Update README.md

README.md

@@ -1,182 +0,0 @@
----
-license: mit
----
-
-# Model Card for SNRAware
-
-<!-- Provide a quick summary of what the model is/does. -->
-
-A deep learning imaging AI model with imaging transformer, for MR denoising.
-
-![image](https://cdn-uploads.huggingface.co/production/uploads/690cf6928b7eacea549fd405/ohWF8O1xe0rOJWvhXmxLf.png)
-
-## Model Details
-
-SNRAware is an imaging transformer model trained to denoise complex MR image data. 
-Imaging transformers use attention modules to capture local, global, and inter-frame 
-signal and noise characteristics. Denoising training used the SNRAware method, 
-generating MR-realistic noises on the fly to create low SNR samples with unitary noise scaling. 
-Model received low SNR complex images and g-factor maps as input, producing high SNR complex images as output. 
-
-Please refer to the publication for technical details.
-
-### Model Description
-
-<!-- Provide a longer summary of what this model is. -->
-
-The published model was instantiated with a high-res net (HRnet) backbone and consists of multiple imaging attention modules. 
-Two models were published: 
-  - SNRAware-small: a 27.7 million parameter model
-  - SNRAware-medium: a 55.1 million parameter model  
-  - SNRAware-large: a 109 million parameter model
-
-Input to model is 5D tensor [B, C, T/F, H, W] for batch, channel, time/frame, height and width. Output tensor is in the shape of 
-[B, C-1, T/F, H, W]. The last channel in input is the g-factor map.
-
-- **Developed by:** Microsoft Research, Health Futures
-- **Model type:** Imaging Transformer
-- **License:** MIT
-
-### Model Sources
-
-<!-- Provide the basic links for the model. -->
-
-- **Repository:** https://github.com/microsoft/SNRAware
-- **Paper:** https://pubs.rsna.org/doi/10.1148/ryai.250227
-
-## Uses
-
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-
-The model should accept the reconstructed MR complex images and g-factor maps and produce the denoising images. 
-The model expects the unitary noise level in the input images. This can be achieved by reconstruction images with 
-Gadgetron framework.
-
-### Direct Use
-
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-
-Denoise the complex MR images.
-
-### Out-of-Scope Use
-
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-
-No out-of-scope use should be attempted. Input to the model should be SNRUnit reconstructed.
-
-## Bias, Risks, and Limitations
-
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-
-No bias and risks are associated with this model. Only limitation is input data should have unitary noise scaling.
-
-### Recommendations
-
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-
-None
-
-## How to Get Started with the Model
-
-Please refer to the documentation in https://github.com/microsoft/SNRAware to get started.
-
-## Training Details
-
-### Training Data
-
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-
-Training data is not shared.
-
-### Training Procedure
-
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-Please refer to the publication for training details.
-
-#### Training Hyperparameters
-
-- **Training regime:** fp32 <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-
-#### Speeds, Sizes, Times
-
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-
-For a typical MR image data, the inference time on H100 is ~3-7s for SNRAware-small and ~5-11s for SNRAware-medium.
-
-## Evaluation
-
-Both in-distribution and generalization tests were performed. For the in-distribution tests, the MR noise was generated and added to clean images to lower its SNR.
-Model was applied to restore the image quality. PSNR and SSIM were computed against the ground-truth. For the generalization tests, new data outside the training cohort
-were acquired for different imaging sequences, field strength, anatomies and resolution. Model outputs were compared to raw inputs and scored by clinicians to judge quality.
-
-Please refer to the publication for evaluation details.
-
-### Results
-
-Imaging transformer model outperformed competing model architectures in the in-distribution tests. 
-The SNRAware training also enabled imaging transformer models to generalize well to unseen applications without further training.
-
-#### Summary
-
-
-## Environmental Impact
-
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-
-- **Hardware Type:** B200
-- **Hours used:** 100
-- **Cloud Provider:** Azure
-- **Compute Region:** westus2
-- **Carbon Emitted:** 7.5
-
-### Model Architecture and Objective
-
-Imaging Transformer with a high-res net backbone
-
-### Compute Infrastructure
-
-Azure GPU VMs
-
-#### Hardware
-
-NVIDIA B200 x16
-
-#### Software
-
-Pytorch 2.8.0+cu128
-
-## Citation
-
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-
-**BibTeX:**
-
-```latex
-@article{
-    doi:10.1148/ryai.250227,
-    author = {Xue, Hui and Hooper, Sarah M. and Pierce, Iain and Davies, Rhodri H. and Stairs, John and Naegele, Joseph and Campbell-Washburn, Adrienne E. and Manisty, Charlotte and Moon, James C. and Treibel, Thomas A. and Hansen, Michael S. and Kellman, Peter},
-    title = {SNRAware: Improved Deep Learning MRI Denoising with Signal-to-noise Ratio Unit Training and G-factor Map Augmentation},
-    journal = {Radiology: Artificial Intelligence},
-    volume = {0},
-    number = {ja},
-    pages = {e250227},
-    year = {0},
-    doi = {10.1148/ryai.250227},
-    note ={PMID: 41123451},
-    URL = {https://doi.org/10.1148/ryai.250227}
-}
-
-@misc{xue2024imagingtransformermridenoising,
-      title={Imaging transformer for MRI denoising with the SNR unit training: enabling generalization across field-strengths, imaging contrasts, and anatomy}, 
-      author={Hui Xue and Sarah Hooper and Azaan Rehman and Iain Pierce and Thomas Treibel and Rhodri Davies and W Patricia Bandettini and Rajiv Ramasawmy and Ahsan Javed and Zheren Zhu and Yang Yang and James Moon and Adrienne Campbell and Peter Kellman},
-      year={2024},
-      eprint={2404.02382},
-      archivePrefix={arXiv},
-      primaryClass={eess.IV},
-      url={https://arxiv.org/abs/2404.02382}, 
-}
-```
-
-## Model Card Contact
-
-Hui Xue, xueh@microsoft.com