Level set image segmentation with a statistical overlap constraint.

Area of Science:

• Medical Imaging
• Computer Vision
• Statistical Modeling

Background:

• Active curve segmentation methods often struggle with regions of overlapping data distributions between foreground and background.
• Existing intensity-driven constraints may be insufficient for complex segmentation tasks.

Purpose of the Study:

• To develop and evaluate a novel statistical overlap constraint for active curve image segmentation.
• To improve segmentation accuracy in medical images with significant foreground-background data overlap.

Main Methods:

• Modeling the overlap between nonparametric data distributions using a Gaussian prior, with parameters learned from training images.
• Minimizing a functional incorporating the statistical overlap constraint and regularization terms.
• Deriving and analyzing the Euler-Lagrange curve evolution equation influenced by the constraint.

Main Results:

• The statistical overlap constraint significantly outperforms the commonly used likelihood prior in level set segmentation.
• The Gaussian prior assumption was found to be sufficient for cardiac Magnetic Resonance (MR) images.
• The method demonstrated flexibility by reducing the need for complex geometric training and accurate background distribution learning.

Conclusions:

• The proposed statistical overlap constraint offers a robust and flexible approach for active curve image segmentation, particularly in challenging scenarios.
• This method enhances segmentation performance and offers practical advantages for clinical applications involving MR and Computed Tomography (CT) imaging.