中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
首页 机构 成果 学者
  
  1. CAS IR Grid
  2. 计算技术研究所
  3. 中国科学院计算技术研究所
  4. 中国科学院计算技术研究所期刊论文
  5. 英文
Adversarial Medical Image With Hierarchical Feature Hiding

文献类型：期刊论文

作者Yao, Qingsong1,2; He, Zecheng3; Li, Yuexiang4,5; Lin, Yi6; Ma, Kai4; Zheng, Yefeng4; Zhou, S. Kevin7,8,9
刊名IEEE TRANSACTIONS ON MEDICAL IMAGING
出版日期2024-04-01
卷号43期号:4页码:1296-1307
关键词Security in machine learning adversarial attacks and defense
ISSN号0278-0062
DOI10.1109/TMI.2023.3335098
英文摘要Deep learning based methods for medical images can be easily compromised by adversarial examples (AEs), posing a great security flaw in clinical decision-making. It has been discovered that conventional adversarial attacks like PGD which optimize the classification logits, are easy to distinguish in the feature space, resulting in accurate reactive defenses. To better understand this phenomenon and reassess the reliability of the reactive defenses for medical AEs, we thoroughly investigate the characteristic of conventional medical AEs. Specifically, we first theoretically prove that conventional adversarial attacks change the outputs by continuously optimizing vulnerable features in a fixed direction, thereby leading to outlier representations in the feature space. Then, a stress test is conducted to reveal the vulnerability of medical images, by comparing with natural images. Interestingly, this vulnerability is a double-edged sword, which can be exploited to hide AEs. We then propose a simple-yet-effective hierarchical feature constraint (HFC), a novel add-on to conventional white-box attacks, which assists to hide the adversarial feature in the target feature distribution. The proposed method is evaluated on three medical datasets, both 2D and 3D, with different modalities. The experimental results demonstrate the superiority of HFC, i.e., it bypasses an array of state-of-the-art adversarial medical AE detectors more efficiently than competing adaptive attacks, which reveals the deficiencies of medical reactive defense and allows to develop more robust defenses in future.
WOS研究方向Computer Science ; Engineering ; Imaging Science & Photographic Technology ; Radiology, Nuclear Medicine & Medical Imaging
语种英语
WOS记录号WOS:001196733400025
出版者IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
源URL[http://119.78.100.204/handle/2XEOYT63/39886]  
专题中国科学院计算技术研究所期刊论文_英文
通讯作者Zhou, S. Kevin
作者单位1.Univ Chinese Acad Sci, Beijing, Peoples R China
2.Chinese Acad Sci, Inst Comp Technol, Beijing 10045, Peoples R China
3.Meta Real Labs, Burlingame, CA 94010 USA
4.Jarvis Res Ctr, Tencent YouTu Lab, Shenzhen 518057, Peoples R China
5.Guangxi Med Univ, Guangxi Key Lab Genom & Personalized Med, Med AI ReS MARS Grp, Nanning 530021, Peoples R China
6.Hong Kong Univ Sci & Technol, Dept Comp Sci & Engn, Hong Kong, Peoples R China
7.Univ Sci & Technol China, Sch Biomed Engn, Hefei 230026, Peoples R China
8.Univ Sci & Technol China, Suzhou Inst Adv Res, Hefei 230026, Peoples R China
9.Chinese Acad Sci, Inst Comp Technol, Beijing 100045, Peoples R China
推荐引用方式
GB/T 7714		 Yao, Qingsong,He, Zecheng,Li, Yuexiang,et al. Adversarial Medical Image With Hierarchical Feature Hiding[J]. IEEE TRANSACTIONS ON MEDICAL IMAGING,2024,43(4):1296-1307.
APA Yao, Qingsong.,He, Zecheng.,Li, Yuexiang.,Lin, Yi.,Ma, Kai.,...&Zhou, S. Kevin.(2024).Adversarial Medical Image With Hierarchical Feature Hiding.IEEE TRANSACTIONS ON MEDICAL IMAGING,43(4),1296-1307.
MLA Yao, Qingsong,et al."Adversarial Medical Image With Hierarchical Feature Hiding".IEEE TRANSACTIONS ON MEDICAL IMAGING 43.4(2024):1296-1307.

入库方式： OAI收割

来源：计算技术研究所

浏览0
下载0
收藏0
其他版本

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。