Related Experiment Video
Updated: Mar 14, 2026
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.9K
Accurate Lungs Segmentation on CT Chest Images by Adaptive Appearance-Guided Shape Modeling
IEEE Transactions on Medical Imaging
|October 6, 2016
Summary
This study introduces a novel 3D Markov-Gibbs random field model for accurate lung segmentation in CT scans, crucial for computer-aided disease diagnostics. The advanced model achieves high accuracy, outperforming existing methods in clinical evaluations.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnostics
- Computational Anatomy
Background:
- Accurate lung segmentation in CT scans is vital for diagnosing various lung pathologies.
- Existing segmentation methods often struggle with the complexity and variability of lung structures and pathologies.
- Reliable segmentation is a prerequisite for quantitative analysis and computer-aided disease diagnostics.
Purpose of the Study:
- To develop and validate a novel, accurate lung segmentation framework for chest CT scans.
- To improve the reliability of computer-aided disease diagnostics through precise pathological and healthy lung segmentation.
- To integrate visual appearance and shape information adaptively for robust segmentation.
Main Methods:
- A learnable 3D Markov-Gibbs random field (MGRF) model integrating visual appearance and adaptive shape sub-models.
- First-order appearance sub-model using original CT and Gaussian scale space (GSS) filtered images with linear combination of discrete Gaussians (LCDG) approximation.
- Second-order appearance sub-model quantifying voxel intensity dependencies and an adaptive shape sub-model trained on diverse lung data.
Main Results:
- The proposed framework achieved high segmentation accuracy across multiple datasets, including public challenges (VESSEL12, LOLA11) and a custom database.
- Quantitative metrics demonstrated excellent performance: Dice similarity coefficients (up to 99.0±0.5%), low 95-percentile bidirectional Hausdorff distances (as low as 2.1±1.6 mm), and minimal volume differences (0.39±0.20%).
- The method ranked first in the LOLA11 competition, achieving an average overlap of 98.0% with expert segmentations.
Conclusions:
- The proposed learnable 3D MGRF model provides highly accurate and robust lung segmentation for chest CT images.
- This framework significantly enhances the reliability of computer-aided disease diagnostics by providing precise lung volume delineation.
- The integration of appearance and adaptive shape models offers a superior approach compared to existing state-of-the-art techniques.
