Automatic liver contouring for radiotherapy treatment planning

Area of Science:

• Medical Imaging
• Radiation Oncology
• Computational Anatomy

Background:

• Accurate liver contouring is essential for effective radiation therapy planning.
• Current manual contouring methods are time-consuming and prone to inter-observer variability.
• Challenges exist in distinguishing liver tissue from surrounding structures due to similar image intensities.

Purpose of the Study:

• To develop an automatic and efficient liver contouring software for 3D-CT and 4D-CT.
• To improve the precision and speed of liver segmentation for radiation therapy planning systems.
• To overcome the challenge of similar intensities between the liver and surrounding tissues.

Main Methods:

• A three-step algorithm combining Total Variation with L1 norm (TV-L1) for noise reduction and edge preservation.
• An improved level set model incorporating global and local energy functions, using Local Correlation Coefficient (LCC) for contour extraction.
• Voxel-based texture characterization for refining liver region segmentation and obtaining final contours.

Main Results:

• The algorithm demonstrated high accuracy on planning CT images, with Dice Similarity Coefficients ranging from 91.01% to 97.21% for normal livers and 86.14% to 93.53% for diseased livers.
• Performance on 4D-CT images showed Dice Similarity Coefficients between 82.23% and 89.44% for diseased livers.
• Low false positive (2.15-3.96%) and false negative (2.96-4.57%) volume percentages were achieved across all tested image types.

Conclusions:

• The proposed three-step method provides efficient and automatic liver contouring for both CT and 4D-CT images.
• This technique is suitable for follow-up treatment planning and has potential for widespread application in future treatment planning systems.
• The algorithm effectively addresses the challenge of similar tissue intensities, enhancing segmentation accuracy.