Update README.md

README.md

@@ -2,8 +2,122 @@ PSIRNet
 
 Last Updated: 06-APR-2026
 
-Model summary
-
-| Developer | Microsoft Corporation, Authorized representative: Microsoft Ireland Operations Limited 70 Sir John Rogerson’s Quay, Dublin 2, D02 R296, Ireland |
+| Model summary | |
 | -------- | -------- |
-| | |
+| Developer | Microsoft Corporation, Authorized representative: Microsoft Ireland Operations Limited 70 Sir John Rogerson’s Quay, Dublin 2, D02 R296, Ireland |
+| Description | PSIRNet accelerates PSIR LGE cardiac MR imaging by eightfold or more while preserving diagnostic image quality. |
+| Model architecture | This model is an instantiation of the variational network architecture (an unrolled method used to solve inverse problems). |
+| Parameters | 845 million trainable parameters |
+| Inputs | Inputs to model are four 4D tensors [B, C, H, W] for batch, channel, height and width. |
+| Context length | Not applicable |
+| Outputs | Output tensor is in the shape of [B, C, H, W]. |
+| Public data summary (or summaries) | Not applicable |
+| Training Dates | Aug 2025  |
+| Release date; Release date in the EU (if different) | Mar 2026  |
+| License | MIT license |
+| Model dependencies: | N/A |
+| List and link to any additional related assets | N/A |
+| Acceptable use policy | N/A |
+
+# Model overview
+   
+PSIRNet is a physics-guided, end-to-end deep learning reconstruction model for free-breathing late gadolinium enhancement (LGE) cardiac MRI that produces a phase-sensitive inversion recovery (PSIR) image (with surface coil correction) from a single interleaved inversion-recovery/proton-density (IR/PD) acquisition, replacing the typical MOCO PSIR workflow that relies on 8–24 averages.
+
+# Usage
+## Primary use cases
+This model is only suited to reconstruct PSIR LGE cardiac MR images. 
+## Out-of-scope use cases
+This model is for research use only. Any clinical or medical decision-making use is out of scope.
+## Distribution channels
+Model source code is available at https://github.com/microsoft/psirnet
+Pre-trained models are available at https://huggingface.co/microsoft/psirnet
+## Input formats
+Inputs to model are four 4D tensor [B, C, H, W] for batch, channel , height and width. 
+## Technical requirements and integration guidance
+Recommend GPU should have >=16GB memory. NVIDIA A100 or newer GPUs are the best.
+## Responsible AI considerations
+This model is for the very specific use case described above. Only domain experts with good knowledge of MR imaging should deploy the model to conduct research. To our knowledge, there are no limitations or risks associated with the model in terms of fairness, representation, or offensive content. PSIRNet is not generative.
+
+# Quality and performance evaluation
+The model was evaluated quantitatively on an external test set using image-similarity metrics (SSIM, PSNR, and NRMSE). Qualitative analysis was performed by two expert cardiologists using a 5-point Likert scale. 
+Model outputs were compared to clinical target images. The standard quality metrics were computed.
+
+# Data overview
+## Training, testing, and validation datasets
+### Size of dataset and characteristics
+- A Text training data size:  Not applicable. Text data is not part of the training data
+- Text training data content: Not applicable.
+- Image training data size: Less than 1 million images
+- Image training data content: 
+The PSIRNet model is trained on public data from the National Institutes of Health Cardiac MRI Raw Data Repository, hosted by the Intramural Research Program of the National Heart Lung and Blood Institute, were curated with the required ethical and/or secondary audit use approvals or guidelines permitting the retrospective analysis of anonymized data without requiring written informed consent for secondary usage for the purpose of technical development, protocol optimization, and/or quality control. 
+The data was fully anonymized and used for training without exclusion. Raw data comprising 800,653 slices from 55,917 patients, acquired on 1.5-T and 3-T MRI scanners across multiple sites from 2016 to 2024, were used in this retrospective study. Data were split by patient: 640,000 slices (42,822 patients) for training and the remainder for validation and testing, without overlap.
+
+### A Audio training data size: 
+Not applicable. Audio data is not part of the training data
+
+### Audio training data content: 
+Not applicable 
+
+### Video training data size: 
+Not applicable. Video data is not part of the training data
+
+### Video training data content: 
+Not applicable 
+
+### Other training data size: 
+Not applicable 
+
+### Other training data content: 
+Not applicable 
+
+## Latest date of data (acquisition/collection for model training): 
+31-AUG-2025
+
+## Is data collection ongoing to update the model with new data collection after deployment? 
+No 
+
+## Date the training dataset was first used to train the model: 
+AUG 2025
+
+## Rationale or purpose of data selection: 
+Publicly available dataset was selected as it allows for replication of scientific findings.
+
+# List of Data Sources
+
+## Publicly available datasets 
+
+### Have you used publicly available datasets to train the model?   
+Yes 
+
+## Private non-publicly available datasets obtained from third parties 
+N/A
+
+### Datasets commercially licensed by rightsholders or their representatives 
+Have you concluded transactional commercial licensing agreement(s) with rightsholder(s) or with their representatives? No
+
+### Private datasets obtained from other third parties 
+- A Have you obtained private datasets from third parties that are not licensed as described in Section 2.2.1, such as data obtained from providers of private databases, or data intermediaries? 
+No
+
+## Personal Data
+- Was personal data used to train the model? Microsoft follows applicable laws and best practices pertaining to personal data.
+No
+
+## Synthetic data 
+- Was any synthetic AI-generated data used to train the model? 
+No 
+
+## Data processing aspects 
+### Respect of reservation of rights from text and data mining exception or limitation 
+- Does this dataset include any data protected by copyright, trademark, or patent? Microsoft follows applicable laws and best practices for processing data protected by copyright, trademark, or patent.
+No
+
+### Other information 
+- Does the dataset include information about consumer groups without revealing individual consumer identities? Microsoft follows applicable laws and best practices for protecting consumer identities.
+- Was the dataset cleaned or modified before model training? 
+No 
+
+# Contact
+Requests for additional information can be directed to MSFTAIActRequest@microsoft.com.
+Authorized representative: Microsoft Ireland Operations Limited 70 Sir John Rogerson’s Quay, Dublin 2, D02 R296, Ireland
+