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

Tang, H.; Sun, N.; Li, Y.

The pandemic of Covid-19 has caused tremendous losses to lives and economy in the entire world. Up until October 2020, it has caused more than 38 million infections and 1.1 million deaths. This has created a severe burden for the health care system worldwide. 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. The recently published dataset of a collection of radiological images for a rural population in United States had made development of such a model a possibility due to the high quality of the radiological images and the consistency in clinical measurements. We manually annotated 221 chest X-ray images with lung fields and opacity regions and trained a segmentation model for the opacity region. The model has a good performance in regarding the overlap between predicted and manually labelled opacity regions for both the testing data set and the validation dataset from very different sources. In addition, the percentage of the opacity region over the area of the total lung fields shows a good predictive power for the patient severity. In view of the above, our model is a successful first try in developing a segmentation model for the opacity regions for the Covid-19 positive chest X-rays. However, careful manual examinations of the model predictions by experienced radiologists show mistakenly predicted opacity regions caused probably by the anatomical complexities. Thus, additional work is needed before a robust and accurate model can be developed for the ultimate goal of implementation in the clinical setting. The model, manual segmentation and other supporting materials can be found in https://github.com/haimingt/opacity_segmentation_covid_chest_X_ray.