Liver tumors segmentation from CTA images using voxels classification and affinity constraint propagation

Area of Science:

• Medical Imaging
• Computational Anatomy
• Radiology

Background:

• Accurate segmentation of liver tumors in CT angiography (CTA) scans is crucial for diagnosis and treatment planning.
• Existing semi-automatic methods often require substantial user interaction, limiting their efficiency.

Purpose of the Study:

• To develop and validate a nearly automatic method for liver tumor segmentation in CTA scans.
• To reduce user interaction while maintaining or improving segmentation accuracy.

Main Methods:

• A Support Vector Machine (SVM) classifier is used to distinguish tumorous from healthy liver tissue based on user-defined seeds.
• An energy function guides seed propagation, optimized using the conjugate gradients method for continuous segmentation.
• The final segmentation is obtained by thresholding the continuous map.

Main Results:

• The method achieved an average symmetric surface distance of 1.76 mm, outperforming other methods on the same dataset (2.0 mm).
• Volumetric overlap error was comparable to existing methods (33.8% vs. 32.6%).
• The approach demonstrated robustness across various tumor types with significantly reduced user interaction.

Conclusions:

• The developed method is accurate, efficient, and robust for liver tumor segmentation in CTA scans.
• It offers comparable or superior performance to other semi-automatic methods with substantially less user effort.
• This technique holds promise for improving clinical workflows in liver tumor management.