Deformable Known Component Model-Based Reconstruction for Coronary CT Angiography

Abstract

Purpose:

Atherosclerosis detection remains challenging in coronary CT angiography for patients with cardiac implants. Pacing electrodes of a pacemaker or lead components of a defibrillator can create substantial blooming and streak artifacts in the heart region, severely hindering the visualization of a plaque of interest. We present a novel reconstruction method that incorporates a deformable model for metal leads to eliminate metal artifacts and improve anatomy visualization even near the boundary of the component.

Methods:

The proposed reconstruction method, referred as STF-dKCR, includes a novel parameterization of the component that integrates deformation, a 3D-2D preregistration process that estimates component shape and position, and a polyenergetic forward model for x-ray propagation through the component where the spectral properties are jointly estimated. The methodology was tested on physical data of a cardiac phantom acquired on a CBCT testbench. The phantom included a simulated vessel, a metal wire emulating a pacing lead, and a small Teflon sphere attached to the vessel wall, mimicking a calcified plaque. The proposed method was also compared to the traditional FBP reconstruction and an interpolation-based metal correction method (FBP-MAR).

Results:

Metal artifacts presented in standard FBP reconstruction were significantly reduced in both FBP-MAR and STF-dKCR, yet only the STF-dKCR approach significantly improved the visibility of the small Teflon target (within 2 mm of the metal wire). The attenuation of the Teflon bead improved to 0.0481 mm⁻¹ with STF-dKCR from 0.0166 mm⁻¹ with FBP and from 0.0301 mm⁻¹ with FBP-MAR – much closer to the expected 0.0414 mm⁻¹.

Conclusion:

The proposed method has the potential to improve plaque visualization in coronary CT angiography in the presence of wire-shaped metal components.

1. INTRODUCTION

Patients with metallic cardiac implants (e.g. pacemakers and cardioverter defibrillators) may benefit from computed tomography (CT) due to its superior resolution and the lack of contraindication to metal. Coronary CT angiography can be useful for not only detecting the occurrence of coronary artery diseases but also potential complications associated with the implants such as lead perforation. Yet in up to 60% of the cases¹, metal artifacts arise from the endocardial leads of an implant severely impair the interpretation of images and prevent reliable coronary artery assessment. Streaks result from beam hardening and “blooming” artifacts around metal components can largely obscure the surrounding anatomy, making it difficult to visualize features such as calcifications and atherosclerotic plaques.

While there has been a great deal of work on metal artifact correction methods, many of these methods utilize interpolation^2,3,4, in-painting^5,6, and similar approaches to replace low-fidelity data associated with the metal implants^7–9. Such manipulation of the data can improve overall image quality; however, such processing can also obscure features that are in close proximity to the metal. Recent work on known-component reconstruction (KCR)^10,11,12 has suggested that integrating models of metal components into the reconstruction permits excellent visualization right up to the boundary of the component. Application of such an approach to coronary CT angiography requires three main elements: 1) A model of the metal component; 2) a strategy for estimating deformations and positioning of the component within the patient; and 3) a characterization of the energy-dependent attenuation due to the component (due to material composition of the metal, beam quality, and detector energy response). In this work we present a reconstruction framework addresses these three issues including: 1) a parametric component model that accommodates deformation of wire-like objects within a constrained volume; 2) a 3D-2D optimization strategy that jointly estimates model and registration parameters; and 3) a polyenergetic KCR reconstruction technique¹³ that jointly estimates energy-dependent parameters of the metal implant as well as the attenuation coefficients for the background anatomy.

2. METHODS

2.1. Known Component Reconstruction Review

The proposed reconstruction scheme extends previously reported KCR techniques to include a class of wire-like deformable components. In this section we provide a brief summary of the polyenergetic KCR (STF-KCR) strategy¹³. The original STF-KCR strategy requires only a component shape model but no knowledge of beam quality or material composition. Instead, the approach presumes a uniform material composition and jointly estimates the component shape and the spectral profile in addition to the reconstruction. The STF-KCR approach is based on a mixed fidelity model that factors attenuation contributions from the component and from the background anatomy. The associated forward model can be defined as:

$$\overline{y}=D\left\{g\right\}\text{exp}\left(-A{\mu }_{*}\right)*f\left(-A{\mu }_I\right)$$ (1)

