Skip to main content

High quality spectral CT using sparse reconstruction methods

Periodic Reporting for period 1 - SUCCESS (High quality spectral CT using sparse reconstruction methods)

Reporting period: 2016-05-01 to 2018-04-30

Spectral CT (SCT), also called “color CT”, provides energy-dependent information, which could translate into higher contrast and material decomposition capabilities, among other benefits, not only improving conventional CT but opening new possibilities for medical diagnosis. This project contemplates two major applications. The first one is material decomposition for K-edge imaging using contrast agents such as iodine, gadolinium, gold or bismuth. The second application is early detection of knee osteoarthritis (OA), as there are not current methods for visualizing with sufficiently high resolution and contrast the whole joint.

SCT has high potential but few challenges need to be addressed before reaching clinical use. 1) Image reconstruction in spectral CT is a nonlinear, nonconvex, multidimensional inverse problem SCT, so using image reconstruction with methods valid for standard CT leads to low image quality. 2) SCT must be validated using experimental preclinical and clinical data, but such systems are currently only available at the stage of prototypes and there is a lack of an energy-based gold standard. 3) Clinical applications must be also identified and validated.

The objective of this proposal is to provide and optimize new algorithms for SCT that will be designed and validated for specific high-impact high-potential applications. According to the general objective, we identify three specific objectives: 1) Research and develop algorithms that optimizes image quality of spectral CT images. 2) Build gold standard images created with monochromatic synchrotron radiation data in order to assess spectral CT. 3) Investigate the feasibility of the proposed method to improve image quality for two applications: k-edge imaging and early detection of osteoarthritis.
The image reconstruction problem in SCT can be divided in to two separated steps: a nonlinear material decomposition step, which decomposes the energy-based data into material-based data, and a linear tomographic reconstruction step. In this work, we investigated robust regularization methods for the material decomposition problem. In [1], we proposed a material-dependent spatial regularization method for material decomposition and evaluated it using a realistic numerical thorax phantom. The proposed method, named regularized weighted least squares Gauss–Newton algorithm (RWLS-GN), improved image quality and contrast-to-noise ratio of the gadolinium image with respect to reference maximum likelihood Nelder–Mead (ML-NM) algorithm, which was very sensitive to noise. In addition, RWLS-GN was 70 times faster than ML-NM. RWLS-GN allowed material decomposition with a number of incident photons equal or larger than 10^5 and with a marker concentration equal or larger than 0.03 gcm^-3.

In [2], we studied nonconvexity and proposed an iterative scheme based on the Bregman distance. We first proved the existence of a convex set where the usual data fidelity term is convex, which provides a guideline for selecting a good initial guess when using convex optimization methods. Using numerical simulations, we showed that the proposed Bregman scheme is robust to the selection of the initial guess. The improvement in global convergence of Bregman iterative scheme combined with other interesting properties of the Bregman distance appears as a compelling strategy for solving nonlinear inverse problems. Fig. 1 shows decomposed images by the proposed method for different number of incident photons. Preliminary results were presented at “IMA Conference on Inverse Problems from Theory to Application 2017” [3] and at “Recherche en Imagerie et Technologies pour la Santé (RITS) 2017”, [4].

On-going work focus on validating these algorithms using experimental data from a clinical spectral CT scanner.

This work has also contributed to other publications for the development of sparse reconstruction methods and multi-dimensional regularization for X-ray imaging [5], [6] and for other imaging modalities [7]-[11]. The work [5] was presented at “SPIE - Developments in X-Ray Tomography XI, 2017”.

We carried out the first experimental validation of SCT using monochromatic synchrotron radiation data. This is part of a competitive national project (MD-1045) that aimed to acquire monochromatic X-ray radiation data at the European Synchrotron Radiation Facility (ESRF), Grenoble. Tissue characterization phantoms and biological samples were acquired both at ESRF beamline ID17, Grenoble, and at CERMEP, Lyon, which holds a Philips spectral CT scanner. Phantom studies show a linear relationship between the two modalities. Processing and analysis is in progress. Preliminary analysis of biological data show the unique, great potential of X-ray energy-based imaging for imaging osteoarthritis (images from ESRF experiment MD-1045 in Fig. 2).

[1] N Ducros et al, Med Phys: 44(9), e174-e187, 2017
[2] JFPJ Abascal et al, hal-01621265 (preprint), 2018
[3] JFPJ Abascal et al, IMA Conference on Inverse Problems from Theory to Application, Cambridge, UK, 2017
[4] JFPJ Abascal et al, Recherche en Imagerie et Technologies pour la Santé (RITS) 2017, Lyon, France. hal-01505326
[5] JFPJ Abascal et al, Proc. SPIE 10391, Developments in X-Ray Tomography XI, San Diego, USA, 2017 (invited speaker)
[6] C Goubet et al, Proc. ISBI, IEEE International Symposium on Biomedical Imaging, Washington, DC, USA, 2018
[7] JFPJ Abascal et al, IEEE Trans Med Imaging, 37(2): 547 – 556, 2017
[8] F Li et al, IEEE Sensors Journal, 17 (4): 976-985, 2016
[9] E Al Hosani et al, Measurement Science and Technology, 27(11): 115402, 2016
[10] B Chen et al, Sensors, 18: 1704, 2018
[11] C Chittenden et al, IEEE Sensors Journal (in press), 2018
This project contributes with several aspects to the development of a new imaging modality. SCT is currently relying on algorithms that are valid for standard CT. This work provides new algorithms specifically designed to account for the nonlinearity and multidimensional nature of spectral CT. Proposed algorithms have led to significantly superior image quality and quantification with respect to the state-of-the art on numerical phantoms. Work is currently being done to assess their improvement on experimental data. Best performing methods are likely to be included in the new generation of spectral CT scanners.

Validating SCT is mandatory to understand the benefit and assessed improvements in sensitivity, resolution and image quality. The ongoing synchrotron radiation experiments provides a novel and accurate methodology to assess and validate SCT for different applications, as well as unique reference data.

OA is the most common chronic condition of the joints, leading to significant disability, reduction in quality of life, and high health care costs. Spectral CT is a unique imaging modality with the potential to depict all components of the joint. This is the first complete study that investigates x-rays based energy-encoded information for the diagnosis of OA. Preliminary results for the study of the feasibility of SCT for osteoarthritis are encouraging and have led to obtaining three national projects MD-1045, MD-1142, and ANR “SALTO”. In new future, positive results from this work will lead to further research such as in-vivo experiments on patients.