Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: Jan 14, 2026

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

Dual adversarial attacks on Explainable Deep Learning in medical image classification.

Wanman Li¹, Wan Mohd Nazmee Wan Zainon¹

¹School of Computer Sciences, USM, 11700, Malaysia.

Computer Methods and Programs in Biomedicine

|October 25, 2025

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Case Report: IgG4-RD-related autoimmune pancreatitis combined with monoclonal gammopathy of undetermined significance.

Frontiers in immunology·2026

Same journal

SynTME: A tumor microenvironment-aware, pharmacology-inspired multi-stage framework for drug synergy prediction.

Computer methods and programs in biomedicine·2026

Same journal

MMFVS-Net: A triple-symmetric cross-attention network for multimodal optical image fusion and high-accuracy virtual staining of breast cancer tissues.

Computer methods and programs in biomedicine·2026

Same journal

A novel Milstein-stochastic epidemiologically-informed neural network for approaching epidemic dynamics: Application to Mpox disease.

Computer methods and programs in biomedicine·2026

Same journal

Accounting for approximation errors using surrogate-based parameter estimation of cardiac mechanics digital twins.

Computer methods and programs in biomedicine·2026

Same journal

Facial iPPG heatmap patterns based on period-aware autoencoder show association with carotid atherosclerosis towards non-contact hemodynamic assessment.

Computer methods and programs in biomedicine·2026

Same journal

Explainable machine learning models predict liver fibrosis risk and outcome in the general population: Development and multi-cohort external validation.

Computer methods and programs in biomedicine·2026

See all related articles

Explainable Deep Learning (XDL) systems in medical imaging are vulnerable to dual adversarial attacks that simultaneously target predictions and explanations. These attacks, using subtle perturbations, can undermine both accuracy and interpretability, posing risks in clinical settings.

Area of Science:

Medical Imaging
Artificial Intelligence
Computer Vision

Background:

Explainable Deep Learning (XDL) enhances clinician trust in medical image classification.
XDL models are vulnerable to adversarial perturbations affecting predictions and explanations.
Existing attacks target predictions or explanations in isolation, not their coupled vulnerability.

Purpose of the Study:

Propose dual adversarial attacks targeting both predictions and explanations simultaneously.
Introduce a novel evaluation metric, Attack Success Rate (ASR), for joint assessment.
Address the overlooked coupled vulnerability in XDL for medical imaging.

Main Methods:

Develop a dual adversarial attack framework with Location and Top-k Attack strategies.

Keywords:

Adversarial robustness Dual adversarial attacks Explainable Deep Learning Interpretability Medical image classification

Related Experiment Videos

Last Updated: Jan 14, 2026

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

Utilize iterative gradient-based optimization under an L-infinity constraint for imperceptible perturbations.

Quantify effectiveness using ASR, measuring misclassification and explanation distortion.

Main Results:

Dual attacks consistently achieve high ASR across diverse medical imaging datasets (Chest X-ray, Fundoscopy, Dermoscopy).
Demonstrate reliable effectiveness and broad applicability in manipulating predictions and explanations.
Experiments involve ResNet50, DenseNet121, and EfficientNet-V2 models with Integrated Gradients, GradCAM, and ScoreCAM.

Conclusions:

XDL systems in medical imaging exhibit coupled vulnerability of prediction and explanation.
Imperceptible perturbations can simultaneously degrade classification accuracy and explanation reliability.
Highlight the urgent need for defenses that protect both accuracy and interpretability in clinical applications.