where g denotes system gain (e.g. fluence) and the system matrix A denotes the linear projection operator. The first factor is based on a monoenergetic forward model and accommodates propagation of x-rays through the anatomical background μ_* using Beer’s law. The second factor models x-ray propagation through the component μ₁. This attenuation model is nonlinear in the exponential and is given by the function, f, which we denote the spectral transfer function (STF). We choose a polynomial form within the exponential such that

$$f(p;\kappa )=\text{exp}\left({\sum }_{\{k=1\}}^K{\kappa }_k{(p)}^k\right)$$ (2)

where f is a function of path length p through the component and is parameterized by the spectral coefficients, κ_k. With this forward model, we may define a reconstruction objective function that jointly estimates background attenuation and STF coefficients:

$$\left\{{\widehat{\mu }}_{*},\widehat{\kappa }\right\}=\text{argmin }\Phi \left({\mu }_{*},\kappa ;y\right)=\text{argmin }L\left({\mu }_{*},\kappa ;y\right)-\beta R\left({\mu }_{*}\right)$$ (3)

$$L\left({\mu }_{*},\kappa ;y\right)={‖A{\mu }_{*}-{\sum }_{\{k=1\}}^K{\kappa }_k{(p)}^k-\text{log}\left(D\left\{g\right\}*{\overline{y}}^{-1}\right)‖}_W^2$$ (4)

where the negative log-likelihood is a weighted (by the inverse of the variance of transformed measurements) 2-norm including forward projection of the background anatomy, the spectrally modified projections through the component, and the normalized and log-processed projection data. Additionally, a general regularization term R and a regularization parameter β are utilized to suppress noise.

2.2. Model Parameterization and Preregistration of Known Component

In order to apply STF-KCR to the coronary CT angiography target, we require a model for the wire-like implants including pacing leads, guidewires, etc. That is, we need a model μ₁. Towards this end we adopt the implant model illustrated in Figure 1. The centerline of a deformed cylindrical volume is defined by a catmull-rom cubic spline with controls points evenly spaced along the central axis. An additional parameter specifies the radius of the component. This deformable model is general and parameterizes the shape and position of a wide class of wire-like components. We refer to the combination of STF-KCR with this deformable model as STF-dKCR. To estimate the deformation and shape parameters we apply a staged approach where we solve the generalized registration problem followed by STF-KCR reconstruction.

Figure 1:

Illustration of implant parameterization that includes deformation and radius variables.

The proposed registration method is a modified version of a 3D-2D registration method reported by Otake et al^14,15. The registration method is a joint estimation of the component centerline shape, location, and the radius. These deformable parameters are updated at each iteration to maximize the image similarity between the acquired 2D projections and the digitally reconstructed radiographs (DRR) of the component model. We chose gradient correlation (GC) reported by Penney et al.¹⁶ as the similarity metric since it leverages the strong boundaries of the metal component. The metric is optimized with Covariance Matrix Adaptation Evolution Strategy (CMA-ES), and the objective function is defined as:

$$\widehat{\lambda }=\underset{\lambda }{\text{arg max}}{\sum }_{\theta }\text{GC}\left(y_{\theta },y_{\theta }^{\prime }\left(\lambda ;{\mu }_I\right)\right)$$ (5)

$$\lambda =\left\{Q_c(t),r\right\}=\left\{{\sum }_i{V}_i{W}_i(t),r\right\}$$ (6)

where the model deformation parameters λ are estimated to maximize the total GC computed from all projections. The DRR y′ is a function of the component centerline Q_c(t) and radius r, where Q_c(t) is composed of vertices V_i and the blending function W_i. The same deformation parameters of the component were used for every projection view.

2.3. Experimental Methods

A dedicated test phantom for the performance evaluation of the proposed method is illustrated in Figure 2. The phantom is made of two parts: a commercial cardiac-lung phantom (Kyoto Kagaku, Kyoto, Japan), and a ~25 cm long tube filled with water. A bare copper wire (~ 2500 HU) was threaded through the tube center, and a 1.59 mm Teflon sphere (~900 HU) was attached to the inner tube wall with epoxy resin glue. This phantom is designed to mimic the scenario of a patient with a cardiac implant that has developed an atherosclerotic plaque accompanied by calcification. The copper wire emulates a worst-case scenario of a pacing lead which lies directly adjacent to an object of interest that ordinarily creates artifacts that obscure the object and the surrounding anatomy.

