A combinatorial Bayesian and Dirichlet model for prostate MR image segmentation using probabilistic image features.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate prostate MR image segmentation is challenging due to blurred boundaries and heterogeneous intensities.
• Existing methods struggle to achieve precise delineation of prostate anatomy in MR images.

Purpose of the Study:

• To develop and evaluate an advanced approach for improving prostate MR image segmentation accuracy.
• To address the limitations of current segmentation techniques by incorporating novel computational methods.

Main Methods:

• A multi-stage approach involving image patch division for homogeneous segmentation, relevance vector machine-based feature classification for probabilistic priors, and a Bayesian-informed graph energy formulation.
• A non-iterative graph energy minimization scheme utilizing matrix differentiation for probabilistic pixel membership optimization.

Main Results:

• The proposed method achieved a mean Dice Similarity Coefficient (DSC) of 0.90 ± 0.02 on the PROMISE-12 dataset.
• The segmentation performance surpassed the five best prior-based methods in the PROMISE-12 challenge.

Conclusions:

• The developed approach significantly enhances prostate MR image segmentation accuracy.
• The novel combination of image processing, machine learning, and graph-based optimization offers a robust solution for clinical applications.