Related Experiment Video
Updated: Jul 21, 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
14.2K
Semi-Supervised Segmentation of Interstitial Lung Disease Patterns from CT Images via Self-Training with Selective
Guang-Wei Cai1, Yun-Bi Liu1, Qian-Jin Feng1
1School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China.
Bioengineering (Basel, Switzerland)
|July 29, 2023
Summary
This study introduces ESSegILD, a semi-supervised deep learning method for segmenting interstitial lung disease (ILD) patterns in CT scans. It effectively uses limited annotations to improve segmentation accuracy for ILD diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Accurate segmentation of interstitial lung disease (ILD) patterns in computed tomography (CT) is crucial for patient management.
- Manual pixel-by-pixel segmentation of ILD patterns is time-consuming, limiting the availability of large annotated datasets for deep learning.
- Data scarcity poses a significant challenge for developing effective data-driven deep learning models for ILD segmentation.
Purpose of the Study:
- To develop an end-to-end semi-supervised learning framework, ESSegILD, for efficient segmentation of ILD patterns from CT images.
- To address the challenge of limited annotated data by leveraging slice-wise sparse annotations.
- To improve the performance of deep learning models for ILD segmentation using partially labeled datasets.
Main Methods:
- Proposed an end-to-end semi-supervised learning framework (ESSegILD) utilizing a Mean-Teacher model with consistency regularization.
- Incorporated a self-training technique with selective re-training to utilize reliable pseudo-labels generated by the teacher model.
- Trained the model on a large CT dataset with slice-wise sparse annotations, expanding training samples iteratively.
Main Results:
- The ESSegILD framework effectively utilized unlabeled data from a partially annotated dataset to enhance segmentation performance.
- Experiments on a dataset of 67 pneumonia patients demonstrated superior performance compared to state-of-the-art methods.
- The method achieved accurate segmentation of eight different ILD lung patterns from over 11,000 CT images.
Conclusions:
- ESSegILD offers a powerful solution for segmenting ILD patterns in CT scans, overcoming data annotation limitations.
- The proposed semi-supervised approach with self-training and selective re-training significantly improves segmentation accuracy.
- This framework holds promise for advancing automated ILD analysis in clinical practice.
