Experiments run1

[melanie-dohmen-bayer]

Script to perform experiments with a config file

Usage:

examples/run_all.py config.yaml

supports loading data from folder of the following datasets:

• fastmri single coil knee
• fastmri multicoil brain
• fastmri prostate T2
• oasis
• cmrXRecon

Store reconstructed images in:

reports/experiments_run1/image_<dataset>_sample_<nr>_<distortion>_<reconstructor>_(un)corrected.tiff`

Store reference images in:

reports/<result_path>/image_<dataset>_sample_<nr>_reference.diff`

kspace images in:

reports/<result_path>/kspace_<dataset>_sample_<nr>_<distortion>.diff
reports/<result_path>/kspace_<dataset>_sample_<nr>_<distortion>.diff

Changes:

• save separate images instead of multi-image plots
• pass parameters, such as selection of datasets, reconstruction algrithms and distortions in a config file
• OasisCenteredFFTPath inherits from deepinv.physics.LinearPhysics and MRIMixin for cropping to reference image size
• Transform oasis centered k-space samples to uncentered (fast-MRI like) k-space and don't use centered FFT path, add tests for this
• Add fast-MRI prostate data by transforming reference images to fast-MRI-like k-space data, similar to oasis data
• adjust selection of algorithms with datasets (models trained with brain/knee datasets should not work on datasets of other regions, oasis model only works with OasisCenteredFFTPath, all other models work with DistortedKSpaceMultiCoilMRI)
• use original reference images, if available, crop reconstructed images to reference image size, if there is a difference
• adjusted fastmri_inference_plot.py accordingly
• add option "add_N4Correction": True/False to run_all.py and run Bias Field Correction (from ITK) on reference images and BiasField - distorted and reconstructed ones