Radiation Image Reconstruction and Uncertainty Quantification Using a Gaussian Process Prior

This repository contains example code for the paper, "Radiation Image Reconstruction and Uncertainty Quantification Using a Gaussian Process Prior." The package imager can be used for image reconstruction and uncertainty quantification methods presented in the paper. The included notebook demonstrates an example measurement scenario where an anisotropic Gaussian source and a ring-shaped source distribution are reconstructed in a single detector mapping.

Installation

To install in an editable mode, use

pip install -e .

Usage

The package imager contains sevaral modules useful for single detector mapping problems.

1. detector.py contains a class, SphericalDetector for simulating a sinlge spherical detector (i.e., isotropic response) with a custom intrinsic efficiency.
2. path.py include two classes: WalkingPath and RasterPath for random walking path and raster path simulation.
3. source.py for various distributed source classes, such as GaussianSource and RingGaussianSource.
4. scenario.py is used to hold a detector, a path and source objects. Then a Scenario object computes the system matrix of a measurement sceanrio. A sceanario object can also be used to plot a grount truth source distribution and count measurement.
5. imager.py contains the Imager class, which is primarily for image reconstruction and uncertainty quantification. The class provides the maximum likelihood expectation maximization (MLEM) and Gaussian Process Prior (GPP) image reconstruction methods. A The Bayesian uncertainty quantification using the Laplace approximation and preconditioned Crank-Nikolson MCMC.
6. utilities.py provides some useful functions.

The notebook file notebooks/example_1.ipynb shows how these different classes can be set up and used for image reconstruction and uncertainty quantificaiton.