[home][people][projects]

Multi-site MRI Harmonization

MRI scanners have measurable differences between units, even for exactly the same make/model. Unfortunately for clinical research, that means that we might be learning more about which hospital is which than any actual disease phenotype. Harmonization is the process of removing these unwanted site/scanner-wise signals.

As it turns out, this is a specific case of a general representation learning problem. Our work has provided a [deep] variational solution to this problem, and, using begun exploring this method’s use in harmonizing MRI, starting with multi-site HARDI (diffusion).

Publications:

1. Invariant Representations without Adversarial Training D Moyer, S Gao, R Brekelmans, G Ver Steeg, A Galstyan, NeurIPS 2018
2. Scanner Invariant Representations for Diffusion MRI Harmonization D Moyer, G Ver Steeg, CMW Tax, PM Thompson, Magnetic Resonance in Medicine 2020

Imaging for Total Joint Replacement

Publications:

1. Dilation-Erosion Methods for Radiograph Annotation in Total Knee Replacement Y Suh, A Mika, J R Martin, D Moyer, MIDL short paper track 2023
2. Label Augmentation Method for Medical Landmark Detection in Hip Radiograph Images Y Suh, P Chan, J R Martin, D Moyer, NeurIPS workshop 2023