Pneumonia and COVID-19 Detection from Chest X-Ray Images Using Multiple DCNN Models

Coronavirus has been the most prominent global plague since the 1920s Bubonic plague. The patients of Coronavirus often develop pneumonia as one of the first symptoms. Doctors suggest that it is reliable to test pneumonia to test the COVID-19. Hence, the task undertaken in the project is to diagnose pneumonia in the patient with the help of Chest X-ray images. This task currently takes the help of an expert who does this manually, which takes a significant amount of time. A lot of delays are created in case of an outbreak as it happened during the COVID-19 spread. Thus, we need an automated system that can process the X-Ray images and produce results within seconds and handle the situation where many cases emerge together. This eradicates the need for a medical expert in every region, especially the underdeveloped or difficult places to reach. This research uses Deep Convolutional Neural Networks for achieving this task. Specifically, this approach uses a fusion of Squeeze and Excitation - ResNeXt and InceptionV3 architectures for classifying the input Chest X-Ray images into three classes – Uninfected (Normal), Pneumonia, and COVID with pneumonia. This service is made available with the help of a web-based application that helps users upload Chest X-Rays and display the results to the user. From the results of the experiments, it was found that the SE-ResNext and InceptionV3 ensemble model outperforms these models when tested individually. The proposed model achieved 96.04% accuracy (ACC), 0.954 precision, 0.956 recall, and an F1Score of 0.955 based on tests done on the Chest X-Ray images test dataset.