An Improved Random Walker with Bayes Model for Volumetric Medical Image Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• The Random Walk (RW) method is common for volumetric medical image segmentation but suffers from large graph sizes and inaccurate segmentation due to poor seed point selection.
• Classical RW algorithms do not effectively utilize organ intensity and shape information.

Purpose of the Study:

• To develop a prior knowledge-based Bayes random walk framework for improved volumetric medical image segmentation.
• To enhance segmentation accuracy by utilizing information from previously segmented slices.

Main Methods:

• A slice-by-slice segmentation approach using a Bayes random walk framework.
• Employing prior shape and intensity knowledge from adjacent segmented slices to guide segmentation.
• Dynamically updating seed points using the narrow band threshold (NBT) method and a Gaussian process-based organ model.

Main Results:

• The proposed Bayes RW framework significantly improved liver segmentation accuracy compared to conventional RW and state-of-the-art interactive methods (p < 0.001).
• The method demonstrated effective utilization of prior knowledge for automated, high-quality image segmentation.

Conclusions:

• The prior knowledge-based Bayes random walk framework offers a more accurate and robust solution for volumetric medical image segmentation.
• This approach addresses limitations of traditional RW methods by incorporating anatomical and intensity priors for improved organ segmentation.