Calibration-free beam hardening correction for myocardial perfusion imaging using CT

Abstract

Purpose: Computed tomography myocardial perfusion imaging (CT-MPI) and coronary CTA have the potential to make CT an ideal noninvasive imaging gatekeeper exam for invasive coronary angiography. However, beam hardening (BH) artifacts prevent accurate blood flow calculation in CT-MPI. BH correction methods require either energy-sensitive CT, not widely available, or typically, a calibration-based method in conventional CT. We propose a calibration-free, automatic BH correction (ABHC) method suitable for CT-MPI and evaluate its ability to reduce BH artifacts in single "static-perfusion" images and to create accurate myocardial blood flow (MBF) in dynamic CT-MPI.

Methods: In the algorithm, we used input CT DICOM images and iteratively optimized parameters in a polynomial BH correction until a BH-sensitive cost function was minimized on output images. An input image was segmented into a soft tissue image and a highly attenuating material (HAM) image containing bones and regions of high iodine concentrations, using mean HU and temporal enhancement properties. We forward projected HAM, corrected projection values according to a polynomial correction, and reconstructed a correction image to obtain the current iteration's BH corrected image. The cost function was sensitive to BH streak artifacts and cupping. We evaluated the algorithm on simulated CT and physical phantom images, and on preclinical porcine with optional coronary obstruction and clinical CT-MPI data. Assessments included measures of BH artifact in single images as well as MBF estimates. We obtained CT images on a prototype spectral detector CT (SDCT, Philips Healthcare) scanner that provided both conventional and virtual keV images, allowing us to quantitatively compare corrected CT images to virtual keV images. To stress test the method, we evaluated results on images from a different scanner (iCT, Philips Healthcare) and different kVp values.

Results: In a CT-simulated digital phantom consisting of water with iodine cylinder insets, BH streak artifacts between simulated iodine inserts were reduced from 13 ± 2 to 0 ± 1 HU. In a similar physical phantom having higher iodine concentrations, BH streak artifacts were reduced from 48 ± 6 to 1 ± 5 HU and cupping was reduced by 86%, from 248 to 23 HU. In preclinical CT-MPI images without coronary obstruction, BH artifact was reduced from 24 ± 6 HU to less than 5 ± 4 HU at peak enhancement. Standard deviation across different regions of interest (ROI) along the myocardium was reduced from 13.26 to 6.86 HU for ABHC, comparing favorably to measurements in the corresponding virtual keV image. Corrections greatly reduced variations in preclinical MBF maps as obtained in normal animals without obstruction (FFR = 1). Coefficients of variations were 22% (conventional CT), 9% (ABHC), and 5% (virtual keV). Moreover, variations in flow tended to be localized after ABHC, giving result which would not be confused with a flow deficit in a coronary vessel territory.

Conclusion: The automated algorithm can be used to reduce BH artifact in conventional CT and improve CT-MPI accuracy particularly by removing regions of reduced estimated flow which might be misinterpreted as flow deficits.

Keywords: CT; MPI-CT; beam hardening correction; cardiovascular imaging; myocardial perfusion.

Figure 1

Beam hardening artifact and its influence on blood flow estimation in a normal pig heart without coronary obstruction. (a) BH artifacts in conventional CT images appear as dark streaks between high‐attenuating structures. (b) Blood flow map calculated from the conventional 120kVp images. A false blood flow deficit can be seen (arrow) at a much lower flow than at about 4 o'clock. (c) Blood flow map calculated from the spectral detector CT (SDCT) 70 keV images. Blood flow is much more homogeneous as SDCT significantly reduces BH artifacts. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 2

Beam hardening causes underestimation of the linear attenuation as given by $ln\left({\mathrm{I}}_0/\mathrm{I}\right)$. Data were simulated using monoenergetic source (ideal), and a polyenergetic source (realistic) passing through Cortical bone. Cortical bone μ(E) values were taken from NIST.42 [Color figure can be viewed at wileyonlinelibrary.com]

Figure 3

Flow chart of the ABHC algorithm. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 4

Simulated porcine phantom. (a) A porcine scanned on a prototype Philips SDCT scanner using a cardiac perfusion protocol. (b) Simulated cardiac porcine phantom. W = 60/L = 360.

