A unified variational segmentation framework with a level-set based sparse composite shape prior.

Area of Science:

• Medical image analysis
• Computer vision
• Computational anatomy

Background:

• Accurate medical image segmentation is crucial for diagnosis and treatment planning.
• Low signal-to-noise ratio (SNR) and artifacts present significant challenges in automated segmentation.
• Shape priors are essential for robust segmentation, especially in complex anatomical regions.

Purpose of the Study:

• To develop a unified variational segmentation framework incorporating a novel level-set based sparse composite prior.
• To introduce a 'dynamic' shape prior that adapts during the segmentation process.
• To evaluate the framework's performance on segmenting the corpus callosum and liver in medical imaging data.

Main Methods:

• A unified variational segmentation framework was developed.
• A level-set based sparse composite prior was utilized for shape regularization.
• A block minimization/descent scheme was employed to solve the variational problem.
• The method was applied to 2D MR images (corpus callosum) and 3D CT volumes (liver).

Main Results:

• The proposed method achieved statistically significant higher accuracy compared to benchmark methods.
• Performance was evaluated using Dice Similarity Coefficient and Hausdorff distance.
• The framework successfully avoided erroneous segmentation of surrounding structures with similar intensities.
• Demonstrated robustness in low SNR conditions and in the presence of artifacts.

Conclusions:

• The unified variational segmentation framework with a dynamic shape prior offers superior performance in medical image segmentation.
• This approach enhances accuracy and robustness, outperforming existing level-set based methods.
• The dynamic shape prior effectively guides segmentation without compromising convergence.