Dongguk University Researchers Reinforce Medical Digital Twins Against Cyberattacks

Imagine a virtual version of your body—an exact digital copy that can predict disease, test treatments, and guide your doctor’s decisions. That’s the idea behind medical digital twins, a growing field in healthcare powered by artificial intelligence and real-time data. But with all their potential, these systems have an Achilles’ heel: they’re vulnerable to cyberattacks. Small, almost invisible tweaks to the input data, like a few altered pixels in a medical image, can cause the AI to misdiagnose a serious illness. In a high-stakes setting like healthcare, that’s more than just a glitch, it’s a dangerous risk.

Now, a team of researchers from Dongguk University in South Korea has developed a new way to fight back.

Digital Twins in Healthcare

Digital twins have already changed industries like aerospace and manufacturing. In healthcare, they’re being used to simulate organs, tissues, and even entire body systems. These virtual models, created using data from scans, sensors, and lab tests, allow doctors and researchers to predict how diseases might progress or how a patient might respond to treatment.

But they’re not invincible. Adversarial attacks, which are sneaky, deliberate data manipulations, can throw off even the most accurate digital twin models. In one case, a breast cancer diagnostic model trained on thermography images dropped from 92% accuracy to a mere 5% when hit by such attacks.

That’s where WBAD—short for Wavelet-Based Adversarial Training—comes in.

Adversarial Robust Image Processing in Medical Digital Twin

A Smart Defense: WBAD

Led by Professor Insoo Sohn, the team at Dongguk University, together with collaborators from Oregon State University, designed WBAD as a two-step defense system to protect medical digital twins.

First, it cleans the data. Attacks usually introduce high-frequency “noise” into images—imperceptible to the eye, but enough to confuse an AI. WBAD uses a method called wavelet denoising to filter out this interference while preserving the important low-frequency features of the image.

Next, the AI model undergoes adversarial training. It’s shown examples of tampered data during training, helping it learn to spot and resist such attacks in the real world.

When tested on a breast cancer digital twin, WBAD proved remarkably effective:

Accuracy shot back up to 98% against Fast Gradient Sign Method (FGSM) attacks

93% against Projected Gradient Descent (PGD) attacks

90% against Carlini & Wagner (C&W) attacks

“Our results demonstrate a transformative approach to medical digital twin security, providing a comprehensive and effective defense against cyberattacks and leading to enhanced system functionality and reliability,” said Professor Sohn.

The team’s findings were published in Information Fusion (March 2025, Volume 115), with the research available online since October 2024.

Dongguk University at the Crossroads of Technology and Tradition

Founded in 1906, Dongguk University sits at the intersection of tradition and innovation. Based in Seoul and with campuses in Gyeongju, Goyang, and Los Angeles, the university is home to over 18,000 students and 1,300 faculty members. It champions cross-disciplinary collaboration across biotechnology, information technology, cultural studies—and even Buddhist philosophy.

This latest research is a reflection of Dongguk’s growing focus on AI and digital health. As the healthcare world embraces technologies like digital twins, solutions like WBAD could play a key role in ensuring they are safe, accurate, and worthy of patient trust.

MedTech Malta 2025

MedTech Malta 2025 is set to take place from 12 to 14 November in Valletta, bringing together innovators, investors, and professionals at the forefront of healthcare and medical technology. Whether you’re a startup founder seeking funding, a MedTech expert eager to connect with peers, or someone looking to stay informed on emerging trends, the event offers a practical and engaging space to get involved. From thought-provoking panels and hands-on networking to opportunities like speaker slots, sponsorship packages, and the MedTech Startup Pitch Competition, there’s something for everyone. To explore how you can participate, reach out to [email protected]. Early bird tickets are now available—secure yours today at the lowest price before they’re gone.