2019/07/30 by Utku Özbulak, Ozbulak, Utku, Arnout Van Messem +3 · 1 citation
Biochemistry, Genetics and Molecular Biology · Computer Science · Medicine · #Adversarial Robustness in Machine Learning #Autopsy Techniques and Outcomes #Bacillus and Francisella bacterial research #Computer Vision and Pattern Recognition (cs.CV) #Cryptography and Security (cs.CR) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1907.13124
openalex publication_date 2019/07/30 · openalex created_date 2022/07/19 · openalex updated_date 2026/07/28
Deep learning models, which are increasingly being used in the field of\nmedical image analysis, come with a major security risk, namely, their\nvulnerability to adversarial examples. Adversarial examples are carefully\ncrafted samples that force machine learning models to make mistakes during\ntesting time. These malicious samples have been shown to be highly effective in\nmisguiding classification tasks. However, research on the influence of\nadversarial examples on segmentation is significantly lacking. Given that a\nlarge portion of medical imaging problems are effectively segmentation\nproblems, we analyze the impact of adversarial examples on deep learning-based\nimage segmentation models. Specifically, we expose the vulnerability of these\nmodels to adversarial examples by proposing the Adaptive Segmentation Mask\nAttack (ASMA). This novel algorithm makes it possible to craft targeted\nadversarial examples that come with (1) high intersection-over-union rates\nbetween the target adversarial mask and the prediction and (2) with\nperturbation that is, for the most part, invisible to the bare eye. We lay out\nexperimental and visual evidence by showing results obtained for the ISIC skin\nlesion segmentation challenge and the problem of glaucoma optic disc\nsegmentation. An implementation of this algorithm and additional examples can\nbe found at https://github.com/utkuozbulak/adaptive-segmentation-mask-attack.\n