A scalable approach to T2-MRI colon segmentation.

Area of Science:

• Medical Imaging
• Gastroenterology
• Computational Anatomy

Background:

• Colonic volume analysis is crucial for gastroenterologists, often requiring MRI of the unprepared colon without contrast.
• Current measurement methods are time-consuming and cumbersome for specialists.
• Accurate segmentation of the unprepared colon is essential for reliable volume assessment.

Purpose of the Study:

• To develop a quasi-automatic segmentation algorithm for the unprepared colon on T2-weighted MRI scans.
• To improve the efficiency and accuracy of colonic volume measurement in clinical settings.
• To provide a substantial step towards fully automated colon segmentation.

Main Methods:

• A three-stage pipeline involving a tubularity filter, medial path estimation, and region of interest delimitation.
• Utilizing custom segmentation algorithms to identify colon neighbors and abdominal fat capsules.
• Employing 3D graph-cuts within a three-stage multigrid approach for final segmentation.
• Testing on MRI scans with varying resolutions to assess accuracy and scalability.

Main Results:

• The algorithm demonstrated accuracy, efficiency, and usability in segmenting the unprepared colon.
• The multigrid architecture proved computationally scalable across different scan resolutions.
• Experimental results showed strong agreement with specialist-provided ground truth segmentations.
• The system is applicable to clinical routines for colon measurement.

Conclusions:

• The developed algorithm offers a significant advancement for quasi-automatic colon segmentation on MRI.
• It addresses the limitations of existing time-consuming manual methods.
• The approach is suitable for clinical integration, paving the way for automated colonic volume analysis.