MITS-GAN: Safeguarding medical imaging from tampering with generative adversarial networks

Area of Science:

• Medical Imaging
• Cybersecurity
• Artificial Intelligence

Background:

• Generative models like Generative Adversarial Networks (GANs) offer advanced image generation capabilities.
• Malicious use of GANs, particularly in sensitive medical imaging, poses significant risks.
• Existing methods for securing medical images against tampering are insufficient.

Purpose of the Study:

• To introduce MITS-GAN, a novel approach for tamper-proofing medical images, specifically CT scans.
• To enhance the resistance of medical images against adversarial attacks from GANs.
• To ensure the integrity and reliability of medical imaging data.

Main Methods:

• Developed MITS-GAN, a system that introduces finely tuned, human-imperceptible perturbations to input data.
• Utilized Gaussian noise as a protective measure against various adversarial attacks targeting CT-GAN architectures.
• Focused on disrupting the output of attacker-controlled CT-GANs to prevent image manipulation.

Main Results:

• MITS-GAN demonstrated superior tamper resistance compared to existing techniques.
• Experimental results on CT scans showed the generation of tamper-resistant images with negligible artifacts.
• The proposed method effectively countered malicious tampering attempts without compromising image quality.

Conclusions:

• MITS-GAN offers a robust solution for securing medical images against sophisticated cyber threats.
• This proactive approach supports the responsible and ethical application of generative models in healthcare.
• The study lays the groundwork for future research in medical imaging cybersecurity and adversarial defense.