Improving initial polyp candidate extraction for CT colonography.

Area of Science:

• Medical Imaging
• Computer-Aided Detection
• Gastroenterology

Background:

• Reducing false positives (FPs) is crucial for effective computer-aided detection (CAD) of colonic polyps.
• Accurate segmentation of initial polyp candidates (IPCs) is essential for subsequent FP reduction steps.

Purpose of the Study:

• To introduce an improved scheme for segmenting IPCs in colonic polyp detection.
• To enhance the accuracy and robustness of polyp candidate segmentation within a CAD pipeline.

Main Methods:

• Developed a novel IPC segmentation strategy using geodesic distance-based merging of suspicious patches (SPs).
• Incorporated a convex dilation component to grow SPs into volumes of interest (VOIs) with a convex inner border.
• Evaluated the method on 100 scans from 50 patients, including 84 polyps and masses.

Main Results:

• The new IPC segmentation strategy demonstrated a lower FP rate (4.78 FPs per scan) at 0.90 sensitivity compared to existing methods (Nappi: 6.37, Zhu: 7.01).
• The method avoided undesirably merging true polyps.
• Provided improved image intensity statistics (mean and variance) from extracted VOIs for better TP/FP classification.

Conclusions:

• The proposed IPC segmentation strategy significantly reduces FPs in colonic polyp CAD systems.
• This method offers a more effective approach to VOI extraction for improved polyp detection accuracy.
• The enhanced segmentation benefits downstream classification tasks in CAD systems.