This Demo is built to let others try to run the processing pipeline used in the abstract Toward a 1-minute high-resolution brain exam - MR Fingerprinting with ML-synthesized contrasts and fast reconstruction presented at the 2022 ISMRM Annual meeting in London (program number 53). You can find the abstract here (requires login).
This project is aimed at translating highly undrsampled MRF to a clinically feasible tool, by building a robust reconstruction pipeline that is portable between different compute systems (research lab, hospital, high performance computing cluster, collaborators, etc...). To achieve this these core objectives were set:
- The pipeline should run smoothly on multiple systems.
- The pipeline should be easy to upgrade when the sequence, the reconstruction method, or the synthesis method is changed.
- The pipeline should be able to provide an image to send back to the scanner within 5 min.
- The pipeline should be able to send a series of images to PACS within ~30min.
- The pipeline should run on hardware available in clinical settings (for now, this means an 11GB GPU).
This modular MRF processing pipeline includes 4 steps:
- Read raw scan data and metadata(In the demo this step is replaced with downloading a demo dataset)
- Reconstruct and get coil compression from calibration scan
- Reconstruct MRF (fast subspace basis reconstruction)
- Synthesize clinical contrasts
Each step will be demonstrated in a Google Colab session, and documentation on how to run the equivalent Docker containers on your own machine will be explained in the final section of the Jupyter notebook.