Accuracy and Uncertainty of Measurements from 3D Tomographic Reconstructions

Summary

NIST has invested into an Innovations in Measurement Science (IMS) project that seeks to develop far field neutron interferometers to create multi-scale images (three-dimensional small-angle neutron scattering) and perform precision measurements of fundamental constants and beyond standard model physics. One of the many challenges of this project is the 3D measurement inference about object properties from neutron imaging interferometry. The project, called INFER due to its dependency on inference and interferometry, must include computational solutions for

• (a) high throughput data stream (~2TB per day),
• (b) phase image reconstruction,
• (c) 3D tomographic reconstruction,
• (d) 3D volumetric image segmentation, and
• (e) pair-correlation function analysis.

There is a need to understand the accuracy and uncertainty of image-based 3D measurements derived from this new far field neutron interferometer using the cutting-edge artificial intelligence (AI) based models.

Description

To address the need, we focus on the following goals:

1. Imaging dose reduction and signal-to-noise (SNR) maximization for 3D tomographic reconstructions using AI-based models
2. Uncertainty estimation of 3D measurements from 3D tomographic reconstructions due to camera defects and dose reduction
3. Automatic volume segmentation using calibration phantoms and AI-based models
4. Pair correlation function fitting using multiple implementations of the Sine Fourier transform of the total (Bragg and diffuse) scattering
5. Multi-scale interactive volumetric image visualization at Gigapixel image scale

Our objectives are to minimize the neutron imaging acquisition time and maximizing the quality of 3D reconstructions and the accuracy of 3D measurements derived from the 3D reconstructions. The challenges of such objectives lie in a complex computational workflow between acquired 2D projection images and 3D reconstructed images used for deriving 3D measurements. The problem is in determining a computational workflow (methodology) that will yield the highest quality of 3D reconstructions with the minimum number of 2D projection images (i.e., limited imaging acquisition time) as illustrated in Figure 1. Since the advances in neutron imaging technology for quicker 3D imaging are limited by the flux of neutrons, we explore dose-reduction and utilization of prior knowledge to decrease the time required for tomographic reconstructions of high-quality 3D images.

Figure 1: A computational workflow to transition from acquired 2D projection images to 3D reconstructed images includes many steps contributing to accuracy and uncertainty of derived 3D measurements.

Our success criteria are defined as tradeoffs between acquisition time and 3D measurement metrics (accuracy and uncertainty) for a range of applications. The applications include measurements of (a) alkali-silica concrete degradation, (b) ion mobility in battery and fuel cell electrodes, (c) macromolecule transport for drug delivery and structural biology to improve implants, and (d) fluid transport in geology. Our approach is leveraging the capabilities of the new far field neutron interferometer to characterize a range of length scales, such as porosity over nanometer to centimeter scales in geological systems.