New paper on adversarial robustness of MR reconstruction algorithms

Our paper on “Adversarial Robustness of MR Image Reconstruction Under Realistic Perturbations” first-authored by Nikolas Morshuis got accepted to the MICCAI workshop on Machine Learning for Medical Image Reconstruction (MLMIR) 2022 as an oral presentation.

In this work, we show that commonly used deep learning based MR reconstruction algorithms are sensitive to realistic perturbations such as noise, and in particular, adversarial patient rotations. The attacks are particularly effective when only targeting diagnostically relevant regions.

Resources: Paper | arXiv pre-print | Code

Machine Learning in Medical Image Analysis
Machine Learning in Medical Image Analysis
Bridging the gap between AI and clinical practice