Uncertainty Analysis in 3D SPECT Reconstruction based on Probabilistic Programming

TL;DR

This paper introduces a probabilistic programming approach using NUTS for 3D SPECT image reconstruction, providing uncertainty quantification that surpasses conventional methods, though computational efficiency needs enhancement.

Contribution

It applies probabilistic programming and NUTS sampling to 3D SPECT reconstruction, incorporating uncertainty estimation into the process.

Findings

Outperforms conventional methods in uncertainty quantification.

Provides detailed uncertainty information in reconstructed images.

Reconstruction time increases with larger phantom sizes.

Abstract

Single Photon Emission Computed Tomography (SPECT) is one of the nuclear medicine imaging modalities used for functional analysis of animal and human organs. Gamma rays emitted from the scanned body are filtered with collimators and detected by the SPECT head that is composed of an array of gamma detectors. The conventional reconstruction algorithms do not deem the uncertainty level that is associated with the field of view of SPECT collimators. In this paper, we incorporate the probabilistic programming approach for SPECT image reconstruction. No-U-Turn Sampler (NUTS) is used to estimate the scanned object system by considering uncertainty. The obtained results indicate that the presented work in 3D SPECT image reconstruction surpassed over the conventional reconstruction methods in terms of generating uncertainty information. However, the reconstruction time needs to be improved further for phantom sizes of 128x128x128 voxels and higher.