Active contours driven by difference of Gaussians.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Computational Imaging

Background:

• Intensity inhomogeneity poses a significant challenge for accurate image segmentation.
• Traditional edge-based active contour methods struggle with global structure analysis in such images.
• Existing methods for intensity inhomogeneous images often have high time complexity or sensitivity to initialization.

Purpose of the Study:

• To propose a novel edge-based active contour method for segmenting intensity inhomogeneous images.
• To leverage the difference of Gaussians (DoG) as an edge-indicator and balloon force.
• To address limitations of traditional methods, including global structure analysis and computational efficiency.

Main Methods:

• A novel active contour model is developed utilizing the difference of Gaussians (DoG) function.
• The DoG function is integrated into the energy functional as an edge-indicator parameter, mimicking a balloon force.
• The internal energy term enforces a signed distance function, while the external energy term guides contour evolution towards object boundaries.

Main Results:

• The proposed DoG-based method effectively captures the global structure of images, enabling global segmentation.
• Achieved lower time complexity compared to state-of-the-art active contour methods for intensity inhomogeneity.
• Demonstrated robustness to the initial placement of the contour.
• Experimental results on synthetic and real brain MR images show superior segmentation performance.

Conclusions:

• The novel edge-based active contour method effectively segments intensity inhomogeneous images.
• The DoG function provides a robust mechanism for global structure analysis and contour evolution.
• The method offers significant advantages in terms of accuracy, efficiency, and initialization insensitivity for medical image segmentation.