Assessment of Dose-Reduction Strategies in Wavelength-Selective Neutron Tomography

Abstract

This study aims to determine an acquisitional and computational workflow that yields the highest quality spatio-spectral reconstructions in four-dimensional neutron tomography studies. The properties of neutrons enable unique image contrast modes, but accessing these modes requires defining the energy of the neutron beam, resulting in long acquisition times. We seek sparse angular tomography approaches to collect of order 100 tomograms at different neutron wavelengths using the minimum number of input projection images. In these computational image workflows, we identified and evaluated the main factors affecting the quality of the tomographic reconstruction such as the projection number, the reconstruction method, and the post-processing method and we report relationships between 3D reconstruction quality metrics and acquisition time. Based on these relationships, the performance of seeded simultaneous iterative reconstruction-based techniques (SIRT and SIRT with total variation regularization) yielded improved image quality and more accurate estimates of the reconstructed attenuation values compared to other methods, which included convolutional neural networks. The methods were then applied to a dose-reduced monochromatic dataset and characterized via signal-to-noise ratio (SNR) and single-voxel resolution.