Related Experiment Video
Updated: Sep 22, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
531
Multi-Dimensional Cascaded Net with Uncertain Probability Reduction for Abdominal Multi-Organ Segmentation in CT
Chengkang Li1, Yishen Mao2, Yi Guo1
1School of Information Science and Technology of Fudan University, Shanghai 200433, China; Key Laboratory of Medical Imaging Computing and Computer Assisted Intervention (MICCAI) of Shanghai, Shanghai 200032, China.
Computer Methods and Programs in Biomedicine
|May 21, 2022
Summary
A novel deep learning algorithm, multi-dimensional cascaded net (MDCNet), improves 3D CT abdominal organ segmentation. This method enhances accuracy for small and irregular organs, offering better preoperative surgical guidance.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Surgical Planning
Background:
- Deep learning for abdominal multi-organ segmentation is crucial for preoperative surgical guidance.
- Existing 3D CT segmentation methods struggle with balancing semantic features and high-resolution details, leading to inaccuracies, especially for small or irregular organs.
- This limitation results in uncertain and rough segmentation outcomes.
Purpose of the Study:
- To propose a novel two-stage algorithm, multi-dimensional cascaded net (MDCNet), for accurate multi-organ segmentation in 3D CT images.
- To address the challenges of balancing semantic completeness and high-resolution detail information in large-volume CT sequences.
- To improve segmentation accuracy for small and irregular abdominal organs.
Main Methods:
- MDCNet integrates a 3D network for semantic feature extraction and a 2.5D network for high-resolution detail recovery.
- Stage 1 employs a prior-guided 3D location net for rough segmentation and uses circular inference with parameter Dice loss to refine boundaries.
- Stage 2 utilizes high-resolution slices and multi-view 2.5D networks to compensate for lost details and spatial information, ensuring smooth and accurate contours.
Main Results:
- MDCNet achieved state-of-the-art performance on two experimental datasets.
- The method demonstrated superior Dice scores for small gallbladders (0.85±0.12) and irregular duodenums (0.77±0.07), outperforming existing methods.
- Improvements of 0.02 and 0.03 in Dice scores were observed for these challenging organs.
Conclusions:
- MDCNet effectively extracts both semantic and high-resolution details from large-volume CT images.
- The algorithm significantly reduces boundary uncertainty and produces smoother segmentation edges.
- The enhanced accuracy and detail indicate strong potential for clinical application in abdominal surgery.
