Detecting and Mitigating Adversarial Attacks on Deep Learning-Based MRI Reconstruction Without Any Retraining

2025-01-03

AI Summary

Deep learning–based MRI reconstruction models are vulnerable to adversarial attacks, and existing defenses typically require model retraining. To address this, we propose an online detection-and-mitigation framework that operates without model retraining. Our method leverages a physics-driven cyclic measurement consistency criterion for automatic adversarial detection and formulates a robust input-domain optimization objective under local spherical constraints to suppress k-space perturbations. Key contributions include: (i) the first retraining-free adversarial detection-and-mitigation paradigm; (ii) an interpretable cyclic consistency verification mechanism grounded in MRI physics; and (iii) a localized optimization strategy balancing stability and reconstruction fidelity. Extensive experiments across multiple datasets, diverse attack types and intensities, and various plug-and-play deep learning (PD-DL) architectures demonstrate significant improvements in PSNR and SSIM over state-of-the-art retraining-dependent methods, with superior visual reconstruction quality.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Adversarial Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Deep learning (DL) methods, especially those based on physics-driven DL, have become the state-of-the-art for reconstructing sub-sampled magnetic resonance imaging (MRI) data. However, studies have shown that these methods are susceptible to small adversarial input perturbations, or attacks, resulting in major distortions in the output images. Various strategies have been proposed to reduce the effects of these attacks, but they require retraining and may lower reconstruction quality for non-perturbed/clean inputs. In this work, we propose a novel approach for detecting and mitigating adversarial attacks on MRI reconstruction models without any retraining. Our detection strategy is based on the idea of cyclic measurement consistency. The output of the model is mapped to another set of MRI measurements for a different sub-sampling pattern, and this synthesized data is reconstructed with the same model. Intuitively, without an attack, the second reconstruction is expected to be consistent with the first, while with an attack, disruptions are present. Subsequently, this idea is extended to devise a novel objective function, which is minimized within a small ball around the attack input for mitigation. Experimental results show that our method substantially reduces the impact of adversarial perturbations across different datasets, attack types/strengths and PD-DL networks, and qualitatively and quantitatively outperforms conventional mitigation methods that involve retraining.

Problem

• Deep Learning
• MRI Image Reconstruction
• Adversarial Attacks Mitigation

Innovation

• Adversarial Attack Mitigation
• MRI Image Reconstruction
• Cyclic Consistency Measurement