Covid-19 and Pneumonia Detection from Chest X-Ray Images by Deep Learning Model

This chapter presents an overview of automatic Covid-19 and Pneumonia detection from Chest X-Ray (CXR) Images by using Deep Learning models. More specifically, in this chapter, a deep analysis of existing Convolution Neural Network (CNN) models (for Covid-19 detection), is presented from the perspective of data imbalance. Covid-19 disease had a disastrous effect on human life, since 2020. One of the effective ways of diagnosing Covid-19 disease is to detect it from Chest X-Ray (CXR) images, because it is observed that Covid-19 may damage the human respiratory system (or, lung region). Hence, the impact changes due to Covid are very prominent on the Chest X-Ray (CXR) images. This chapter is further divided into two types of Covid-19 detection categories: (I) Covid-19 and Pneumonia detection by pre-trained CNN models, (II) A framework for alleviating class imbalance problem from the CXR dataset: Here, the architecture of the standard CNN model is not changed, but modified (or, weighted) loss function or, Ensemble models or, Re-sampling techniques are deployed in order to overcome the class imbalance problem. Furthermore, some of the existing CNN model's results are implemented and analyzed on an imbalanced (and readily available) CXR dataset in which there are four classes: (a) Covid, (b) Normal, (c) Viral Pneumonia (VP), (d) Lung Opacity (LO). Experimental results reveal that, Xception model has outperformed other existing pre-trained models. Furthermore, experimental results suggest that, Mobile Freeze-Net with Attention-based loss function, not only provides the highest accuracy on this CXR dataset, but also, has improved its generalization ability considerably.