EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Comparing Detection Schemes for Adversarial Images against Deep Learning Models for Cancer Imaging

Marina Z Joel, Arman Avesta iD, Daniel X Yang, Jian-Ge Zhou, Antonio Omuro, Roy S Herbst, Harlan M Krumholz, Sanjay Aneja iD

Cancers · 01.03.2023 · 15 (5) · 1548

DOI: 10.3390/cancers15051548 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Cancers, online erschienen am 01.03.2023. Verfasst von Marina Z Joel und 7 weiteren Autor:innen. Beteiligte Einrichtung: Department of Dermatology, Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA. Die Arbeit wurde bisher 14-mal zitiert (Crossref) und nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NCI NIH HHS, NCATS NIH HHS, Developmental Research Program Grant (PI: Aneja).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Deep learning (DL) models have demonstrated state-of-the-art performance in the classification of diagnostic imaging in oncology. However, DL models for medical images can be compromised by adversarial images, where pixel values of input images are manipulated to deceive the DL model. To address this limitation, our study investigates the detectability of adversarial images in oncology using multiple detection schemes. Experiments were conducted on thoracic computed tomography (CT) scans, mammography, and brain magnetic resonance imaging (MRI). For each dataset we trained a convolutional neural network to classify the presence or absence of malignancy. We trained five DL and machine learning (ML)-based detection models and tested their performance in detecting adversarial images. Adversarial images generated using projected gradient descent (PGD) with a perturbation size of 0.004 were detected by the ResNet detection model with an accuracy of 100% for CT, 100% for mammogram, and 90.0% for MRI. Overall, adversarial images were detected with high accuracy in settings where adversarial perturbation was above set thresholds. Adversarial detection should be considered alongside adversarial training as a defense technique to protect DL models for cancer imaging classification from the threat of adversarial images.

Schlagwörter

Stichwörter der Autor:innen

artificial intelligencecancer classificationdeep learningmedical imaging

Autor:innen und Institutionen

  • Department of Dermatology, Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA.
  • Department of Therapeutic Radiology, Yale School of Medicine, New Haven, CT 06510, USA.
  • Department of Chemistry, Physics and Atmospheric Science, Jackson State University, Jackson, MS 39217, USA.
  • Department of Neurology, Yale School of Medicine, New Haven, CT 06510, USA.
  • Department of Medicine, Yale School of Medicine, New Haven, CT 06510, USA.
  • Center for Outcomes Research and Evaluation (CORE), Yale School of Medicine, New Haven, CT 06510, USA.

ORCID-Profile: Arman Avesta, Sanjay Aneja

Förderung

  • NCI NIH HHS
  • NCATS NIH HHS
  • Developmental Research Program Grant (PI: Aneja)
  • William O. Seery Mentored Research Award for Cancer Research (PI: Aneja)
  • Conquer Cancer Career Development Award (PI: Aneja)
  • Hayden Family Foundation
  • National Science Foundation NSF Award

Zitieren

APA

Joel, M. Z., Avesta, A., Yang, D. X., Zhou, J. G., Omuro, A., Herbst, R. S., Krumholz, H. M., & Aneja, S. (2023). Comparing Detection Schemes for Adversarial Images against Deep Learning Models for Cancer Imaging. Cancers, 15(5), 1548. https://doi.org/10.3390/cancers15051548

Vancouver

Joel MZ, Avesta A, Yang DX, Zhou JG, Omuro A, Herbst RS, et al. Comparing Detection Schemes for Adversarial Images against Deep Learning Models for Cancer Imaging. Cancers. 2023;15(5):1548. doi:10.3390/cancers15051548

BibTeX
@article{Joel2023,
  author = {Joel, Marina Z and Avesta, Arman and Yang, Daniel X and Zhou, Jian-Ge and Omuro, Antonio and Herbst, Roy S and Krumholz, Harlan M and Aneja, Sanjay},
  title = {Comparing Detection Schemes for Adversarial Images against Deep Learning Models for Cancer Imaging},
  journal = {Cancers},
  year = {2023},
  volume = {15},
  number = {5},
  pages = {1548},
  doi = {10.3390/cancers15051548}
}
Kostenlos starten →