The Deep Learning-Based Framework for Automated Predicting COVID-19 Severity Score.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Disease Research

Background:

• The COVID-19 pandemic necessitates research into disease severity prediction.
• Diagnostic chest imaging offers a rapid method for assessing COVID-19 severity.

Purpose of the Study:

• To develop a deep learning model for automated prediction of COVID-19 severity using chest X-rays.
• To assist clinicians in developing patient treatment strategies and monitoring disease progression.

Main Methods:

• A two-phase deep learning approach was employed, involving feature extraction and outcome prediction.
• A DenseNet backbone network extracted 18 lung disease-related features from chest X-rays (CXRs).
• An MLP regression model selected key features for severity score prediction, validated against standard regression models.

Main Results:

• The model achieved a Mean Absolute Error (MAE) of 1.02 for geographic extent score prediction.
• The model achieved an MAE of 0.85 for lung opacity score prediction.
• Performance was evaluated on a dataset comprising 2373 CXRs.

Conclusions:

• The developed deep learning model effectively predicts COVID-19 severity from chest X-rays.
• This AI-driven approach shows promise for clinical decision support in managing COVID-19 patients.