COVID-19 diagnosis from chest x-rays: developing a simple, fast, and accurate neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Diseases

Background:

• Chest X-rays are a rapid, cost-effective diagnostic tool.
• Current limitations include low accuracy and difficulty distinguishing COVID-19 from other pneumonias.
• Deep learning, specifically CNNs, offers potential for improved diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate a CNN model for diagnosing COVID-19 using chest X-ray images.
• To assess the model's ability to differentiate COVID-19 from normal lungs and other viral pneumonias.

Main Methods:

• Utilized the COVID-19 Radiography database containing 15,153 X-ray images.
• Developed a CNN by adding a dense layer to a pre-trained EfficientNetB0 model.
• Employed data augmentation and fine-tuning for model optimization and validation.

Main Results:

• The CNN achieved 95% accuracy, 90% sensitivity, and 97% specificity in differentiating COVID-19 from normal lungs.
• The model differentiated COVID-19 from other viral pneumonias and normal lungs with 93% accuracy, 94% sensitivity, and 95% specificity.

Conclusions:

• A parsimonious CNN can accurately differentiate COVID-19 from other conditions on chest X-rays.
• This AI-driven approach can aid clinicians in diagnosis and support chest X-rays as a valuable early screening tool.