MedicalGuard: U-Net Model Robust against Adversarially Perturbed Images

Figure 2

Overview of the proposed method. (a) Generating adversarial example. (b) Training processing. (c) Inference.