biasgen 0.1.0

Utility functions to produce and visualize simulated bias fields generated according to Kern et. al's sinusoidal sensitivity model.

biasgen

biasgen is a Python package which simulates magnetic resonance (MR) non-uniformities. This package allows users to generate custom bias fields by providing radiofrequency (RF) coil spatial information as well as frequency sampling information.

Bias fields are constructed from a sum-of-squares approach using sensitivity maps which follow the sinusoidal model of Kern et. al. Direct computation of sinusoidal sensitivities are done using the closed form solution of segment sources derived by Vinas and Sudhyadhom.

Installation

Package biasgen can be installed using pip:

pip install biasgen

A CUDA accelerated version of biasgen is also provided and can be installed with:

pip install biasgen[gpu]

Usage

biasgen requires a set of coil positions and sampling frequencies in order to work. Coil positioning can be initialized using several CoilSegment objects or through the predefined constructor routine cage_constructor. An example using cage_constructor is provided below:

import biasgen

ph = biasgen.phantom3d(n=128)
coils = biasgen.cage_constructor(n_coils=2, center=(64,64,64), coil_height=128,
                                 length_to_space_ratio=0.35, ellip_axes=(90,65))
biasgen.view_coil_positioning(ph, coils)

Here function view_coil_positioning provides a top-down view of the coils defined by our cage constructor. Next is to define a sampling grid for the sinusoidal sensitivity model:

sens_settings = biasgen.SensitivitySettings(grid_lengths=(5,5,5), grid_spacings=(1,1,1))

These sampling settings can be used in conjunction to the coils to produce a set of sensitivity maps:

# biasgen.use_gpu(True) # Uncomment if GPU is available
sens = biasgen.compute_sensitivity(coils, sens_settings, ph.shape, batch_sz=1, scale_fctr=0.5)

Arguments "batch_sz" and "scale_fctr" can be helpful for memory-limited devices. "batch_sz" determines how many segments are loaded into memory while "scale_fctr" provides a temporary spatial downsampling during computation. The sensitivity maps can be viewed using function view_center_axes and a boolean mask:

ph_mask = ph > 0
biasgen.view_center_axes(abs(sens), ph_mask, ['Z-slice','Y-slice','X-slice'])

Finally the bias field is constructed through a sum-of-squares procedure:

bias = biasgen.bias_sum_of_squares(sens)
biasgen.view_center_axes(bias*ph, ph_mask, ['Z-slice','Y-slice','X-slice'])

More detailed examples can be found in the examples/bias.ipynb notebook.

References

1. Guerquin-Kern M, Lejeune L, Pruessmann KP, Unser M. Realistic Analytical Phantoms for Parallel Magnetic Resonance Imaging. IEEE Transactions on Medical Imaging. 2012;31(3):626-636.
2. TBD