Death by Retrospective Undersampling - Caveats and Solutions for Learning-Based MRI Reconstructions

This study challenges the validity of retrospective undersampling in MRI data science by analysis via an MRI physics simulation. We demonstrate that retrospective undersampling, a method often used to create training data for reconstruction models, can inherently alter MRI signals from their prospective counterparts. This arises from the sequential nature of MRI acquisition, where undersampling post-acquisition effectively alters the MR sequence and the magnetization dynamic in a non-linear fashion. We show that even in common sequences, this effect can make learning-based reconstructions unreliable. Our simulation provides both, (i) a tool for generating accurate prospective undersampled datasets for analysis of such effects, or for MRI training data augmentation, and (ii) a differentiable reconstruction operator that models undersampling correctly. The provided insights are crucial for the development and evaluation of AI-driven acceleration of diagnostic MRI tools.