Unmasking adversarial attacks using a robust XAI-driven approach for secure medical image classification
- 1Department of Mathematical and Physical Sciences, East West University
- 2Department of Computer Science and Engineering, East West University
Abstract
Deep learning models have revolutionized medical image analysis but remain vulnerable to adversarial attacks. This work presents FusionStack, an explainability-driven framework for detecting adversarial perturbations through multi-method XAI stability analysis. We introduce three novel stability metrics (LIME Stability Metric, SHAP Stability Metric, Grad-CAM Stability Metric) quantifying explanation consistency across convolutional feature hierarchies, augmented with frequency-domain and gradient-based features. Tested on five medical imaging datasets, five CNN models, and five adversarial attacks with rigorous patient-level separation, FusionStack achieves 98.71% accuracy and 0.998 AUC. Cross-attack evaluation shows DeepFool-trained detectors achieve 96.85% transferability as universal defenders. Cross-dataset evaluation demonstrates 87.05% mean accuracy, with histopathology-trained detectors achieving 100% transfer to all modalities. Lightweight frequency-domain detection (10 features) achieves 95.46% accuracy with 5.75× speedup. Our findings establish XAI stability analysis as a principled pathway toward adversarially robust medical AI systems for safety-critical applications.
Related
- A robust adversarial learning framework for EEG emotion recognition
- Privacy-preserving collaborative transfer learning with generative adversarial networks (CTL-GAN) for classification of dry eye disease
- A Robust and Trustable Approach to Incorporate Medical Domain Knowledge in Machine Learning Models for Diabetic Retinopathy Screening Using Routine Lab Results
- Efficient Alzheimer’s disease classification with robust rank pooling
- Topological signatures vs. gradient histograms: A comparative study for medical image classification
BibTeX
@article{saha2026unmasking,
title = {{Unmasking adversarial attacks using a robust XAI-driven approach for secure medical image classification}},
author = {Sudarshan Saha and Sohel Rana and Ahmed Wasif Reza and M. Ruhul Amin},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-61357-1},
url = {https://doi.org/10.1038/s41598-026-61357-1},
note = {CC BY},
}
RIS
TY - JOUR
TI - Unmasking adversarial attacks using a robust XAI-driven approach for secure medical image classification
AU - Sudarshan Saha
AU - Sohel Rana
AU - Ahmed Wasif Reza
AU - M. Ruhul Amin
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-61357-1
UR - https://doi.org/10.1038/s41598-026-61357-1
AB - Deep learning models have revolutionized medical image analysis but remain vulnerable to adversarial attacks. This work presents FusionStack, an explainability-driven framework for detecting adversarial perturbations through multi-method XAI stability analysis. We introduce three novel stability metrics (LIME Stability Metric, SHAP Stability Metric, Grad-CAM Stability Metric) quantifying explanation consistency across convolutional feature hierarchies, augmented with frequency-domain and gradient-based features. Tested on five medical imaging datasets, five CNN models, and five adversarial attacks with rigorous patient-level separation, FusionStack achieves 98.71% accuracy and 0.998 AUC. Cross-attack evaluation shows DeepFool-trained detectors achieve 96.85% transferability as universal defenders. Cross-dataset evaluation demonstrates 87.05% mean accuracy, with histopathology-trained detectors achieving 100% transfer to all modalities. Lightweight frequency-domain detection (10 features) achieves 95.46% accuracy with 5.75× speedup. Our findings establish XAI stability analysis as a principled pathway toward adversarially robust medical AI systems for safety-critical applications.
SN - 2045-2322
ER -