Related Experiment Video
Updated: Nov 7, 2025
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
Deep Learning to Quantify Pulmonary Edema in Chest Radiographs.
Steven Horng1,2, Ruizhi Liao1,2, Xin Wang1,2
1Department of Radiology, Beth Israel Deaconess Medical Center, Harvard Medical School, 330 Brookline Ave, Boston, MA 02215 (S.H., S.J.B.); Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, Mass (R.L., P.G.); and Clinical Informatics Solutions and Services, Philips Research, Cambridge, Mass (X.W., S.D.).
Machine learning models accurately graded pulmonary edema severity on chest radiographs. These deep learning approaches achieved high performance in classifying edema from vascular congestion to alveolar edema.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Cardiology
Background:
- Pulmonary edema is a critical condition often diagnosed using chest radiographs.
- Accurate grading of pulmonary edema severity is essential for effective patient management.
- Existing methods for pulmonary edema classification may have limitations in speed and accuracy.
Purpose of the Study:
- To develop and evaluate machine learning models for classifying pulmonary edema severity on chest radiographs.
- To compare the performance of semisupervised and supervised deep learning models.
Main Methods:
- Retrospective analysis of 369,071 chest radiographs from the MIMIC-CXR dataset.
- Development of a semisupervised model using a variational autoencoder and a pretrained supervised model using a dense neural network.
- Classification of pulmonary edema into four ordinal levels: no edema, vascular congestion, interstitial edema, and alveolar edema.
Main Results:
- The semisupervised model achieved an AUC of 0.99 for differentiating alveolar edema from no edema.
- Performance varied with edema severity, with higher accuracy for more severe cases.
- Both models demonstrated high performance, with the semisupervised model generally outperforming the pretrained model.
Conclusions:
- Deep learning models can effectively grade pulmonary edema severity from chest radiographs.
- The developed models show high performance on a large dataset.
- These findings support the potential of AI in improving pulmonary edema diagnosis.
