CLASSIFICATION OF COVID-19 FROM CHEST X-RAY IMAGES USING RESQNET

The infectious SARS-COV2 or COVID-19 is one of the potentially harmful and deadliest virus for human beings. The infectious coronavirus disease has seriously impacted the health of humans, severely affecting the respiratory system. The viral COVID-19 pandemic has impacted the lives and health of several communities. This extremely contagious disease is spreading quickly around the world; thus, it is critical to diagnose it early and accurately. Medical imaging technologies are widely used by radiologists to see complex structures and how the body functions. An accurate diagnosis can help detect the COVID-19 virus early on, especially in cases where the patient shows no symptoms. Therefore, it is necessary to build effective methods with artificial intelligence right away to predict COVID-19. This paper presents a novel approach that utilizes chest X-ray images for COVID-19 detection. This approach offers a framework for using the many kinds of deep features, maintaining complementary and correlated information, and achieving precise detection throughout the classification stage. In the proposed approach, a model named ReSqNet50 is implemented for identification. The four-class classifications included in the dataset and utilized to train the model were confirmed COVID-19 cases, Normal (NO) (healthy lung) individuals, Lung Opacity (LO) cases, and Viral Pneumonia (VP) cases. The dataset was gathered from the Kaggle database, a publicly available data source found at . Before the dataset was fed into the suggested model, the aggregated dataset was initially pre-processed and then augmented. Extensive analysis of the model was done utilizing the dataset of COVID-19 and the accuracy was compared with state-of-the-art models. The objective of the study is to detect infected COVID-19 individuals from various classes like VP, LO, and NO (Healthy Lung) individuals. Although the dataset was imbalanced, the proper balanced dataset is considered with various changes in the image sizes as well as proper image boosting. For four-class classification (COVID-19/Normal/Lung Opacity/Viral Pneumonia), our model achieved an accuracy of 96.9%. This promising accuracy offers radiologists a novel and potentially valuable tool to help them quickly diagnose COVID-19 patients using simple and readily available equipment.