Chapter 1 Magnetic resonance image re-parameterization on real data
Melika Abdollahi, Heidar Davoudi, Mehran Ebrahimi · 2025
Magnetic resonance (MR) image re-parameterization involves creating an MR image with a modified set of magnetic resonance imaging (MRI) scanning parameters. Adjusting these parameters produces varying contrasts between tissues, aiding in the detection of pathological changes. While multiple scans are often necessary for accurate diagnosis, obtaining them can be expensive, time-consuming and challenging for patients. As an alternative, MR image re-parameterization provides a method to predict and estimate contrast in imaging scans. In this work, we introduce an innovative model for MRI re-parameterization that combines autoencoders with cycle-generative adversarial networks. Unlike prior work that relied on synthetic data, we validate the method on Duke’s Breast Cancer MRI dataset, showing the effectiveness of the model in generating translated MR images with different acquisition parameters.