Deep Learning Algorithm for COVID-19 Classification Using Chest X-Ray Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Early diagnosis of SARS-CoV-2 is crucial for controlling COVID-19 spread.
• Chest X-ray (CXR) imaging offers reliable disease classification, complementing RT-PCR.
• Deep learning models like CNNs show promise for COVID-19 detection but require extensive training data.

Purpose of the Study:

• To propose a novel deep learning model integrating CNNs and DCGANs for classifying CXR images into normal, pneumonia, and COVID-19 categories.
• To address the challenge of limited training data for COVID-19 detection models.
• To enhance the accuracy and generalization of COVID-19 classification using CXR.

Main Methods:

• Developed a hybrid CNN-DCGAN model with eight convolutional layers, four max-pooling layers, and two fully connected layers.
• Utilized DCGAN for generating synthetic CXR images to augment imbalanced datasets and extract deep features.
• Trained and evaluated the proposed model on four diverse public CXR datasets.

Main Results:

• The proposed CNN model, enhanced with DCGAN-generated synthetic images, achieved high classification accuracies (94.8% to 98.6%) across four datasets.
• The DCGAN-CNN approach outperformed existing pretrained models like AlexNet and GoogLeNet.
• DCGAN effectively enlarged the dataset and improved model generalization.

Conclusions:

• The proposed DCGAN-CNN approach demonstrates a promising and efficient solution for the automated diagnosis of COVID-19 from chest X-ray images.
• The integration of generative adversarial networks significantly enhances the performance of deep learning models in medical image classification.
• This hybrid model offers a robust tool for supporting clinical decision-making in COVID-19 diagnosis.