MMIF: GPU-Accelerated Parallel Cryptographic Permutation Network for Secure Multimodal Medical Image Fusion in Brain Tumor Detection

In the modern hospital, a doctor's ability to see inside the human body has never been sharper. By combining images from different scanners—such as the detailed structural maps from a CT scan, the soft-tissue clarity of an MRI, and the metabolic activity shown by a PET scan—clinicians can construct a single, comprehensive picture of a brain tumor. This process, known as multimodal image fusion, is vital for accurate diagnosis and surgical planning. However, this powerful tool creates a new vulnerability. These fused images contain highly sensitive patient data, and as they move between computers or are stored for later review, they are at risk of being stolen or viewed by unauthorized eyes. Protecting this data is essential, but traditional methods of locking it away often come with a heavy price: they are slow. In a busy clinic, even a few seconds of delay while a computer scrambles an image can disrupt the flow of care, forcing doctors to wait before they can see the results they need.

A team of researchers at VIT-AP University has developed a new way to secure these medical images that solves this speed problem without sacrificing safety. Their approach, which they call a parallel cryptographic permutation network, fundamentally changes how the encryption process works. Instead of treating the image as a single block of data that must be locked up step-by-step, they break the image down into eight separate layers, or bit-planes. Imagine a digital photograph as a stack of transparent sheets, where each sheet holds a specific piece of the visual information. The researchers realized that these sheets could be shuffled independently and simultaneously, rather than one after another. By doing this on powerful graphics processors, which are designed to handle thousands of tasks at once, they turned a slow, sequential process into a rapid, parallel one.

The core of their method involves a clever use of chaos and biology-inspired coding to create the shuffling patterns. The system first generates a chaotic sequence, a stream of numbers that appears random and is extremely sensitive to its starting conditions, much like how a tiny change in the weather can lead to a completely different storm days later. This chaotic stream is then translated into a code based on the four building blocks of DNA. These biological symbols are used to create a unique set of instructions for rearranging the pixels in each of the eight image layers. Because the layers are processed at the same time, the computer does not have to wait for one layer to finish before starting the next. Once the layers are shuffled, they are reassembled into a single image and given one final layer of protection using a simple mathematical operation that mixes the data with a second stream of random numbers. This final step ensures that even a tiny change in the original image results in a completely different encrypted version, making it highly difficult for an attacker to guess the original content, though the researchers note that a determined attacker with access to specific plaintext-ciphertext pairs could potentially reconstruct the underlying sequences.

When the researchers tested their system on brain tumor images, the results were striking. The new method achieved a level of security that rivals the most complex existing techniques, with the encrypted images showing no discernible patterns and appearing as random static to any observer. More importantly, the speed was transformative. While older methods took over twelve milliseconds to encrypt a standard medical image, this new system completed the task in just 0.42 milliseconds. This is a speed increase of roughly thirty times compared to traditional approaches. The system utilized the computer's graphics processor to its full capacity, reaching 92.4 percent utilization, whereas older methods struggled to use even a fifth of the available power because their step-by-step nature prevented them from working in parallel.

Crucially, the researchers confirmed that this rapid encryption did not harm the medical value of the images. Because the process is mathematically reversible, the original fused image can be recovered perfectly after decryption, with no loss of detail or quality. This means that a surgeon can view the encrypted image, decrypt it instantly, and see the tumor boundaries exactly as they were before, without any distortion that could lead to a misdiagnosis. The study also showed that the system is robust against various types of cyberattacks, including those where an attacker tries to guess the key by analyzing how small changes in the input affect the output. The combination of the DNA-inspired shuffling and the final mixing step created a strong shield that kept the data secure while moving at the speed of modern clinical workflows.

This work suggests that the bottleneck of slow encryption in medical imaging can be removed by rethinking the architecture of the security system itself. By moving away from sequential, step-by-step locking and embracing parallel processing, it is possible to protect patient privacy without slowing down the doctors who need to save lives. The researchers acknowledge that while their method is highly effective on the specific hardware they tested, future work will need to explore how it performs on different types of computers and with different kinds of medical images. They also note that for the system to be widely adopted in hospitals, it will need to be integrated with standard security protocols and key management systems that are already in use. Nevertheless, the study provides a clear path forward: a secure, fast, and lossless way to handle the sensitive data that drives modern brain tumor detection, ensuring that the technology used to heal patients remains safe from those who would do them harm.