Research

Variational diffusion models for blind MRI inverse problems

Diffusion models have demonstrated state-of-the-art results in solving inverse problems in various domains including medical imaging. However, existing works generally consider the cases where the forward operator is fully known. Therefore, blind inverse problems with unknown forward operator parameters require modifications on existing methods. In this work, we present an extension of the recently developed regularization by denoising diffusion process (RED-diff) algorithm to blind inverse problems. Similarly to RED-diff, our method can reconstruct images without model re-training or fine-tuning for arbitrary acquisition settings. Tested in fieldmap-corrected MR image reconstruction, our blind RED-diff framework can successfully approximate the unknown forward model parameters and produce fieldmap-corrected reconstructions accurately.

Oscanoa JA*, Alkan C*, Abraham D, Nurdinova A, Gao M, Setsompop K, Pauly JM, Mardani M, Vasanawala SS, 2023. Variational diffusion models for MRI blind inverse problems. NeurIPS 2023 Workshop on Deep Learning and Inverse Problems, New Orleans, USA. https://openreview.net/forum?id=Cei9ee2zfJ. *Equal contribution.

Coil sketching for computationally-efficient MR iterative reconstruction

Parallel imaging and compressed sensing reconstruction of large datasets has a high computational cost, especially for 3D non-Cartesian acquisitions. This work is motivated by the success of iterative Hessian sketching methods in machine learning. Herein, we develop Coil Sketching to lower computational burden by effectively reducing the number of coils actively used during iterative reconstruction. Tested with 2D radial and 3D cones acquisitions, our method yields considerably faster reconstructions (around 2x) with virtually no penalty on reconstruction accuracy.

Oscanoa JA, Ong F, Iyer SS, Li Z, Sandino CM, Ozturkler B, Ennis DB, Pilanci M, Vasanawala SS. Coil Sketching for computationally-efficient MR iterative reconstruction. Magnetic Resonance in Medicine.

Coil-sketched unrolled networks for computationally-efficient deep MRI reconstruction

Deep unrolled networks can outperform conventional compressed sensing reconstruction. However, training unrolled networks has intensive memory and computational requirements, and is limited by GPU-memory constraints. We propose to use our previously developed “coil-sketching” algorithm to lower the computational burden of the data consistency step. Our method reduced memory usage and training time by 18% and 15% respectively with virtually no penalty on reconstruction accuracy when compared to a state-of-the-art unrolled network.

Oscanoa JA, Ozturkler B, Iyer SS, Li Z, Sandino CM, Pilanci M, Ennis DB, Vasanawala SS. Coil-sketched unrolled networks for computationally-efficient deep MRI reconstruction. Poster presentation at ISMRM 30th Annual Meeting, London, United Kingdom, 2022

Deep learning-based reconstruction for accelerated 2D phase contrast MRI

We propose a modified DL-ESPIRiT reconstruction framework for 2D PC-MRI, comprised of an unrolled neural network architecture with a Complex Difference estimation (CD-DL). CD-DL was trained on 155 fully sampled 2D PC-MRI pediatric clinical datasets. The fully sampled data (n=29) was retrospectively undersampled (6–11x) and reconstructed using CD-DL and a parallel imaging and compressed sensing method (PICS). Measurements of peak velocity and total flow were compared to determine the highest acceleration rate that provided accuracy and precision within (+/- 5%). Feasibility of CD-DL was demonstrated on prospectively undersampled datasets acquired in pediatric clinical patients (n=5) and compared to traditional parallel imaging (PI) and PICS.

The retrospective evaluation showed that 9x accelerated 2D PC-MRI images reconstructed with CD-DL provided accuracy and precision within +/-5%. CD-DL showed higher accuracy and precision compared to PICS for measurements of peak velocity and total flow. The prospective feasibility study showed that CD-DL provided higher accuracy and precision than PICS for measurements of peak velocity and total flow.

Oscanoa JA*, Middione MJ*, Syed AB, Sandino CM, Vasanawala SS, Ennis DB, 2022: Accelerated two-dimensional phase-contrast for cardiovascular MRI using deep learning-based reconstruction with complex difference estimation. Magnetic Resonance in Medicine. doi: 10.1002/mrm.29441. *Equal contribution