Parallel qMRI Reconstruction

TL;DR

This project enhances the SPICER framework by applying a supervised learning approach to clinical qMRI datasets from Washington University School of Medicine. By redesigning the original Deep Unfolding architecture into streamlined, non-cascading U-Net and Attention U-Net models with automatic coil sensitivity estimation, we achieved a 4x reduction in parameters. Furthermore, we developed novel normalization techniques (ACS region-specific and coil-instance) for under-sampled k-space data, resulting in high-fidelity reconstructions with 37 dB PSNR and 0.923 SSIM from 4x accelerated acquisitions.

Paper

Parallel qMRI Reconstruction from 4x Accelerated Acquisitions
Mingi Kang
Washington University in St. Louis, Bowdoin College

Available on arXiv: 2511.18232

Citation

@article{kang2025parallel,
  title={Parallel qMRI Reconstruction from 4x Accelerated Acquisitions},
  author={Kang, Mingi},
  journal={arXiv preprint arXiv:2511.18232},
  year={2025}
}

Presentation & Poster

Presented at McKelvey School of Engineering Poster Palooza and Washington University Summer Research Symposium on July 2025. Awarded Best Poster Presentation at Poster Palooza.

Here are some photos from the summer in St. Louis!