Related Experiment Video
Updated: Aug 20, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.7K
Automatic 3D CT liver segmentation based on fast global minimization of probabilistic active contour
Renchao Jin1, Manyang Wang1, Lijun Xu2
1School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, Hubei, China.
Medical Physics
|November 22, 2022
Summary
This study introduces a novel, fast, and explainable method for automatic liver segmentation in CT images using a probabilistic active contour model. The approach achieves high accuracy, outperforming traditional methods and rivaling deep learning techniques for liver disease diagnosis.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Computational Anatomy
Background:
- Automatic liver segmentation from CT images is crucial for diagnosing liver diseases.
- Challenges include variations in liver shape, low contrast, and presence of abnormalities.
- Existing methods often struggle with accuracy and speed.
Purpose of the Study:
- To present an accurate and fast liver segmentation method using a novel probabilistic active contour (PAC) model and its fast global minimization scheme (3D-FGMPAC).
- To provide an explainable alternative to deep learning methods for liver segmentation.
- To improve preprocessing for computer-aided diagnosis of liver diseases.
Main Methods:
- A slice-indexed-histogram localizes the volume of interest and estimates voxel probability for the liver.
- A new contour indicator function combines gradient-based and Hessian-matrix-based detection.
- A fast numerical scheme evolves the probabilistic image to a global minimizer, followed by region growing and B-spline fitting to extract the liver mask.
Main Results:
- The method achieved high performance on public datasets, with average Dice scores of 0.96 (Sliver07) and 0.95 (3Dircadb).
- Excellent accuracy was demonstrated with low volume overlap error (7.35%, 8.89%) and symmetric surface distance (1.17 mm, 1.45 mm).
- Processing times were efficient, averaging 19.8 s and 23.08 s per dataset.
Conclusions:
- The fully-automatic approach effectively segments livers from complex CT images.
- The method surpasses traditional algorithms and is competitive with deep learning approaches.
- This technique offers a promising, explainable solution for liver segmentation in medical imaging.
