Overview

Code for tomographic reconstruction built around the ASTRA toolbox (https://www.astra-toolbox.com/). This code implements model-based computed tomography (CT) reconstruction algorithms based on a Markov random field (MRF) prior and requires systems with atleast one GPU. The algorithms involve formulating and finding a minimum of the cost function of the form:

$$c(x) = l(y;Ax)+r(x)$$

where $l(y;Ax)$ is a data-fidelity term, $A$ is the tomographic projector and $r(x)$ is a regularizer based on the q-generalized Markov random field [1]. Currently, the package supports parallel beam tomography, laminography [2], arbitary view CT with a point-spread function as in cryo-EM [3] and conventional cone-beam CT for the choice of the $A$ matrix (see examples folder).

Requirements

astra toolbox : Core GPU based projection and back-projection

numpy, scipy, matplotlib, time, gc, concurrent, psutil, ctypes

pyqtgraph (optional): For displaying 3D volumes

gcc

Installation

1. Install the above packages (conda env create -f environment.yml --name pymbir python=3.7 creates a fresh conda environment with the dependencies installed)

2. Run conda activate pymbir

3. From the base folder, run python setup.py install

Getting started

We highly recommend starting with the examples in the examples/sim directory to develop an understanding of the pyMBIR package and how to set the different parameters.

References

[1] Thibault, Jean‐Baptiste, et al. "A three‐dimensional statistical approach to improved image quality for multislice helical CT." Medical physics 34.11 (2007): 4526-4544.

[2] Venkatakrishnan, Singanallur V., et al. "Model-based iterative reconstruction for neutron laminography." 2017 51st Asilomar Conference on Signals, Systems, and Computers. IEEE, 2017.

[3] Singanallur Venkatakrishnan, Puneet Juneja, Hugh O’Neill, “Model-based Reconstruction for Single Particle Cryo-Electron Microscopy”, Proc. of IEEE Asilomar Conference of Signals, Systems and Computer 2020

License

pyMBIR is distributed as open-source software under a GPL License (see the LICENSE file for details)