteristics of suspicious colonic regions.

Main Methods:

• Introduction of a new 'global curvature' shape descriptor for improved detection of suspicious patches on the colon wall.
• Extraction of detected patches as fitted ellipsoid models using a ray-driven edge finder.
• Analysis of CT image density distribution within ellipsoid models to extract geometrical, morphological, and textural internal features for FP elimination.

Main Results:

• The CAD method achieved 100% sensitivity for polyp detection across various sizes (4-30 mm) in 153 patient datasets.
• An average of 2.68 FPs per patient dataset was recorded, with 93.1% of FPs from surface-based measures being eliminated.
• For polyps sized 10-30 mm, the method yielded 2.0 FPs per dataset; for 4-10 mm polyps, it resulted in 3.44 FPs per dataset, both with 100% sensitivity.

Conclusions:

• The proposed CAD method effectively reduces false positives in CT colonography by incorporating internal feature analysis.
• This approach enhances the reliability of CT colonography for colonic polyp screening, offering high sensitivity with a manageable FP rate.