Related Experiment Video
Updated: Jul 30, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
457
Contour-aware network with class-wise convolutions for 3D abdominal multi-organ segmentation
Hongjian Gao1, Mengyao Lyu2, Xinyue Zhao3
1Image Processing Center, Beihang University, Beijing 102206, China.
Medical Image Analysis
|May 17, 2023
Summary
This study introduces a novel method for segmenting multiple abdominal organs in 3D CT scans, improving accuracy and efficiency for medical procedures. The approach enhances organ boundary delineation and highlights specific anatomical features for better segmentation results.
Area of Science:
- Medical Imaging
- Computer Vision
- Radiology
Background:
- Accurate multi-organ segmentation is crucial for medical procedures but is often operator-dependent and time-consuming.
- Existing methods struggle with simultaneous segmentation of diverse organ shapes and sizes, lacking exploitation of multi-organ task-specific traits.
Purpose of the Study:
- To develop an improved method for accurate and efficient multi-organ segmentation in 3D CT scans.
- To address limitations of current techniques by considering unique characteristics of multi-organ segmentation tasks.
Main Methods:
- Supplementing a region segmentation backbone with a contour localization task to enhance boundary certainty.
- Utilizing class-wise convolutions to address anatomical variability and highlight organ-specific features.
- Developing and validating the method on a multi-center dataset of 110 3D CT scans with 14 abdominal organs.
Main Results:
- Achieved state-of-the-art performance for most abdominal organs.
- Obtained an average 95% Hausdorff Distance of 3.63 mm and Dice Similarity Coefficient of 83.32%.
- Extensive ablation and visualization studies confirmed the method's effectiveness.
Conclusions:
- The proposed method effectively improves multi-organ segmentation accuracy by integrating contour localization and class-wise convolutions.
- Demonstrated superior performance on a comprehensive multi-center dataset, advancing the field of medical image analysis.
