What is it about?
This research introduces an improved encryption method that allows medical images, such as chest X-rays, to be safely transmitted and stored in the cloud for AI-powered Tuberculosis diagnosis while keeping patient data private. Most current encryption methods either make files too large or only work with color images. Because medical images are usually grayscale, they are harder to protect efficiently. Our study solves this by creating a "Perceptual Encryption" system that works for both color and grayscale images. It achieves high security with only a 12% increase in file size. This allows a "Smart Hospital" to securely outsource data to the cloud, where the images are decrypted and analyzed by an AI model (EfficientNetV2) to detect Tuberculosis with high accuracy. We also developed a new way to "augment" or expand small medical datasets using noise, which helped our AI become 10% more accurate. Keywords: medical image privacy, healthcare cloud security, tuberculosis AI diagnosis, perceptual encryption, EfficientNetV2, JPEG-compliant encryption, Smart hospital technology. You can access the full technical details via the DOI [https://doi.org/10.3390/electronics11162514] or view more of my related research on my ORCID profile [https://orcid.org/0000-0002-6022-7413].
Featured Image
Photo by CDC on Unsplash
Why is it important?
This work bridges the gap between high-security data protection and high-performance AI in smart hospitals. Compared to existing Perceptual Encryption schemes, our approach is unique because: 1. Universal Compatibility: We solved the problem where medical images were previously inadequate for secure transmission because existing PE methods required color inputs for better security. It is the first perceptual encryption method optimized specifically for grayscale medical images, making it practical for real-world radiology. 2. Extreme Efficiency: By reducing the "bitrate overhead" from 113% to just 12%, we make it financially and technically viable for hospitals to use cloud storage at scale. 3. Resilient AI Diagnosis: We demonstrated that our encryption and compression steps do not compromise the final diagnostic quality, even improving AI accuracy by 10% through a custom noise-based data augmentation method. 4. Format Compliance: Unlike chaos-based encryption that produces a raw bitstream, our encryption output remains a standard JPEG image. This allows the encrypted files to be managed by any standard cloud storage or photo-handling service that is, existing hospital software does not need to be replaced.
Perspectives
We wanted to solve a specific paradox: How can we give AI the data it needs to save lives while ensuring that same data remains protected from external threats during transmission and storage? In our research, we noticed that while color image encryption was advancing (optimized for color social media photos), the medical field, which relies heavily on grayscale images, was being left behind. Conventional Perceptual Encryption often forced a choice between security and storage efficiency—leading to massive files that would clog hospital networks. We believe that privacy should not be a barrier to innovation. By creating a method that respects the intrinsic properties of the JPEG standard, we’ve created a system where security and efficiency finally coexist. This allows hospitals to keep the "secret key" for decryption while leveraging the massive power of the cloud for deep learning.
Dr Ijaz Ahmad
Korea University
Read the Original
This page is a summary of: A Perceptual Encryption-Based Image Communication System for Deep Learning-Based Tuberculosis Diagnosis Using Healthcare Cloud Services, Electronics, August 2022, MDPI AG,
DOI: 10.3390/electronics11162514.
You can read the full text:
Contributors
The following have contributed to this page
