Robust extraction for low-contrast liver tumors using modified adaptive likelihood estimation.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Surgical Planning

Background:

• Accurate liver tumor segmentation is crucial for effective liver ablation surgery planning and treatment.
• Existing segmentation methods may struggle with tumors exhibiting low contrast, high noise, or heterogeneous densities.

Purpose of the Study:

• To propose a semiautomatic method for accurate and robust liver tumor segmentation.
• To enhance liver tumor segmentation for improved surgical planning and treatment outcomes.

Main Methods:

• A semiautomatic approach utilizing adaptive likelihood classification with a modified likelihood model.
• Initialization via a minimal enclosing ellipse/quasi-ellipsoid.
• Hybrid intensity likelihood modification with nonparametric density estimation and prior shape constraints.
• Adaptive classification for handling low-contrast, high-noise, or heterogeneous tumors.

Main Results:

• Experiments on 3Dircadb and LiTS datasets yielded average volumetric overlap errors of 27.05% and 35.72%, respectively.
• Algorithm robustness confirmed through validation by 5 operators with multiple selections.
• Successful segmentation of diverse tumors, including those with low contrast and blurred boundaries.

Conclusions:

• The proposed semiautomatic method demonstrates good performance across various tumor types.
• Reliable and consistent segmentation results are achievable, even with different initializations and operators.
• The method effectively addresses challenges posed by low-contrast and heterogeneous liver tumors.