Balancing the data term of graph-cuts algorithm to improve segmentation of hepatic vascular structures

Area of Science:

• Medical Imaging
• Computational Anatomy
• Image Segmentation

Background:

• Accurate hepatic vessel segmentation is crucial for liver disease diagnosis and treatment planning.
• Extracting small hepatic vessels remains a challenge in medical image analysis.

Purpose of the Study:

• To improve the segmentation accuracy of hepatic vessels, particularly small structures.
• To adaptively calculate the data term in the conventional graph-cuts algorithm for enhanced vessel delineation.

Main Methods:

• An adaptive approach was developed to calculate the data term in graph-cuts for hepatic vessel segmentation.
• Statistical parameters of vessels were adaptively estimated using Gaussian distributions and Expectation-Maximization on CT images.
• A medial-axes enhancement algorithm identified vessel axes, which informed the cost function for graph-cuts.

Main Results:

• The proposed method achieved a mean Dice score of 95.51% on synthetic and clinical data.
• It successfully segmented small hepatic vessels missed by traditional methods, including conventional graph-cuts.
• Performance was robust in the presence of image noise.

Conclusions:

• The adaptive graph-cuts method significantly improves the segmentation of small hepatic vessels.
• This technique enhances the delineation of liver vasculature in noisy CT images.
• The findings support its utility in clinical diagnosis and treatment planning.