Deep-COVID: Predicting COVID-19 from chest X-ray images using deep transfer learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Diseases

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Chest radiography is a key method for identifying COVID-19 related abnormalities.
• Early detection is crucial for patient management and limiting disease spread.

Purpose of the Study:

• To investigate the efficacy of deep learning models for detecting COVID-19 from chest X-ray images.
• To evaluate the performance of popular convolutional neural networks (CNNs) in identifying COVID-19 indicators.

Main Methods:

• A dataset of 5000 chest X-rays was curated, with COVID-19 cases identified by radiologists.
• Transfer learning was applied to train ResNet18, ResNet50, SqueezeNet, and DenseNet-121 models.
• Model performance was evaluated using sensitivity, specificity, ROC curves, and heatmaps.

Main Results:

• Most trained CNN models achieved a sensitivity of 98% (±3%) and a specificity of approximately 90%.
• Generated heatmaps highlighted potentially infected lung regions, correlating with radiologist annotations.
• The study provides a publicly available dataset and model implementations.

Conclusions:

• Deep learning models demonstrate high potential for accurate and rapid COVID-19 detection using chest X-rays.
• Further validation on larger datasets is recommended for robust accuracy assessment.
• The developed tools can support clinical decision-making in pandemic scenarios.