History

The SPECT forward and back-projection method implemented here is based on the 1992 paper by GL Zeng & GT Gullberg "Frequency domain implementation of the three-dimensional geometric point response correction in SPECT imaging" (DOI).

Historically there have been relatively few open-source libraries for SPECT image reconstruction, and to the best of our knowledge source code for the Zeng & Gullberg approach has not been available prior to this package.

Starting in about 1995, the ASPIRE library, developed in the early 1990s at the University of Michigan, began providing precompiled binaries (from C99 source code) for 2D SPECT image reconstruction with documentation.

In about 1997, a 3D version of ASPIRE for SPECT reconstruction became available, again as precompiled binaries with documentation. Anastasia Yendiki was a key contributor to the SPECT code. As noted in a 2001 technical report we took pains to ensure that the forward and back-projector were (adjoint) consistent pairs. Around 2001 the work was extended to consider blob basis functions, leading to a 2004 comparison paper.

Somewhere during that period the 3D SPECT projector / backprojector became available as precompiled MEX files for use with the Matlab version of the Michigan Image Reconstruction Toolbox.

The 3D version in ASPIRE precomputes rotated versions of the attenuation map, to save computation at the price of substantially more memory. That trade-off was reasonable in the era before machine learning. Today, with a focus on end-to-end training of image reconstruction methods in all modalities, including SPECT, it is desirable to have methods that use less memory to facilitate GPU implementations. This open-source Julia package is designed for the machine learning era.

Development work on this package is supported in part by the following projects led by Dr. Yuni Dewaraja:

• NIH Grant R01 EB022075
• NIH Grant R01 CA240706