MITS-GAN: Safeguarding Medical Imaging from Tampering with Generative Adversarial Networks

TL;DR

MISS-GAN is introduced, a novel approach to prevent tampering in medical images, with a specific focus on CT scans, that involves the introduction of appropriate Gaussian noise to the input as a protective measure against various attacks.

Abstract

The progress in generative models, particularly Generative Adversarial Networks (GANs), opened new possibilities for image generation but raised concerns about potential malicious uses, especially in sensitive areas like medical imaging. This study introduces MITS-GAN, a novel approach to prevent tampering in medical images, with a specific focus on CT scans. The approach disrupts the output of the attacker's CT-GAN architecture by introducing finely tuned perturbations that are imperceptible to the human eye. Specifically, the proposed approach involves the introduction of appropriate Gaussian noise to the input as a protective measure against various attacks. Our method aims to enhance tamper resistance, comparing favorably to existing techniques. Experimental results on a CT scan demonstrate MITS-GAN's superior performance, emphasizing its ability to generate tamper-resistant images with negligible artifacts. As image tampering in medical domains poses life-threatening risks, our proactive approach contributes to the responsible and ethical use of generative models. This work provides a foundation for future research in countering cyber threats in medical imaging. Models and codes are publicly available on https://iplab.dmi.unict.it/MITS-GAN-2024/.

Figures (7)

• Figure 1: Overview of the GAN architecture and training process.
• Figure 2: Qualitative results comparison between real and tampered CT scans. Columns 1 and 3 show the original images, whereas Columns 2 and 4 depict the manipulated images. The red bounding boxes highlight the manipulations introduced by CT-GAN, wherein tumors have been added to the scans. This visual representation underscores the impact and detectability of manipulations within the medical imaging context.
• Figure 3: Comparison between Real unprotected CT scans and protected CT scans generated by the proposed model MITS-GAN. As can be noted, the protected images, which embed the protection noise pattern, are similar to the original one.
• Figure 4: Model Architecture Overview: The generator receives the input image $x$ and perturbation noise $\delta$ to produce the protected image $x^p$. Subsequently, $x^p$ is forwarded to the manipulation model and discriminator.
• Figure 5: Example of a CT scan.