Tours

Standalone examples beyond the Course. Each names the lessons it assumes.

Acquisition and corrections

Corrections before and after reconstruction. Most are tested on a simulated artefact with a known truth; the EPI susceptibility tour uses measured data and an independent field map. These tours also need the following packages, and the susceptibility tour downloads about 2 MB of data into ~/.cache/bartorch-examples:

pip install PyHySCO

Tour	Contents
Readout oversampling and apodization	Readout oversampling removal; Fermi and Hann apodization
EPI Nyquist ghost and ramp sampling	EPI Nyquist ghost correction; ramp-sampling regridding
Receive bias field	Receive bias field estimation with N4
Gradient nonlinearity	Gradient nonlinearity correction from spherical-harmonic coefficients
Off-resonance correction of spiral imaging	Spiral off-resonance deblurring: multifrequency interpolation, time-segmented reconstruction
Rigid head motion from navigators	Rigid head motion from orthogonal navigator planes
Susceptibility distortion in EPI	EPI susceptibility distortion correction from reversed phase encoding
Noise prewhitening	Noise prewhitening and SNR

Dynamic and quantitative workflows

The Course’s encodings applied to a time series and to quantitative mapping. Non-Cartesian transform and Model-based reconstruction give the background.

Tour	Contents
Dynamic golden-angle radial MRI	Dynamic golden-angle radial reconstruction with temporal total variation
Subspace-constrained T1 mapping	Subspace \(T_1\) mapping from a continuous inversion-recovery scan
Parameter maps from scanner images	\(T_2\) mapping from DICOM magnitude images

Learning workflows

Training and use of learned regularizers beyond lessons 10 and 11. Learned reconstruction explains where a network enters a reconstruction. These tours also need:

pip install lightning torchio monai deepinv

Tour	Contents
Networks for complex volumes	Networks for complex multi-contrast volumes; patchwise execution
Staged training of an unrolled network	Staged training of unrolled networks
Training without a reference	Self-supervised training from undersampled data
Annealed plug-and-play	Plug-and-play with an annealed noise level
Uncertainty estimation	Calibrated voxel-wise uncertainty