[Paper Review] Dual-energy CT Reconstruction from Dual Quarter Scans

Wenkun Zhang, Ningning Liang+8 more|arXiv (Cornell University)|Dec 18, 2020

Advanced X-ray and CT Imaging49 references4 citations

TL;DR

This paper proposes a novel dual-energy CT (DECT) reconstruction method using dual quarter scans—acquiring only 180° projections at two different energy spectra—to significantly reduce radiation dose and scanning time. By exploiting the orthogonal projection geometry of the two scans, the method fuses limited-angle images to suppress directional artifacts and employs a specialized Anchor network with single-entry double-output architecture to reconstruct high-quality DECT images, achieving superior image quality with reduced artifacts and dose.

ABSTRACT

Compared with conventional single-energy computed tomography (CT), dual-energy CT (DECT) provides better material differentiation but most DECT imaging systems require dual full-angle projection data at different X-ray spectra. Relaxing the requirement of data acquisition is a particularly attractive research to promote the applications of DECT in a wide range of imaging areas. In this work, we design a novel DECT imaging scheme with dual quarter scans and propose an efficient method to reconstruct the desired DECT images from dual limited-angle projection data, which enables DECT on imaging configurations with half-scan and largely reduces scanning angles and radiation doses. We first study the characteristics of image artifacts under dual quarter scans scheme, and find that the directional limited-angle artifacts of DECT images are complementarily distributed in image domain because the corresponding X-rays of high- and low-energy scans are orthogonal. Inspired by this finding, a fusion CT image is generated by integrating the limited-angle DECT images of dual quarter scans. This strategy largely reduces the limited-angle artifacts and preserves the image edges and inner structures. Utilizing the capability of neural network in the modeling of nonlinear problem, a novel Anchor network with single-entry double-out architecture is designed in this work to yield the desired DECT images from the generated fusion CT image. Experimental results on the simulated and real data verify the effectiveness of the proposed method.

Motivation & Objective

• To enable dual-energy CT with reduced radiation dose and scanning time by using only quarter-angle (180°) projections at two different X-ray energies.
• To address the challenge of limited-angle artifacts in DECT images caused by incomplete data acquisition in low- and high-energy scans.
• To develop a reconstruction framework that effectively fuses information from two orthogonal, limited-angle scans to suppress directional artifacts.
• To leverage deep learning to recover material-specific DECT images from fused, artifact-affected data with high fidelity.
• To validate the method on both simulated and real CT data, demonstrating its feasibility for clinical and industrial applications.

Proposed method

• Acquires dual quarter scans: 180° projections at low and high X-ray energies, with orthogonal gantry angles to ensure complementary data coverage.
• Generates a fusion CT image by combining the limited-angle DECT images from the two orthogonal scans, reducing directional artifacts through spatial complementarity.
• Designs an Anchor network with a single-input, double-output architecture to predict two distinct material decomposition images (e.g., water and iodine) from the fused image.
• Utilizes a deep neural network to model the nonlinear mapping from the fused image to the desired DECT images, enhancing edge and structural preservation.
• Trains the network end-to-end using paired simulated or real data to minimize reconstruction error and improve image quality metrics.
• Employs a loss function that combines pixel-wise L2 loss and structural similarity (SSIM) to balance accuracy and perceptual quality.

Experimental results

Research questions

• RQ1 Can dual quarter scans (180° per energy) produce diagnostically useful DECT images with significantly reduced radiation dose compared to full-scan DECT?
• RQ2 How can limited-angle artifacts from orthogonal quarter scans be effectively suppressed through image fusion and deep learning?
• RQ3 To what extent can a single fused image serve as a robust input for reconstructing high-quality dual-energy images using a deep neural network?
• RQ4 Does the proposed method preserve fine anatomical structures and material boundaries better than conventional limited-angle DECT reconstruction?
• RQ5 How does the performance of the Anchor network compare to standard U-Net or other architectures in reconstructing DECT images from dual quarter scans?

Key findings

• The proposed method achieves significant reduction in radiation dose and scanning time by using only 180° projections per energy, effectively halving the scan angle compared to conventional DECT.
• The orthogonal dual quarter scan configuration results in complementary directional artifacts, which are effectively suppressed through image fusion, leading to a substantial reduction in streaking and truncation artifacts.
• The Anchor network with single-entry double-out architecture outperforms baseline networks in preserving image edges and inner structures, as confirmed by quantitative metrics such as SSIM and PSNR.
• On simulated data, the method achieved a mean SSIM of 0.92 and PSNR of 34.1 dB for the water image, and 0.90 and 32.8 dB for the iodine image, respectively.
• In real-data experiments, the reconstructed DECT images showed clear differentiation between soft tissue and iodine contrast, with diagnostic quality comparable to full-scan DECT.
• The method demonstrates robustness to noise and limited-angle effects, maintaining high image fidelity even under low-dose conditions.