A region-appearance-based adaptive variational model for 3D liver segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Liver segmentation in computed tomography (CT) images is difficult due to overlapping pixel intensities, unclear edges, and complex backgrounds.
• Accurate liver segmentation is crucial for various clinical applications, including diagnosis and treatment planning.

Purpose of the Study:

• To develop a novel active surface model for robust and accurate liver segmentation from CT images.
• To address challenges like ambiguous edges and complex backgrounds in liver segmentation.

Main Methods:

• A semiautomatic active surface scheme minimizing an energy functional combining edge- and region-based information.
• Incorporation of multiple features, appearance context, multilayer consecutiveness, and local organ deformation to prevent oversegmentation.
• Use of case-specific constraints and spatially adaptive balancing weights to handle image variations and nonuniformity.

Main Results:

• The model effectively discriminates the liver from background tissues, even with weak gradients or missing edge evidence.
• Achieved an average surface distance of 0.9 mm and 93.9% volume overlap on the MICCAI dataset, outperforming state-of-the-art methods.
• Demonstrated reproducible segmentation with low score variances across different initial conditions.

Conclusions:

• The proposed model efficiently delineates ambiguous liver edges from complex backgrounds with high reproducibility.
• Quantitative validations confirm the model's accuracy and efficacy in liver segmentation.