COVID-19 Diagnosis from Chest X-Ray Images Using Convolutional Neural Networks and Effects of Data Poisoning

Menon, K.; Bohra, V. K.; Murugan, L.; Jaganathan, K.; Arumugam, C.

At the end of 2019, a new type of virus called SARS-CoV-2 began spreading resulting in a global pandemic. As of June 2021, almost 175 million people were affected worldwide. Symptom-wise, it is very difficult to diagnose if a person has Covid or just a viral infection. But, taking a close look at chest X-Rays is extremely helpful in the diagnostic process. The proposed methodology in this paper helps in classification of chest X-Ray images into 3 categories: ‘Covid’, ‘Viral’ and ‘Normal’. The dataset was created by integrating 3 pre-existing evergrowing datasets and the ResNet-18 model was adopted to train it. The experimental results show that the classification of the chest X-Ray images was done with an accuracy of 0.9648. An adversarial machine learning approach was employed to poison the train data after which the classification accuracy dropped to 0.8711. © 2021, Springer Nature Switzerland AG.