Modeling a deep transfer learning framework for the classification of COVID-19 radiology dataset

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• The COVID-19 pandemic, caused by SARS-CoV-2, necessitates rapid and accurate diagnostic tools.
• Chest X-rays are a crucial imaging modality for diagnosing respiratory illnesses, including COVID-19.
• Developing automated systems for analyzing X-ray images can significantly aid in early detection and management.

Purpose of the Study:

• To evaluate the effectiveness of Deep Transfer Learning (DTL) models for classifying COVID-19 from chest X-ray images.
• To compare the performance of fine-tuned VGG-16 and VGG-19 Convolutional Neural Networks (CNNs) in binary and three-class classification scenarios.
• To assess the diagnostic accuracy and reliability of the proposed DTL models.

Main Methods:

• Utilized a real-life dataset of chest X-ray images for training and validation.
• Employed fine-tuned VGG-16 and VGG-19 CNNs integrated with DTL for image classification.
• Trained models using Adam optimizer, categorical cross-entropy loss, batch size of 10, over 40 epochs.
• Performed binary (COVID-19 vs. Normal) and three-class (COVID-19 vs. Viral Pneumonia vs. Normal) classifications.

Main Results:

• The VGG-16 DTL model achieved 99.23% accuracy in binary classification and 93.85% in three-class classification.
• The VGG-19 DTL model achieved 98.00% accuracy in binary and 92.92% in three-class classification.
• Both models showed high Matthews Correlation Coefficient (MCC) and Kappa values, indicating strong performance and agreement between predictions and true labels, with VGG-16 outperforming VGG-19.
• The best-performing VGG-16 DTL model achieved 98% accuracy on an independent, unlabeled test dataset.

Conclusions:

• The fine-tuned VGG-16 based DTL model is highly effective for the accurate classification of COVID-19 from chest X-ray images.
• DTL models, particularly VGG-16, offer a promising approach for automated COVID-19 detection, potentially outperforming existing methods.
• The study validates the utility of deep learning in medical image analysis for pandemic response and diagnostics.