Related Experiment Video
Updated: May 23, 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
14.0K
A novel 3D lightweight model for COVID-19 lung CT Lesion Segmentation
Jingdong Yang1, Shaoyu Huang1, Han Wang1
1School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China.
Medical Engineering & Physics
|March 8, 2025
Summary
This study introduces a lightweight 3D medical image segmentation model for improved lung area segmentation. The novel approach enhances accuracy and generalization while maintaining computational efficiency.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- 3D medical image segmentation offers superior spatial data over 2D methods.
- Challenges include limited data, imbalanced pixels, and poor generalization.
- Existing models struggle with efficiency and accuracy in 3D segmentation tasks.
Purpose of the Study:
- To develop a lightweight and efficient segmentation model for 3D medical images.
- To improve lung area segmentation accuracy and model generalization.
- To address challenges of small datasets and imbalanced data distributions.
Main Methods:
- Proposed a lightweight segmentation model using K-means for Region of Interest (ROI) extraction.
- Implemented Focal loss with Dice coefficient to mitigate overfitting.
- Incorporated parallel attention mechanisms and depth-wise separable convolutions for enhanced feature extraction and efficiency.
- Utilized Ghost-inspired 1x1 convolution and residual connections for gradient flow and feature consistency.
Main Results:
- Achieved superior performance on a dataset of 199 COVID-19-Seg cases via 5-fold cross-validation.
- Key metrics: Average Surface Distance (ASD) 19.880, accuracy 99.90%, sensitivity 58.90%, Dice coefficient 56.10%, Intersection over Union (IOU) 41.00%.
- Demonstrated improved segmentation accuracy and generalization compared to state-of-the-art models with minimal parameter increase.
Conclusions:
- The proposed lightweight model effectively addresses challenges in 3D medical image segmentation.
- It achieves high accuracy and generalization with improved computational efficiency.
- This model shows promise for clinical applications requiring precise lung segmentation.
