Deep Learning Models for Early Discovery of Covid-19 with Radiology Modalities

Due to the COVID-19 pandemic, the whole world is experiencing a health catastrophe that is unprecedented in its scope. Since coronavirus spreads rapidly, investigators are worried about finding or assisting in the development of treatments to save lives and reduce the pandemic. The key issues in the present COVID-19 situation are the initial identification and diagnosis of COVID-19, as well as the precise parting of non-COVID-19 patients in cost-effective methods during the initial period of the illness. Artificial intelligence (AI), for example, has been modified to solve the issues posed by pandemics. Deep learning (DL) models’ computation power has aided healthcare procedures to be more rapid, accurate, and well-organized. DL networks are revolutionizing patient care, and they play a crucial role in clinical practice for health systems. DL approaches in healthcare include computer vision, Natural language processing (NLP), and fortification knowledge. DL is a strategy to tackle the COVID-19 outbreak because there are so many bases of medicinal pictures (e.g., X-ray, CT, and MRI). As a result of this statistic, numerous studies have been projected and produced for the first months of 2020. We create a DL method to extract features and identify COVID-19 from Radiology Modalities in this chapter. In spite of their widespread use in diagnostic centers, diagnostic procedures created on radiological examinations have flaws when it comes to the disease's uniqueness. As a result, DL models are commonly used to evaluate radiological pictures in order to identify the disease in the early stage. AI, notably DL, is being used to solve this challenge. A novel approach of entailing FJCovNet2 which is a DL method built on DenseNet121 to detect COVID-19 using CT-Scan and X-Ray pictures is effectuated. The comparative study of various forms of radiology modalities in deep learning determines the most accurate method to detect the disease earlier using the predefined models namely InceptionV3, ResNet50, and VGG 16. The maximum accuracy of 98.23% is attained through the proposed model RJCovNet2.