PREDATOR-milestones

Deliverables for PREDATOR project milestones.

The notebooks folder contains browser-executable Jupyter notebook showcasing the deliverables planned for the milestone M1 of the project. Specifically:

• Milestone M1 - development of a computational framework for quantitative MR reconstruction.
  • M1a: implementation of data I/O routines.
  • M1b: implementation of MR encoding operators.
  • M1c: implementation of the main regularizers and solvers for iterative reconstruction.
  • M1d: implementation of a numerical Bloch simulator for synthetic data generation and parameter fitting.

Milestone M2 consisted in the creation of a database of quantitative maps (T1, T2, T2*) of at least 10 subjects to be used for synthetic data generation for DL-based reconstruction training.

The database was built as follows:

1. Whole-head quantitative data (MP2RAGE-based T1 mapping, MRF-based M0/T2 mapping and multi-echo GRE-based T2* mapping, resolution: 0.6-1.1 mm), together with high-resolution (0.8 mm) anatomical images (T1w and T2*w), were collected for 10 healthy subjects using 3T and 7T MRI scanners (GE HealthCare).
2. Anatomical images were resampled to 0.5mm, and a super-resolution template was built using ANTs (nonlinear registration).
3. Sub-millimetric quantitative parameter distributions were obtained by aligning the parametric maps to the anatomical template space.
4. A 25-subjects 0.5mm isotropic whole head quantitative dataset (PD, T1, T2, T2*) was built by aligning our anatomical template to the Open Science CBS Neuroimaging database and using our parameter distribution to create realistic randomized quantitative maps with different anatomical features.

The database can be freely downloaded on the PREDATOR project OSF repository.