Related Experiment Video
Updated: May 28, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.1K
Point-annotation supervision for robust 3D pulmonary infection segmentation by CT-based cascading deep learning.
Yuetan Chu1, Jianpeng Wang2, Yaxin Xiong2
1Center of Excellence for Smart Health (KCSH), King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia.
Computers in Biology and Medicine
|February 9, 2025
Summary
This study introduces a novel point-annotation framework for segmenting pulmonary infections in CT scans. The method significantly improves accuracy with less manual labeling, outperforming existing weakly-supervised techniques.
Area of Science:
- Medical Imaging
- Computer Vision
- Radiology
Background:
- Pulmonary infection segmentation is vital for diagnosis and treatment monitoring.
- Manual annotation of large datasets is labor-intensive and time-consuming.
- Existing weakly-supervised methods struggle with the complex variations in pulmonary infection imaging.
Purpose of the Study:
- To develop a weakly-supervised framework for pulmonary infection segmentation using sparse point annotations.
- To address challenges like high variability, ambiguous boundaries, and poor contrast in 3D medical images.
- To achieve high performance comparable to fully-supervised methods with reduced annotation effort.
Main Methods:
- A cascading point-annotation framework utilizing sparse annotations.
- Regularization strategies based on point-voxel comparison and global uncertainty.
- An enhancement module for anatomical perception and spatial anisotropy adaptation.
- A texture-aware variational module for consistent boundary delineation.
Main Results:
- Outperformed state-of-the-art weakly-supervised methods by 3%-6% in dice score on a large dataset (1,072 CT volumes).
- Achieved performance comparable to fully-supervised methods on external datasets.
- Demonstrated robust performance on an unseen infection subtype (Mycoplasma pneumoniae).
Conclusions:
- The proposed framework offers a highly effective and efficient solution for pulmonary infection segmentation.
- It significantly reduces annotation burden while maintaining high accuracy.
- The method shows broad applicability for emerging pulmonary infections.
