Medical X-Ray Image Classification Employing DCGAN and CNN Transfer Learning Techniques

Abstract

Over the decades, a typical imaging test that has been used is an X-ray. It allows doctors to see into the body without an incision. As a result, an X-ray can aid in diagnosing, monitoring, and treating a variety of medical disorders by detecting diseases beforehand. Among the diseases, pneumonia got major heed because of its intensity. As the lungs are the most vulnerable part of the body when it comes to pneumonia, doctors rely on the chest X-ray to diagnose the disease. In this research, we have worked on the X-ray images to discern pneumonia using our custom CNN model and different types of transfer learning models and manifested a comparison of those methods in terms of their ability to detect the disease. Furthermore, we performed generative adversarial networks (GAN) with deep convolutional layers to generate and merge a new training dataset using existing image data. Then, we executed the models anew after acquiring a new artificial dataset. Before using GAN, we got accuracy of 94%, 94%, 73%, 73%, 96%, 97%, and 94% in Custom CNN, InceptionV3, ResNet50, EfficientNetB0, VGG16, DenseNet201, and Xception, respectively. However, we observed improved accuracy from all models applying GAN except for DenseNet201. Moreover, VGG16, DenseNet201, and custom CNN acquired the higher accuracy overall.