COVID-19 Pneumonia Diagnosis Using Medical Images: Deep Learning-Based Transfer Learning Approach

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Disease Diagnostics

Background:

• SARS-CoV-2 (COVID-19) continues to pose a global health threat due to its transmissibility and evolving variants.
• Rising global test positivity (11% as of Feb 2025) highlights ongoing challenges in diagnostics despite widespread vaccination.
• Newer variants exhibit increased infectivity and diagnostic complexity due to enhanced host cell binding.

Purpose of the Study:

• To evaluate deep transfer learning for rapid, accurate, and mutation-resilient COVID-19 diagnosis using medical imaging.
• To assess the scalability and accessibility of AI-driven diagnostic tools.

Main Methods:

• Developed an automated detection system using convolutional neural networks (CNNs).
• Evaluated models including VGG16, ResNet50, ConvNeXtTiny, MobileNet, NASNetMobile, and DenseNet121.
• Utilized chest X-ray and computed tomography (CT) images for COVID-19 detection.

Main Results:

• DenseNet121 demonstrated superior performance, achieving 98% accuracy.
• Key performance metrics for DenseNet121 included 96.9% precision, 98.9% recall, 97.9% F1-score, and 99.8% AUC.
• The model exhibited minimal false positives/negatives, indicating robustness for clinical deployment.

Conclusions:

• AI-powered diagnostics hold significant potential for early disease detection and pandemic response.
• Optimized deep learning models can bridge testing gaps, especially in resource-limited settings or with new variants.
• DenseNet121 serves as a benchmark for future AI diagnostic tool development.