Dual-Energy Cone-Beam CT Using Two Complementary Limited-Angle Scans with A Projection-Consistent Diffusion Model
Abstract
Background: Dual-energy imaging on cone-beam CT (CBCT) scanners has great potential in different clinical applications, including image-guided surgery and adaptive proton therapy. However, the clinical practice of dual-energy CBCT (DE-CBCT) has been hindered by the requirement of sophisticated hardware components. Purpose: In this work, we aim to propose a practical solution for single-scan dual-energy imaging on current CBCT scanners without hardware modifications, using two complementary limited-angle scans with a projection-consistent diffusion model. Methods: Our approach has two major components: data acquisition using two complementary limited-angle scans, and dual-energy projections restoration with subsequent FDK reconstruction. Two complementary scans at different kVps are performed in a single rotation by switching the tube voltage at the middle of the source trajectory, acquiring the mixed-spectra projection in a single CBCT scan. Full-sampled dual-energy projections are then restored by a projection-consistent diffusion model in a slice-by-slice manner, followed by the DE-CBCT reconstruction using the FDK algorithm. Results: The proposed method was evaluated in a simulation study of digital abdomen phantoms and a study of real rat data. In the simulation study, the proposed method produced DE-CBCT images at a mean absolute error (MAE) of 20 HU. In the small-animal study, reconstructed DE-CBCT images using the proposed method gave an MAE of 25 HU. Conclusion: This study demonstrates the feasibility of DE-CBCT imaging using two complementary limited-angle scans with a projection-consistent diffusion model in both half-fan and short scans. The proposed method may allow quantitative applications of DE-CBCT and enable DE-CBCT-based adaptive proton therapy.
Cite
@article{arxiv.2403.11890,
title = {Dual-Energy Cone-Beam CT Using Two Complementary Limited-Angle Scans with A Projection-Consistent Diffusion Model},
author = {Junbo Peng and Chih-Wei Chang and Richard L. J. Qiu and Tonghe Wang and Justin Roper and Beth Ghavidel and Xiangyang Tang and Xiaofeng Yang},
journal= {arXiv preprint arXiv:2403.11890},
year = {2024}
}