COVID-19 diagnosis from chest X-ray images using transfer learning: Enhanced performance by debiasing dataloader

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Disease Detection
• Radiology and Diagnostic Imaging

Background:

• Chest X-ray imaging is a vital tool for diagnosing COVID-19 due to its accessibility and speed.
• Current diagnostic methods can be time-consuming, highlighting the need for efficient screening tools.

Purpose of the Study:

• To enhance the efficacy of COVID-19 screening using chest X-ray images.
• To develop and validate a deep convolutional neural network (CNN) model, nCoV-NET, for accurate COVID-19 detection.

Main Methods:

• Trained and evaluated nCoV-NET using three diverse datasets comprising COVID-19 pneumonia and non-COVID-19 cases.
• Employed transfer learning with re-trained ResNet, DenseNet, and VGG architectures to address class imbalance.
• Utilized Activation Mapping to visualize and interpret the model's decision-making process.

Main Results:

• The nCoV-NET model, optimized with DenseNet-161, achieved a high classification accuracy of 97.1% for COVID-19 cases.
• All re-trained architectures demonstrated comparable performance, with nCoV-NET showing superior results.
• Activation maps provided insights into the critical regions identified by the model in chest X-rays.

Conclusions:

• The developed nCoV-NET model reliably detects COVID-19 using chest X-ray images.
• This AI-powered approach can accelerate patient triaging and support radiologists in diagnosis.
• nCoV-NET offers a valuable tool for efficient disease control and management.