A Novel COVID-19 Detection Model Based on DCGAN and Deep Transfer Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• The global outbreak of Coronavirus disease (COVID-19) necessitates accurate and efficient diagnostic tools.
• Medical image datasets, particularly for rare diseases, often suffer from class imbalance, hindering deep learning model performance.
• Traditional data augmentation may be insufficient for limited medical imaging datasets.

Purpose of the Study:

• To develop a novel data generation model for creating synthetic COVID-19 X-ray images.
• To address the challenge of unbalanced datasets in deep learning for medical image analysis.
• To enhance the performance of deep learning models for COVID-19 detection using augmented data.

Main Methods:

• Implementation of a Deep Convolutional Generative Adversarial Network (DCGAN) for synthetic data generation.
• Evaluation of generated data quality using Fréchet Inception Distance (FID), achieving a score of 23.78.
• Utilizing deep transfer learning models (VGG-16, Inceptionv3, MobilNet) as backbones for COVID-19 detection.

Main Results:

• The DCGAN model successfully generated synthetic X-ray images with properties comparable to original data.
• The FID score of 23.78 indicates high fidelity of the generated images.
• The study demonstrated the potential of DCGAN-generated data to improve classifier performance.

Conclusions:

• DCGAN is a viable technique for augmenting limited medical imaging datasets, specifically for COVID-19 detection.
• Addressing data imbalance through synthetic data generation can lead to more robust and accurate deep learning models.
• This approach offers a promising solution for improving diagnostic capabilities in the context of global health crises.