COVID-19 Severity Prediction from Chest X-ray Images Using an Anatomy-Aware Deep Learning Model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• COVID-19 pandemic strains hospital patient management systems globally.
• Radiological imaging, particularly chest X-rays (CXRs), is crucial for assessing COVID-19 severity.
• Automated assessment is vital due to increasing patient numbers and radiologist shortages.

Purpose of the Study:

• To develop an anatomy-aware (AA) deep learning model for COVID-19 severity prediction from CXR images.
• To address limitations of previous models that did not explicitly consider anatomical attributes.
• To provide an effective tool for low-resource settings and areas with limited access to skilled radiologists.

Main Methods:

• Proposed an anatomy-aware (AA) deep learning model incorporating lung segmentation masks.
• The model learns generic features from X-ray images, considering anatomical information.
• Utilized four open-source datasets and an in-house annotated test set for training and evaluation.

Main Results:

• The AA model improved the geographical extent score by 11% (MSE).
• The model maintained benchmark performance for lung opacity scoring.
• Demonstrated effectiveness in predicting COVID-19 severity from chest X-ray images.

Conclusions:

• The proposed AA deep learning model accurately predicts COVID-19 severity from CXR.
• The model's ability to consider anatomical information enhances prediction accuracy.
• This approach offers a valuable tool for COVID-19 severity assessment in resource-limited healthcare environments.