Explainable COVID-19 Detection Based on Chest X-rays Using an End-to-End RegNet Architecture

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Diseases

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Chest X-rays (CXRs) are vital for assessing respiratory conditions.
• Early identification of COVID-19 is crucial for controlling its spread.

Purpose of the Study:

• To validate and test a deep learning model for COVID-19 detection using CXR images.
• To adapt and train the RegNetX032 convolutional neural network (CNN) for this diagnostic task.
• To evaluate the model's performance against RT-PCR reference standards.

Main Methods:

• A deep convolutional neural network (CNN) model, RegNetX032, was customized and trained on over 15,000 CXR images from five datasets.
• The model was tested on an independent dataset from Montfort Hospital.
• Performance was assessed using metrics including Area Under the Curve (AUC), sensitivity, and specificity, with multi-binary classifications.

Main Results:

• The fine-tuned RegNetX032 model achieved 96.0% accuracy and 99.1% AUC for COVID-19 detection, with 98.0% sensitivity and 93.0% specificity.
• In classifying COVID-19 with pneumonia versus normal, the model achieved 99.1% AUC, 96.0% sensitivity, and 93.0% specificity.
• Validation sets showed high average accuracy (98.6%) and AUC (98.0%), demonstrating robust generalization.

Conclusions:

• The deep learning model demonstrates excellent performance and generalization for detecting COVID-19 from chest X-rays.
• This AI tool can automate COVID-19 detection, assisting in patient triage and isolation decisions.
• The model serves as a valuable complementary aid for radiologists and clinicians in diagnosing COVID-19.