Predicting COVID-19 Pneumonia Severity on Chest X-ray With Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonology

Background:

• Coronavirus disease-19 (COVID-19) patient management requires efficient monitoring of disease progression.
• Chest X-rays (CXRs) offer a non-invasive method for assessing lung involvement in COVID-19 pneumonia.
• Accurate severity assessment is crucial for timely clinical decision-making and resource allocation.

Purpose of the Study:

• To develop and validate a predictive model for COVID-19 pneumonia severity using frontal chest X-ray images.
• To create a tool that quantifies lung involvement and opacity for objective severity scoring.
• To support clinical decisions regarding patient care escalation/de-escalation and treatment efficacy monitoring.

Main Methods:

• Utilized a public COVID-19 chest X-ray database for model training and evaluation.
• Employed three blinded expert radiologists to retrospectively score lung involvement and opacity.
• Leveraged a pre-trained neural network on large chest X-ray datasets to extract predictive image features.

Main Results:

• The developed regression model achieved a Mean Absolute Error (MAE) of 1.14 for geographic extent score (0-8).
• The model demonstrated an MAE of 0.78 for lung opacity score (0-6).
• Feature extraction from a pre-trained model proved effective for predicting COVID-19 severity.

Conclusions:

• The predictive model accurately gauges COVID-19 lung infection severity from chest X-rays.
• This tool can assist in optimizing patient care pathways and monitoring therapeutic responses, particularly in critical care settings.
• The study's code, labels, and data are publicly available to facilitate further research and development.