An Efficient Transferred Cascade System for COVID-19 Detection from Chest X-ray Images

Abstract

Analysing x-ray images for detecting Covid-19 presents one cost-effective approach. To automate this task, deep learning techniques have been suggested to reduce doctors workload. However, existing datasets classify X-ray images into three categories: Normal, Pneumonia, and COVID-19, but it is crucial to differentiate
between bacterial and viral pneumonia due to their distinct treatment approaches. This paper introduces three novel cascade systems designed to distinguish between COVID-19 and non-COVID-19 pneumonia, as well as to classify bacterial and viral pneumonia, using a newly compiled dataset. The
proposed Transferred Cascade Convolutional Neural Network (TCCNN) system enables the model to efficiently recognize complex concepts by combining various convolutional neural networks in two or three stages. Furthermore, TCCNN incorporates transfer learning within the cascade structure, allowing each
convolutional neural network to exploit the trained model from the previous stage. The comparative analysis demonstrated the efficiency of the proposed systems, where the two-stage PN_CBV system achievd an accuracy of 96.27% using the DenseNet201_DenseNet121 combination.