Related Experiment Video
Updated: Aug 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
14.3K
A Systematic Evaluation of Ensemble Learning Methods for Fine-Grained Semantic Segmentation of
Sivaramakrishnan Rajaraman1, Feng Yang1, Ghada Zamzmi1
1National Library of Medicine, National Institutes of Health, Bethesda, MD 20892, USA.
Bioengineering (Basel, Switzerland)
|September 22, 2022
Summary
Fine-grained annotations improve tuberculosis lesion segmentation in chest X-rays. Stacking ensembles of U-Net models achieved superior performance, enhancing diagnostic accuracy for tuberculosis (TB).
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Automated segmentation of tuberculosis (TB)-consistent lesions in chest X-rays (CXRs) aids clinical decision-making.
- Current methods often use coarse bounding box annotations, leading to pixel-level inaccuracies.
- This can negatively impact the performance of deep learning (DL) semantic segmentation models.
Purpose of the Study:
- To evaluate the benefits of fine-grained annotations for TB lesion segmentation.
- To compare U-Net model variants and their ensembles for segmenting TB lesions in CXRs.
- To assess segmentation performance on both original and bone-suppressed CXRs.
Main Methods:
- Trained U-Net model variants using fine-grained annotations of TB-consistent lesions.
- Constructed ensembles of these U-Net models.
- Evaluated segmentation performance using ensemble methods: bitwise-AND, bitwise-OR, bitwise-MAX, and stacking.
- Tested on original and bone-suppressed frontal CXRs.
Main Results:
- The stacking ensemble demonstrated superior segmentation performance compared to individual models and other ensemble methods.
- Achieved a Dice score of 0.5743 (95% CI: 0.4055, 0.7431).
- Fine-grained annotations led to improved semantic segmentation of TB lesions.
Conclusions:
- Fine-grained annotations significantly enhance the performance of deep learning models for TB lesion segmentation in CXRs.
- Stacking ensembles offer a robust approach to improve segmentation accuracy.
- This study is the first to apply ensemble learning for fine-grained TB lesion segmentation.
