Joint Lossless Compression and Steganography for Medical Images via Large Language Models

TL;DR

This work addresses the need for secure, efficient lossless compression of medical images by introducing a dual-path framework that partitions data into global and local modalities via adaptive bit-plane slicing. Global information is compactly encoded with a latent-variable path using bits-back coding, while local information is modeled by an LLM-based path that also facilitates integrated, segmented steganography to invisibly embed privacy messages; a visual-prompt mechanism and A-LoRA fine-tuning further boost performance. Across diverse medical datasets, the method achieves state-of-the-art compression (lower $bpp$) with favorable runtimes, while preserving high visual fidelity and enabling secure steganographic embedding. The approach demonstrates practical impact for secure medical data transmission and storage, and points to future work on lightweight LLMs to enable deployment in resource-constrained settings.

Abstract

Recently, large language models (LLMs) have driven promising progress in lossless image compression. However, directly adopting existing paradigms for medical images suffers from an unsatisfactory trade-off between compression performance and efficiency. Moreover, existing LLM-based compressors often overlook the security of the compression process, which is critical in modern medical scenarios. To this end, we propose a novel joint lossless compression and steganography framework. Inspired by bit plane slicing (BPS), we find it feasible to securely embed privacy messages into medical images in an invisible manner. Based on this insight, an adaptive modalities decomposition strategy is first devised to partition the entire image into two segments, providing global and local modalities for subsequent dual-path lossless compression. During this dual-path stage, we innovatively propose a segmented message steganography algorithm within the local modality path to ensure the security of the compression process. Coupled with the proposed anatomical priors-based low-rank adaptation (A-LoRA) fine-tuning strategy, extensive experimental results demonstrate the superiority of our proposed method in terms of compression ratios, efficiency, and security. The source code will be made publicly available.

Figures (6)

• Figure 1: Standard Lossless Compression and Steganography vs. our proposed joint framework.
• Figure 2: Overview of the proposed framework. In the encoding process, the image is firstly decomposed into global and local modalities via adaptive modalities decomposition strategy. Then, the global modalities are compressed using VAE with bits-back coding scheme. The local modalities are compressed by leveraging the unprecedented intelligence of LLM. The final bitstream are concatenated from the dual-path outputs. The decoding order is the reverse of the encoding process.
• Figure 3: Visualize each bit plane, with the planes organized from the MSP to the LSP, arranged in a left-to-right and top-to-bottom manner.
• Figure 4: An example of the encryption and decryption process. We send $L_1$, $L_2$, $L_s$, and Bitmap $\mathbf{B}$ to the receiver to ensure image $\mathbf{x}$ and message $\mathbf{M}$ can be losslessly recovered.
• Figure 5: Visualization of stego images. The proposed segmented message steganography algorithm achieves an invisible message encryption.