Patient-specific probabilistic atlas combining modified distance regularized level set for automatic liver...

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Automatic liver segmentation from CT scans is crucial for computer-assisted clinical applications.
• Existing methods struggle with variable liver shapes and low contrast in CT images.
• Manual segmentation is time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop a robust and precise automatic liver segmentation method using a patient-specific probabilistic atlas (PA) and a modified distance regularized level set model.
• To address the limitations of current automatic liver segmentation techniques, particularly concerning image quality and anatomical variability.
• To provide an efficient and accurate alternative to manual liver segmentation for clinical use.

Main Methods:

• A patient-specific probabilistic atlas (PA) was generated by calculating similarities between training atlases and testing patient images, creating weighted atlases.
• A most likely liver region (MLLR) was determined from the patient-specific PA.
• A modified distance regularized level set model, utilizing both edge and region information as balloon force, was employed for final segmentation refinement.

Main Results:

• The proposed method demonstrated robust and precise liver segmentation capabilities.
• Evaluation on 35 public datasets confirmed the effectiveness of the PA-based approach combined with the level set model.
• The method successfully addressed challenges posed by variable liver shapes and low contrast.

Conclusions:

• The developed patient-specific probabilistic atlas (PA)-based method offers a significant advancement in automatic liver segmentation from CT.
• This technique provides a reliable and accurate alternative to manual segmentation, potentially improving efficiency in clinical workflows.
• The combined approach of PA generation and modified level set refinement ensures high precision and robustness in liver segmentation.