Abstract

The COVID-19 pandemic has emphasized the urgent need for accurate and timely detection of pneumonia, which is commonly associated with the virus. In recent years, various studies have been conducted in the field of artificial intelligence to develop precise neural network models for identifying pathologies. However, the lack of medical data for research purposes is a significant challenge that impedes the development of deep learning models with high accuracy. To address the issue of sparse data, several techniques have been employed, such as using various augmentation techniques to increase the image volume and utilizing transfer learning models for similar classification tasks. These pre-trained models are trained over millions of images, and this knowledge can be transferred to similar problems. To tackle the problem of overfitting when training a small dataset, we suggest a new approach that involves fine-tuned transfer learning (TL) using the VGG16 model architecture. Additionally, we introduce a data augmentation technique that incorporates generative adversarial networks (GANs). GANs generate realistic synthetic images, enriching the training dataset and providing better performance than traditional methods. Furthermore, we compared traditional data augmentation techniques with GAN-based augmentation, and the comparative analysis underlines the effectiveness of GANs.Experimental results confirm the validity of the proposed methodology, achieving an impressive 99% accuracy. We believe that our proposed approach offers a valuable solution to the challenge of limited medical data while enhancing the accuracy of pneumonia classification models.