IRC-SC: Improving Robustness of CNN-Based MRI System Using SRGAN and CNN

Abstract

Convolutional Neural Networks (CNNs) have demonstrated remarkable accuracy in automating diagnostic tasks in medical imaging. However, they are highly vulnerable to adversarial attacks, where imperceptible perturbations in input images lead to misclassification. This paper proposes a novel hybrid framework, named IRC-SC combining Super Resolution Generative Adversarial Networks (SRGAN) with CNN s to enhance robustness against adversarial attacks. Using brain MRI datasets, the framework demonstrates over 95% improvement in robustness compared to state-of-the-art methods. It effectively defends against both white-box and black-box attacks, showcasing its potential for secure and reliable diagnostic applications in healthcare.