Endoluminal surface registration for CT colonography using haustral fold matching

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Gastroenterology

Background:

• Computed Tomographic (CT) colonography aids in detecting bowel cancer and polyps.
• Differentiating fixed pathology from mobile residue requires prone and supine patient positioning.
• Colonic deformations during repositioning complicate manual landmark matching.

Purpose of the Study:

• To develop an automated method for establishing correspondence between prone and supine CT colonography acquisitions.
• To overcome challenges in manual landmark matching caused by colonic deformations.

Main Methods:

• Utilized graph cuts and curvature-based metrics on segmented colonic lumen surface meshes to detect haustral folds.
• Employed a virtual camera to generate matching metrics and optimized image patch registration.
• Applied a Markov Random Field (MRF) model with unary and pair-wise costs for fold labeling.

Main Results:

• Achieved high fold matching accuracy: 96.0% with local colonic collapse and 96.1% without.
• Improved upon existing surface-based registration algorithms by providing initialization.
• Reduced mean error in surface correspondence from 11.9 mm to 6.0 mm.

Conclusions:

• The proposed automated method accurately matches colonic folds in CT colonography.
• This technique enhances the reliability of CT colonography for polyp and cancer detection.
• The method offers significant improvements in surface correspondence accuracy and efficiency.