Chaos theory encrypts medical scans in seconds, even if hackers breach the network

University of East Anglia

In 2024, a ransomware attack on Synnovis, a pathology services provider, crippled NHS laboratory services and cost an estimated 30 million pounds. The attackers did not need to break sophisticated encryption. They reached systems that were never designed to be exposed to the internet, accessed unprotected data, and caused months of disruption.

Medical imaging systems are a known weak point. Many hospitals store X-rays, CT scans, and MRIs on legacy systems using protocols that predate modern cybersecurity. If an attacker reaches a picture archiving and communications system (PACS), unencrypted images can be accessed, copied, or leaked in minutes.

A team at the University of East Anglia has built a system that changes the equation. Their approach encrypts each medical image individually, so that even if an attacker breaches the network, the images themselves remain unreadable. The method, published in the Wiley Journal of Computational and Mathematical Methods, processes images in two to four seconds, fast enough for routine clinical use.

Borrowing from the butterfly effect

Chaos theory describes systems that follow deterministic rules but are so sensitive to initial conditions that their outputs appear random. The butterfly effect, where small perturbations cascade into large-scale changes, is the popular example. In encryption, that extreme sensitivity is useful: it means that without knowing the exact starting conditions (the key), the encrypted output is practically impossible to reverse.

The UEA system uses chaotic mathematical functions to generate S-Boxes, substitution tables that determine how parts of an image are scrambled during encryption. Because these S-Boxes are regenerated for each image, attackers cannot rely on fixed patterns to crack the code.

The system also employs Galois Field arithmetic, a mathematical framework commonly used in cryptography, to mix and transform image data in precise but extremely difficult-to-reverse ways. And XNOR diffusion blends each pixel's data with its neighbors, ensuring that even a single-pixel change ripples across the entire image, destroying any recognizable structure.

Fast enough for emergency radiology

Speed has been the persistent weakness of previous medical image encryption research. Many proposed systems work in theory but are too slow for clinical environments where radiologists need images within seconds, not minutes. Hassan Malik, Associate Professor in Computing Sciences at UEA, described speed as a core design requirement, not an afterthought.

At two to four seconds per image, the system is designed to handle high-volume environments like emergency radiology departments without creating bottlenecks. It integrates with existing hospital systems, works across image types including X-rays and MRIs, and runs on standard hospital server hardware.

A complement, not a replacement

The system is designed to add a layer of protection on top of existing NHS cybersecurity measures, not to replace them. Network-level defenses, access controls, and staff training remain important. What image-level encryption adds is resilience: even when other defenses fail, and recent NHS history demonstrates they do fail, the images themselves stay protected.

Jawaid Iqbal, Associate Professor at Riphah International University in Pakistan and a collaborator on the project, noted that the approach is particularly valuable when older equipment or external suppliers create vulnerabilities that are difficult to eliminate entirely.

Not yet tested in live NHS environments

The encryption method has been developed and validated in research settings but has not yet been deployed in actual NHS hospitals. The team is preparing pilot deployments with NHS partners to assess performance across different setups, measure impact on day-to-day radiology workflows, and explore integration with national cybersecurity guidance.

Real-world deployment will introduce challenges that laboratory testing cannot fully replicate. Different NHS trusts use different PACS systems, different network architectures, and different workflow patterns. Key management, the secure generation, distribution, and storage of encryption keys, becomes critical at scale and must be robust against both technical failures and human error.

The pilot phase will also need to address how encrypted images interact with AI diagnostic tools, which increasingly process medical images for automated screening and analysis. If encryption adds latency or requires images to be decrypted before AI processing, the workflow implications could be significant.

The team has invited NHS trusts, imaging technology vendors, and cybersecurity teams to participate in shaping the rollout.