Automatic colon segmentation on T1-FS MR images.

Area of Science:

• Medical Imaging
• Gastroenterology
• Computational Biology

Background:

• Colonic content analysis via Magnetic Resonance Imaging (MRI) is crucial for understanding diet's impact on gut microbiota.
• T2-weighted MRI allows colon lumen segmentation, but T1-weighted Fat-Sat is needed to distinguish fecal and gas contents.
• Manual segmentation of T1-weighted Fat-Sat images is difficult, and automatic methods are lacking.

Purpose of the Study:

• To develop a non-supervised algorithm for accurate colon segmentation in T1-weighted Fat-Sat MRI.
• To enable automated differentiation of fecal and gas contents within the colon.
• To provide a tool for enhanced clinical diagnosis and gut microbiota research.

Main Methods:

• A two-phase algorithm combining deformable registration and a novel Iterative Colon Registration process.
• Utilizes mesh deformation guided by a probabilistic model for colon boundary likelihood.
• Incorporates a shape preservation process from T2-weighted segmentations for T1-weighted images.

Main Results:

• Achieved 93±5% accuracy in identifying ground truth labeled feces.
• Demonstrated accurate colon segmentation in T1-weighted Fat-Sat images through iterative refinement.
• Validated on 154 scans across 3 acquisition machines, showing accuracy, usability, and stability.

Conclusions:

• The proposed non-supervised algorithm provides accurate T1-weighted Fat-Sat colon segmentation.
• The method is suitable for clinical applications and research due to its accuracy and stability.
• Facilitates improved analysis of colonic contents for gut microbiota studies and diagnostics.