Automatic segmentation of breast MR images through a Markov random field statistical model

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate segmentation of breast magnetic resonance images (MRIs) is crucial for diagnosis and treatment planning.
• Existing segmentation methods often struggle with noise and partial volume effects, limiting their clinical utility.

Purpose of the Study:

• To develop and evaluate an automated algorithm for segmenting breast MRIs.
• To enhance segmentation accuracy by incorporating advanced denoising and statistical modeling techniques.

Main Methods:

• A pipeline combining anisotropic diffusion for denoising and a Markov random field model for statistical segmentation.
• Continuous parameter updating for automatic denoising and addressing partial volume effects during labeling.
• Validation using synthetic data and clinical breast MRI datasets from high-risk populations.

Main Results:

• The algorithm demonstrated superior accuracy compared to traditional clustering algorithms like K-means.
• Significant improvements in Jaccard similarity coefficients were observed for adipose, glandular, muscle, and skin components (e.g., 114-140% increase).
• Good agreement was achieved between automated segmentations and manual segmentations by an experienced radiologist (Jaccard coefficients of 0.769, 0.756, and 0.694).

Conclusions:

• The proposed algorithm effectively segments breast MRIs, benefiting from data-guided denoising and Markov random field-based spatial dependency.
• The method shows high relevance for clinical datasets and offers a promising tool for automated breast MRI analysis.
• The integration of denoising and spatial modeling significantly enhances segmentation performance over conventional approaches.