Safeguarding Medical Image Segmentation Datasets against Unauthorized Training via Contour- and Texture-Aware Perturbations

TL;DR

An Unlearnable Medical image generation method, termed UMed, which integrates the prior knowledge of MIS by injecting contour- and texture-aware perturbations to protect images.

Abstract

The widespread availability of publicly accessible medical images has significantly propelled advancements in various research and clinical fields. Nonetheless, concerns regarding unauthorized training of AI systems for commercial purposes and the duties of patient privacy protection have led numerous institutions to hesitate to share their images. This is particularly true for medical image segmentation (MIS) datasets, where the processes of collection and fine-grained annotation are time-intensive and laborious. Recently, Unlearnable Examples (UEs) methods have shown the potential to protect images by adding invisible shortcuts. These shortcuts can prevent unauthorized deep neural networks from generalizing. However, existing UEs are designed for natural image classification and fail to protect MIS datasets imperceptibly as their protective perturbations are less learnable than important prior knowledge in MIS, e.g., contour and texture features. To this end, we propose an Unlearnable Medical image generation method, termed UMed. UMed integrates the prior knowledge of MIS by injecting contour- and texture-aware perturbations to protect images. Given that our target is to only poison features critical to MIS, UMed requires only minimal perturbations within the ROI and its contour to achieve greater imperceptibility (average PSNR is 50.03) and protective performance (clean average DSC degrades from 82.18% to 6.80%).

Figures (3)

• Figure 1: Illustration of using our proposed UMed to prevent an MIS dataset from unauthorized usage for AI model training. By adding protective perturbations to images of the MIS dataset, UMed can effectively reduce the clean segmentation performance of models trained on this dataset.
• Figure 2: Illustration of the (a) pipeline of the proposed UMed, (b) contour perturbator $\mathcal{G}_c$ of UMed, and (c) texture perturbator $\mathcal{G}_t$ of UMed. $\mathcal{G}_c$ injects contour-aware perturbations using an encoder-decoder structured generator $\mathcal{F}_c$ integrated with central difference convolution (CDC) kernels. $\mathcal{G}_t$ perturbs textures within the ROI constrained by a texture-aware adaptive bound.
• Figure 3: Visualization of the perturbations generated by different protectors. From left to right, each column represents original images, perturbations generated by EM, TAP, LSP, AR, SEP, $\mathcal{G}_c$ of UMed, $\mathcal{G}_t$ of UMed, and the images protected by UMed, respectively.