Deep learning segmentation model for automated detection of the opacity regions in the chest X-rays of the Covid-19 positive patients and the application for disease severity
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (ScreenIT)
Abstract
Purpose
The pandemic of Covid-19 has caused tremendous losses to lives and economy in the entire world. The machine learning models have been applied to the radiological images of the Covid-19 positive patients for disease prediction and severity assessment. However, a segmentation model for detecting the opacity regions like haziness, ground-glass opacity and lung consolidation from the Covid-19 positive chest X-rays is still lacking.
Methods
The recently published collection of the radiological images for a rural population in United States had made the development of such a model a possibility, for the high quality images and consistent clinical measurements. We manually annotated 221 chest X-ray images with the lung fields and the opacity regions and trained a segmentation model for the opacity region using the Unet framework and the Resnet18 backbone. In addition, we applied the percentage of the opacity region over the area of the total lung fields for predicting the severity of patients.
Results
The model has a good performance regarding the overlap between the predicted and the manually labelled opacity regions. The performance is comparable for both the testing data set and the validation data set which comes from very diverse sources. However, careful manual examinations by experienced radiologists show mistakes in the predictions, which could be caused by the anatomical complexities. Nevertheless, the percentage of the opacity region can predict the severity of the patients well in regards to the ICU admissions and mortality.
Conclusion
In view of the above, our model is a successful first try in the development of a segmentation model for the opacity regions for the Covid-19 positive chest X-rays. However, additional work is needed before a robust model can be developed for the ultimate goal of the implementations in the clinical setting.
Model and supporting materials can be found in https://github.com/haimingt/opacity_segmentation_covid_chest_X_ray .
Article activity feed
-
-
SciScore for 10.1101/2020.10.19.20215483: (What is this?)
Please note, not all rigor criteria are appropriate for all manuscripts.
Table 1: Rigor
NIH rigor criteria are not applicable to paper type.Table 2: Resources
Software and Algorithms Sentences Resources 2.5 Model architecture and training: “Segmentation models” is a python library with neural networks for image segmentation based on Keras and Tensorflow[35]. pythonsuggested: (IPython, RRID:SCR_001658)Results from OddPub: We did not detect open data. We also did not detect open code. Researchers are encouraged to share open data when possible (see Nature blog).
Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:The clear delimitation of the abnormality boundaries are usually difficult, especially for lighter representations like …
SciScore for 10.1101/2020.10.19.20215483: (What is this?)
Please note, not all rigor criteria are appropriate for all manuscripts.
Table 1: Rigor
NIH rigor criteria are not applicable to paper type.Table 2: Resources
Software and Algorithms Sentences Resources 2.5 Model architecture and training: “Segmentation models” is a python library with neural networks for image segmentation based on Keras and Tensorflow[35]. pythonsuggested: (IPython, RRID:SCR_001658)Results from OddPub: We did not detect open data. We also did not detect open code. Researchers are encouraged to share open data when possible (see Nature blog).
Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:The clear delimitation of the abnormality boundaries are usually difficult, especially for lighter representations like ground glass opacity or haziness. Secondly, some anatomical structures may compromise the model predictions. Figure 5.b shows that the chest X rays have opacities in both lung bases, however, the model prediction also contains a small region in the right lung apex. However, this is a wrong prediction probably caused by the rib shadow. Figure 5.c shows several tiny opacities in bilateral lung bases. However, the model may mistakenly predict the high density as an opacity. The high density is more likely due to the overlap of lung tissues instead of the ground glass opacities or consolidations caused by infections and exudates. Thirdly, as shown in Figure 5.d, the model may mistake the markings of bronchial trees for opacities. The lungs in Figure 5.d are actually clear, but the markings of bronchial trees are prominent. They were likely to have been mistakenly predicted as opacities due to the higher densities.
Results from TrialIdentifier: No clinical trial numbers were referenced.
Results from Barzooka: We did not find any issues relating to the usage of bar graphs.
Results from JetFighter: We did not find any issues relating to colormaps.
Results from rtransparent:- Thank you for including a conflict of interest statement. Authors are encouraged to include this statement when submitting to a journal.
- Thank you for including a funding statement. Authors are encouraged to include this statement when submitting to a journal.
- No protocol registration statement was detected.
-