SpinalXNet: Transfer Learning with Modified Fully Connected Layer for X-Ray Image Classification

Abstract

Over the past year, COVID-19 has become a global pandemic and people across the globe have suffered a lot from this pandemic. The rate of transmitting the coronavirus in people is very quick. A rapid diagnosis can potentially help governments in identifying the pattern of transmission. There are some tests available but those tests take a long time to give the report. So, in this work, we have proposed a model that will distinguish between normal people, COVID affected people, and pneumonia affected people with the help of an X-ray. X-ray images are considered because taking an X-ray image is very little time-consuming. In this work, we have trained the X-ray images with a novel Deep Learning approach with SpinalNet architecture, and that distinguishes normal people, COVID affected people, and pneumonia affected people. After training the model we have achieved a very good accuracy for the SpinalNet architecture that is 96.12% while the traditional model provides 95.50% accuracy. We present precision, recall, and F1 scores of COVID and Pneumonia classes. We also present our results and codes with execution details. This paper contains the link to Kaggle notebooks with execution details. The applied Spinalnet transfer learning code is available in our GitHub repository: https://github.com/dipuk0506/SpinalNet