3D liver segmentation using multiple region appearances and graph cuts.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Computational Anatomy

Background:

• Accurate 3D liver segmentation is crucial for treating hepatic diseases.
• Challenges include inhomogeneous appearances, ambiguous boundaries, and anatomical variations.
• Liver abnormalities and tumors complicate segmentation tasks.

Purpose of the Study:

• To develop an efficient and accurate 3D liver segmentation method for contrast-enhanced computed tomography (CT) images.
• To address challenges posed by inhomogeneous liver appearances and abnormalities.
• To segment livers with multiple subregions of distinct appearances.

Main Methods:

• A novel multiregion-appearance based approach using graph cuts for liver surface delineation.
• A geodesic distance-based scheme for selecting appearance constraints in multi-subregion livers.
• Energy functions incorporating boundary and region information with adaptive balancing weights.
• Method requires only initialization within the liver, no additional user interaction.

Main Results:

• Validated on 50 3D CT images across three datasets.
• Achieved a total score of 83.4 ± 3.1 on the MICCAI testing set, outperforming manual segmentation (75.0).
• Yielded high Dice similarity coefficients (DSC): 97.7% ± 0.5% (MICCAI training) and 97.5% ± 0.4% (local dataset).
• Demonstrated good reproducibility in user operator variability experiments.

Conclusions:

• A multiregion-appearance based method effectively segments livers from CT images.
• Eliminates the need for prior model construction and matching.
• Achieves results comparable to state-of-the-art methods.
• Results suggest suitability for clinical use.