Dynamic Regulation of Level Set Parameters Using 3D Convolutional Neural Network for Liver Tumor Segmentation

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Accurate liver tumor segmentation is crucial for effective treatment planning and monitoring in liver disease.
• Existing segmentation methods often lack full automation or struggle with complex tumor characteristics.

Purpose of the Study:

• To develop a fully automatic and robust liver tumor segmentation method using a dynamic level set approach.
• To enhance segmentation accuracy and reliability for diverse and challenging tumor cases.

Main Methods:

• A 3D convolutional neural network (CNN) estimates tumor contour locations.
• CNN outputs dynamically regulate energy functional parameters within a level set framework.
• Automatic initialization and local window sizing are derived from contour probabilities for full automation.

Main Results:

• The proposed method achieved a significant average DICE improvement of 0.15 compared to fixed-parameter level sets.
• On the challenging 3DIRCADb dataset, a competitive DICE score of 0.85 ± 0.06 was obtained.
• The method demonstrated robustness on datasets with low contrast, heterogeneous tumors, and noisy images.

Conclusions:

• The dynamic regulation algorithm offers a significant advancement in automated liver tumor segmentation.
• This approach enhances the precision and efficiency of segmentation for clinical applications.
• The method shows strong potential for improving liver disease management through accurate tumor delineation.