arXiv 2406.084862024-09-02eess.IVcs.CV

关于评估体积医学分割模型的对抗鲁棒性

On Evaluating Adversarial Robustness of Volumetric Medical Segmentation Models

Mohamed bin Zayed University of AI（穆罕默德·本·扎耶德人工智能大学）；Australian National University（澳大利亚国立大学）；Linköping University（林雪平大学）；Khalifa University（哈利法大学）

Hashmat Shadab Malik, Numan Saeed, Asif Hanif, Muzammal Naseer, Mohammad Yaqub, Salman Khan, Fahad Shahbaz Khan

备注：: Accepted at British Machine Vision Conference 2024

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文实证评估了卷积、Transformer 和 Mamba 体积医学分割模型在四个数据集上的对抗鲁棒性，发现 Transformer 模型更鲁棒、Mamba 模型最脆弱，且大规模训练能提升鲁棒性。

摘要：

近年来，体积医学分割模型在器官和肿瘤分割任务上取得了显著成功。然而，它们对对抗攻击的脆弱性在很大程度上仍未得到探索，这引发了人们对在医疗领域部署采用此类模型的工具的严重担忧。这凸显了研究现有模型鲁棒性的重要性。在此背景下，我们的工作旨在实证考察当前体积分割架构的对抗鲁棒性，涵盖卷积、Transformer 和 Mamba 模型。我们将这一研究扩展到四个体积分割数据集，在白盒和黑盒对抗攻击下评估鲁棒性。总体而言，我们观察到，虽然基于像素和基于频率的攻击在白盒设置下表现都相当不错，但后者在基于迁移的黑盒攻击下表现明显更好。在我们的实验中，我们观察到基于 Transformer 的模型比基于卷积的模型表现出更高的鲁棒性，而基于 Mamba 的模型最为脆弱。此外，我们表明，体积分割模型的大规模训练提高了模型对对抗攻击的鲁棒性。代码和鲁棒模型可在 https://github.com/HashmatShadab/Robustness-of-Volumetric-Medical-Segmentation-Models 获取。

英文摘要：

Volumetric medical segmentation models have achieved significant success on organ and tumor-based segmentation tasks in recent years. However, their vulnerability to adversarial attacks remains largely unexplored, raising serious concerns regarding the real-world deployment of tools employing such models in the healthcare sector. This underscores the importance of investigating the robustness of existing models. In this context, our work aims to empirically examine the adversarial robustness across current volumetric segmentation architectures, encompassing Convolutional, Transformer, and Mamba-based models. We extend this investigation across four volumetric segmentation datasets, evaluating robustness under both white box and black box adversarial attacks. Overall, we observe that while both pixel and frequency-based attacks perform reasonably well under \emph{white box} setting, the latter performs significantly better under transfer-based black box attacks. Across our experiments, we observe transformer-based models show higher robustness than convolution-based models with Mamba-based models being the most vulnerable. Additionally, we show that large-scale training of volumetric segmentation models improves the model's robustness against adversarial attacks. The code and robust models are available at https://github.com/HashmatShadab/Robustness-of-Volumetric-Medical-Segmentation-Models.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。