Figure 2.

The heart-lung complex was modified to include a water-filled tube to emulate a coronary vessel. A metal wire was placed in the vessel and a small Teflon bead was included to emulate a calcification.

Images of the test phantom was acquired on an experimental CBCT test bench (Figure 3). The system geometry was set to have a source-to-detector distance (SDD) of ~1100 cm and a source-to-axis distance (SAD) of ~ 850 cm. The detector was a PaxScan 4030CB (Varian Medical Systems, Palo Alto CA) with 1536×1536 pixels at 0.278 mm pixel spacing. The x-ray source was operated at 80 kVp (+1 mm Al, +0.5 mm Cu) and 1.6 mAs/frame. 360 projections were acquired over 360° and a 650×700×305 reconstruction volume with isotropic voxel size of 0.5 mm was obtained with filtered backprojection (FBP) with a Hamming filter (α = 0.5) and a cutoff frequency at 0.8. The proposed reconstruction method was also applied along with a scatter correction step in which the scatter was empirically estimated as a constant in each projection view but a sinusoidal function of rotation angle. The preregistration step involves 80 iterations with a population size of 80, and convergence criteria is achieved when change in the best solution is less than 10⁻⁴. A quadratic penalty was employed and the regularization strength β = 0.05. The proposed KCR used 40 separable quadratic surrogates iterations with 20 ordered-subsets.

Figure 3.

Physical measurement data of the test phantom were acquired on an experimental CBCT bench.

3. RESULTS AND BREAKTHROUGH WORK

Figure 4 illustrates the performance of the STF-dKCR method. The wire-like implant model is shown as a red overlay in the STF-dKCR reconstruction (Figure 4C) to demonstrate outcome of the preregistration. Streak artifacts evident at the boundaries of the copper wire in the FBP reconstruction (Figure 4A) are significantly reduced in both the FBP-MAR reconstruction (Figure 4B) and the STF-dKCR volume. These two methods also suppress shading artifacts and subsequently enhanced the attenuation homogeneity in the central region of the tube, which is filled with water and is expected to have an attenuation value close to that of the cardiac phantom. Only the STF-dKCR method allows good visualization of the Teflon sphere.

Figure 4.

Performance assessment of the STF-dKCR method. Axial, sagittal, and coronal views are shown for (A) the FBP reconstruction, (B) the FBP-MAR reconstruction, and (C) the STF-dKCR volume. The registered, wire-like component is shown as a red overlay in the STF-dKCR reconstruction. The Teflon sphere is indicated in each view by a yellow arrow.

A quantitative measurement was performed to further validate the improvement provided by the STF-dKCR method (Figure 5A). The average attenuation coefficient of the Teflon sphere measured in the FBP reconstruction, in the FBP-MAR reconstruction, and in the STF-dKCR reconstruction is 0.0166 mm⁻¹, 0.0301 mm⁻¹, and 0.0481 mm⁻¹, respectively. The STF-dKCR value is approximately double the FBP value and closely approximates the expected attenuation of Teflon (0.0414 mm⁻¹), assuming an effective energy of 60 keV and Teflon density of 2.2 g/cm³. The estimated STF profile (Figure 5B) demonstrates the classic beam-hardening effect where high-energy x-ray photons have relatively greater survival probability for longer path lengths.

Figure 5.

(A) quantitative measurement of the Teflon volume attenuation for FBP, FBP-MAR, and STF-dKCR and (B) the estimated spectral coefficient values.

4. CONCLUSIONS

We proposed and evaluated a polyenergetic Known-Component Reconstruction method that incorporates a deformable model of a wire-like implant. The algorithm uses a staged estimation process in which the deformable parameters of the implant are determined using an iterative gradient-based registration method prior to the reconstruction. The STF-KCR method estimates the spectral parameters of the component and a nearly artifact-free image. Not only is the visualization improved, but quantitative measurements on the anatomical background and Teflon target are more accurate. Ongoing and future work includes combining the proposed method with an image-based motion estimation process,^17,18 collecting and processing data of phantoms with various shapes and sizes of metal implants, and placing both low- and high-attenuating targets close to the implants. The proposed method has the potential to enhance visualization and diagnosis of atherosclerotic plaque and calcifications in patients with cardiac implants.