Graph-Based COVID-19 Detection Using Conditional Generative Adversarial Network

Coronavirus (SARS-CoV-2) is a novel global pandemic, which requires rapid and accurate identification techniques to curb its spread. COVID-19, the disease induced by the virus, causes severe respiratory complications, necessitating advanced diagnostic tools for early detection. Recent research indicates the potential of radiographic imaging in unravelling critical insights into the characteristics of this formidable pathogen. Leveraging the advancements in Computer Vision (CV) and deep learning methodologies, an automated system can be devised to discern respiratory anomalies from X-ray images, enhancing conventional diagnostic methods. In this study, we propose a pioneering approach for COVID-19 diagnosis utilizing chest radiographs. The proposed methodology encompasses four distinct phases: initial segmentation of raw chest radiographs employing Conditional Generative Adversarial Networks (CGAN), followed by feature extraction through a tailored pipeline integrating both manual computer vision algorithms and pre-trained Deep Neural Network (DNN) models. Subsequently, a graph-based feature reconstruction technique amalgamates these extracted features across the network, culminating in a comprehensive representation. These reconstructed features serve as input to a classification module, comprising a multi-layer neural network, GCN, adept at processing graph-structured data, alongside conventional machine learning classifiers such as Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and Random Forest (RF), facilitating categorization of chest X-ray images into COVID-19, pneumonia, and normal cases. Furthermore, we conduct an exhaustive evaluation of the selected DNN architectures to ascertain the efficacy of our proposed models vis-à-vis existing research, thus ensuring the deployment of the most robust diagnostic framework.