A deep convolution neural network for automated COVID-19 disease detection using chest X-ray images

Abstract

COVID-19 is a virus that can cause severe pneumonia, and the severity varies based on the patient's immune system. The rapid spread of the disease can be mitigated through automated detection, addressing the shortage of radiologists in medicine. This paper introduces the Modified-Inception V3 (MIn-V3) model, which utilizes feature fusion from the internal layers of Inception V3 to classify different diseases, including normal cases, COVID-19 positivity, viral pneumonia, and bacterial pneumonia. Additionally, transfer learning and fine-tuning techniques are applied to enhance accuracy. The performance of MIn-V3 is assessed by comparing it with pre-trained Deep Learning (DL) models, such as Inception-ResNet V2 (InRN-V2), Inception V3, and MobileNet V2. Experimental results demonstrate that the MIn-V3 model surpasses other pre-trained models with a classification accuracy of 96.33 %. Furthermore, integrating the MIn-V3 model into a mobile application enables rapid and accurate detection of COVID-19, thus playing a crucial role in advancing early diagnostics, which is essential for timely intervention and effective disease management.

Keywords

• Deep convolution neural network
• Transfer learning
• COVID-19
• Feature fusion
• Multi-class classification