Researchers use LLMs for lossless medical image compression and steganography
New method embeds hidden data while compressing medical images with no loss...
A new research paper from a team led by Pengcheng Zheng introduces a framework that jointly performs lossless compression and steganography on medical images using large language models (LLMs). Traditional LLM-based compressors for medical images struggle with balancing compression performance and efficiency, and often lack security features. The proposed method leverages bit plane slicing to separate image data, then uses an adaptive modalities decomposition strategy to split the image into two segments (global and local) for dual-path lossless compression.
During the local modality path, the authors introduce a segmented message steganography algorithm that invisibly embeds private metadata or messages. To further optimize performance, they develop an anatomical priors-based low-rank adaptation (A-LoRA) fine-tuning technique. Experimental results show significant improvements in compression ratios, processing speed, and security over existing methods. The source code will be released publicly, potentially enabling secure and efficient medical image storage and transmission.
- LLMs power lossless compression of medical images while enabling hidden data embedding.
- Bit plane slicing and adaptive modality decomposition split images for dual-path processing.
- A-LoRA fine-tuning combines anatomical priors with low-rank adaptation for better efficiency.
Why It Matters
Enables secure, efficient medical image archiving and transmission, critical for patient privacy and telemedicine.