COVID-19 detection and disease progression visualization: Deep learning on chest X-rays for classification and coarse...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Disease Diagnosis

Background:

• Chest X-rays are crucial for COVID-19 diagnosis.
• Limited labeled data hinders automated classification of COVID-19 chest X-rays.
• Distinguishing COVID-19 from other lung inflammations like pneumonia is challenging.

Purpose of the Study:

• To implement a transfer learning pipeline for automated COVID-19 detection from chest X-rays.
• To improve classification accuracy despite limited data by using data augmentation.
• To provide visual explanations for model predictions and monitor disease progression.

Main Methods:

• Utilized transfer learning with pre-trained convolutional neural networks (VGG16, ResNet50, EfficientNetB0) as feature extractors.
• Employed a generative adversarial network (CycleGAN) for data augmentation of the COVID-19 class.
• Applied Gradient Class Activation Mapping (Grad-CAM) for visual interpretability.

Main Results:

• Achieved high detection accuracies: 90% (VGG16), 94.3% (ResNet50), and 96.8% (EfficientNetB0).
• Successfully distinguished COVID-19 and pneumonia from normal cases.
• Generated visualizations highlighting key regions for accurate predictions.

Conclusions:

• The developed transfer learning pipeline offers a robust solution for automated COVID-19 classification from chest X-rays.
• Data augmentation and interpretability techniques enhance the reliability and clinical utility of the model.
• Visualizations aid in understanding disease impact and monitoring progression.