Optimization-based phase retrieval for material decomposition with multi-energy computed tomography

TL;DR

This paper introduces a 3D optimization-based method for material decomposition in multi-energy phase-contrast CT, improving accuracy over traditional analytical approaches and demonstrating benefits of phase contrast in simulations.

Contribution

It presents a novel, natively 3D, optimization-based approach leveraging automatic differentiation for improved phase retrieval in multi-energy CT.

Findings

Quantitative improvement over existing analytical methods

Enhanced material differentiation with phase contrast

Validation through simulation studies

Abstract

Multi-energy CT has long demonstrated its ability to enhance image quality with material decomposition. Yet, it has largely been limited to applications that already have high contrast. More recently, x-ray phase-contrast (XPC) imaging has gained interest for its potential to improve detectability in tasks lacking such contrast. Previous work has demonstrated the benefit of combining multi-energy imaging with XPC for material decomposition. While the existing method is promising, its analytical approach requires several approximations that limit its broad applicability, and it is based on projection imaging, requiring separate tomographic reconstruction for three-dimensional (3D) volumes. We propose a natively 3D optimization-based solution that leverages modern computational advances, namely automatic differentiability, to efficiently solve the multi-energy CT phase retrieval problem with a more accurate forward model. In a simulation study, we quantitatively demonstrate the improvement of our proposed technique relative to the existing analytical standard, and we affirm the benefits of phase contrast over traditional absorption-only x-ray imaging.