Plaintext-Free Deep Learning for Privacy-Preserving Medical Image Analysis via Frequency Information Embedding

TL;DR

The paper tackles privacy risks in DL-based medical image analysis by eliminating plaintext exposure during training and inference through a frequency-domain approach. It introduces Frequency-domain Exchange Style Fusion (FESF), combining an Image Hidden Module (IHM) that embeds partial plaintext frequency information into a host image and an Image Quality Enhancement Module (IQEM) that uses adversarial learning to align surrogate images with the host-domain distribution, then refines them with a second IHM pass. The framework employs FFT to obtain amplitude $ A$, phase $ P$, a low-frequency mask $ M$ defined by $ alpha$, and amplitude mixing ratio $ beta$ to produce surrogate images via $x_i^{sy}= F^{-1}( A_{x_i^{sy}}, P_{x^{ho}})$, with $ L_{GAN}$ guiding the generator $ G: S ightarrow K$ against discriminator $ D_K$ to yield high-fidelity surrogates $x_i^{su'}$. Experiments across multiple datasets and architectures show that surrogate images preserve essential diagnostic content with reasonable degradation relative to plaintext upper bounds, while IQEM improves image quality and maintains privacy, enabling practical, privacy-preserving DL in medical imaging.

Abstract

In the fast-evolving field of medical image analysis, Deep Learning (DL)-based methods have achieved tremendous success. However, these methods require plaintext data for training and inference stages, raising privacy concerns, especially in the sensitive area of medical data. To tackle these concerns, this paper proposes a novel framework that uses surrogate images for analysis, eliminating the need for plaintext images. This approach is called Frequency-domain Exchange Style Fusion (FESF). The framework includes two main components: Image Hidden Module (IHM) and Image Quality Enhancement Module~(IQEM). The~IHM performs in the frequency domain, blending the features of plaintext medical images into host medical images, and then combines this with IQEM to improve and create surrogate images effectively. During the diagnostic model training process, only surrogate images are used, enabling anonymous analysis without any plaintext data during both training and inference stages. Extensive evaluations demonstrate that our framework effectively preserves the privacy of medical images and maintains diagnostic accuracy of DL models at a relatively high level, proving its effectiveness across various datasets and DL-based models.

Figures (3)

• Figure 1: The concepts of the traditional medical image disease diagnosis method and ours.
• Figure 2: The overview of the proposed FESF framework.
• Figure 3: Visual comparison of host, plaintext, synthetic, and surrogate images across different datasets.