Automatic detection of COVID-19 from chest radiographs using deep learning

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Infectious Disease Diagnostics

Background:

• The COVID-19 pandemic necessitates rapid and scalable diagnostic methods.
• Conventional PCR testing struggles to meet the demand for widespread COVID-19 detection.
• Timely isolation of infected individuals is crucial for controlling disease spread.

Purpose of the Study:

• To develop a deep neural network model for expedited COVID-19 detection using chest radiographs.
• To evaluate the efficacy of a non-contact, image-based diagnostic approach.
• To assess the potential of transfer learning with the VGG-16 model for classifying COVID-19 cases.

Main Methods:

• A deep neural network, specifically the VGG-16 model, was employed for image classification.
• Transfer learning with fine-tuning was utilized to train the model on a dataset of 1428 chest radiographs.
• The dataset included confirmed COVID-19 positive, bacterial pneumonia, and healthy cases.

Main Results:

• The model achieved promising results in expediting COVID-19 detection.
• An accuracy of 96% was obtained for a two-class classification (COVID-19 vs. non-COVID-19).
• An accuracy of 92.5% was achieved for a three-class classification (COVID-19, bacterial pneumonia, healthy).

Conclusions:

• The proposed deep neural network model serves as a potential tool for initial COVID-19 screening.
• Its simplicity, ability to work with small datasets, and non-contact nature make it a viable alternative for rapid testing.
• This approach can assist healthcare professionals in timely disease detection and patient management.