Tomosipo algorithms

A collection of common tomographic reconstruction algorithms implemented using the tomosipo package.

The following algorithms are implemented:

• FBP
• FDK
• SIRT
• tv_min: Total-variation regularized least squares with the Champbolle Pock method
• nag_ls: l2-regularized least squares with the Nesterov accelerated gradient descent(nag) method

Free software: GNU General Public License v3

Getting Started

It takes a few steps to setup Tomosipo algorithms on your machine. We recommend installing Anaconda package manager for Python 3.

Installing with conda

Install with:

conda create -n tomosipo tomosipo pytorch==2.0.1 pytorch-cuda=11.7 tqdm -c pytorch -c nvidia -c astra-toolbox/label/dev -c aahendriksen -c defaults

conda activate tomosipo
pip install git+https://github.com/ahendriksen/ts_algorithms.git

Running

import torch
import tomosipo as ts
from ts_algorithms import fbp, sirt, tv_min2d, fdk, nag_ls

# Setup up volume and parallel projection geometry
vg = ts.volume(shape=(1, 256, 256))
pg = ts.parallel(angles=384, shape=(1, 384))
A = ts.operator(vg, pg)

# Create hollow cube phantom
x = torch.zeros(A.domain_shape)
x[:, 10:-10, 10:-10] = 1.0
x[:, 20:-20, 20:-20] = 0.0

# Forward project
y = A(x)

# reconstructions made with different algorithms
rec_fbp = fbp(A, y)
rec_sirt = sirt(A, y, num_iterations=100)
rec_tv_min = tv_min2d(A, y, 0.0001, num_iterations=100)
rec_nag_ls = nag_ls(A, y, num_iterations=100)

Authors and contributors

• Allard Hendriksen - Initial work
• Dirk Schut - FDK and nag_ls implementations, current maintainer

See also the list of contributors who participated in this project.

How to contribute

Contributions are always welcome. Please submit pull requests against the master branch.

If you have any issues, questions, or remarks, then please open an issue on GitHub.

License

This project is licensed under the GNU General Public License v3 - see the LICENSE.md file for details.