ResoNet: Noise-Trained Physics-Informed MRI Off-Resonance Correction
Authors: Alfredo De Goyeneche Macaya, Shreya Ramachandran, Ke Wang, Ekin Karasan, Joseph Y. Cheng, Stella X. Yu, Michael Lustig
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of our approach through results on phantom and in-vivo data. To evaluate the performance of our model, we acquired phantom data, as well as brain, knee, and abdominal in-vivo data using a GE 3T MR750W System. We compared the performance of our proposed approach and the DL Baseline by measuring the NRMSE (Normalized Root Mean Squared Error) and PSNR (Peak Signal to Noise Ratio) on the magnitude of the combined output images within the validation set. We report average metrics and their standard deviation over the 1000 examples in Table 1. |
| Researcher Affiliation | Academia | Alfredo De Goyeneche1, Shreya Ramachandran1, Ke Wang1, Ekin Karasan1, Joseph Cheng2, Stella X. Yu1,3, Michael Lustig1 1 Electrical Engineering and Computer Sciences, University of California, Berkeley 2 Radiology, Stanford University 3 Computer Science and Engineering, University of Michigan {adg, shreyar, kewang, ekinkarasan, stellayu, lustig}@berkeley.edu jycheng@stanford.edu, stellayu@umich.edu |
| Pseudocode | No | The paper describes the model architecture and training process but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is publicly available at: https://github.com/mikgroup/ResoNet |
| Open Datasets | Yes | We conducted an evaluation on a validation set created using the fast MRI brain dataset [30]. This set consists of brain anatomy images that were corrupted with simulated random off-resonance field variations and fat/water partial volume effects. The validation set comprised 1000 examples. |
| Dataset Splits | Yes | We conducted an evaluation on a validation set created using the fast MRI brain dataset [30]. This set consists of brain anatomy images that were corrupted with simulated random off-resonance field variations and fat/water partial volume effects. The validation set comprised 1000 examples. |
| Hardware Specification | Yes | We trained our model using Py Torch [27] in a Nvidia RTX3090 GPU, employing ℓ1 losses and the Adam optimizer [28]. |
| Software Dependencies | No | We trained our model using Py Torch [27] in a Nvidia RTX3090 GPU... We used the torchkbnufft library [29] for NUFFT operations. The paper mentions software but does not specify their version numbers (e.g., "PyTorch 1.9"). |
| Experiment Setup | Yes | The network architecture featured four unrolls. Each CNN module in these unrolls included three residual blocks, each containing two convolutional layers with 128 filters and a kernel size of 5, employing Re LU activation. The DC module used 12 conjugate gradient iterations with a trainable regularization parameter λ, empirically initialized to ensure that a ground truth input remained uncorrupted (λ = 1). We trained our model using Py Torch [27] in a Nvidia RTX3090 GPU, employing ℓ1 losses and the Adam optimizer [28]. |