Comparative Evaluation of Deep Learning and Compressed Sensing Methods for Dynamic Contrast-Enhanced MRI Reconstruction

Motivation: Deep Learning (DL) techniques have recently shown considerable promise in addressing MRI reconstruction challenges, frequently surpassing conventional methods. However, their effectiveness for Dynamic Contrast-Enhanced MRI (DCE-MRI) - where precise estimation of pharmacokinetic parameter is critical - remains unclear. Goal(s): The goal is to compare DL-based reconstruction with established compressed sensing techniques to find out whether the superiority of DL-based reconstructions and their quantitative results still holds true for the more challenging DCE-MRI data. Approach: We use synthetic and in-vivo DCE-MRI data. Both DL (VarNet) and conventional (TTV-SENSE) methods will be applied, and their performance will be assessed using quantitative metrics. Impact: DCE-MRI is the most accurate tool for diagnosing breast cancer, but its potential is limited by acquisition techniques that cannot achieve high spatial and temporal resolution simultaneously. This work explores whether DL or conventional compressed sensing can overcome these limitations.