Figure 5

Cost function value as a function of correction parameters “a” and “b,” with and without filtration. The data cursor on both panels shows the optimal correction parameters. Left: without filtration there are several local minima and the global minimum is not the same as that with noise reduction. Right: after filtration, there is a clear global minimum. Visual examination of resulting images supported the global minimum after noise reduction. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 6

Reduction of BH artifacts using ABHC in a simulated phantom. (a) Original phantom simulated without noise in order to isolate BH artifacts. (b) Conventional FBP reconstruction. (c) ABHC corrected image. W = 80/L = 0. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 7

Reduction of BH artifacts using ABHC in physical phantom. (a) Acrylic phantom with four inserts filled with 15, 22, 26.5, and 77 mgI/ml (starting at bottom right and moving clock wise) reconstructed with FBP. (b) BH‐corrected image using ABHC. (c) Cupping artifact comparison. BH streak artifacts were reduced from 48 ± 6 HU to 1 ± 5 HU and cupping was reduced by 86%. W = 100/L = 100. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 8

Comparison of mean HU values between (a) Conventional, (b) ABHC and 70 keV images in a baseline (FFR = 1.0) porcine model. Standard deviations of mean HU across ROIs are 13.26, 6.86, and 4.54 respectively. ROIs are shown in panel (a). At baseline, enhancement in the myocardium should be uniform. Reduction in HU in ROI3 is due to BH artifact and reduced by ABHC. W = 180/L = 50. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 9

Comparison of absolute blood flow calculated from (a) conventional, (b) ABHC, and (c) 70 keV simulated perfusion scans. Simulated data have constant flow of 100 ml/min/100 g. Conventional had a less homogeneous blood flow compared to ABHC and 70 keV with coefficient of variations of 22%, 9%, and 5% respectively. In conventional images, BH artifacts causes false hypo‐perfusion (red arrow) and false hyper‐perfusion (blue arrow) which is corrected by ABHC. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 10

Effect of iodine‐filled LV on TACs. There is an observed depression (red arrow) in the TAC of a BH affected ROI near 11 o'clock [Fig. 9(a)] when the LV is filled with iodine. This will manifest as low blood flow in this ROI. ABHC reduces BH artifact in all affected time points, correcting this depression. For comparison, in a remote ROI at 1 o'clock, no depression is observed. The overall difference between affected ROI to remote ROI (~3 HU) is from BH artifact originated from the bones, regardless of iodine content. Note that the AIF taken from the ventricle was scaled down by 4 for presentation purposes. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 11

Comparison of absolute blood flow in porcine (FFR = 1.0) calculated from conventional (a), ABHC (b) and 70 keV (c) perfusion scans. Blood flow was calculated using the Johnson‐Wilson model. Conventional shows less homogeneous blood flow compared to ABHC and 70 keV, with flow ratios of 0.59, 0.85, and 0.93 respectively. Flow ratios were calculated as the ratio between the mean flow in ROI1 to the mean flow in ROI2. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 12

Comparison of absolute blood flow in porcine model at FFR = 0.7, calculated from conventional (a), ABHC (b) and 70 keV (c) images. BH artifact was effectively reduced without obscuring the flow deficit in the ischemic region. Blood flow calculated from ABHC images is closer to 70 keV than conventional, as the sum of the squared difference was reduced by 46%. Flow ratios for R0I‐6/ROI‐1 are 1.7, 1.48, and 1.31, for conventional, ABHC, and 70 keV image data, respectively. Data obtained from pig 17.

Figure 13

Blood flow estimation in clinical MPI‐CT, calculated from conventional (a) and ABHC (b) images. Blood flow calculated from ABHC images is more homogeneous. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 14

Example of the cost function sensitivity and BH artifact to the parameters “a” and “b.” The colored contours show the values calculated using the cost function. The black horizontal lines show the measured BH artifact. The red cross shows the minimum of the cost function as found by ABHC. Here, a change of ∼4 × 10⁻⁶ in “b” parameter causes a change of ~5 HU in BH artifact, while the “a” parameter does not affect BH artifact. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 15

Comparison of simulated blood flow maps at different kVps. Simulated data have a constant flow of 100 ml/min/100 g. (a) and (c) were calculated from the conventional reconstruction of MPI‐CT images generated at 80 and 140 kVp respectively. (b) and (d) are the blood flow maps generated from ABHC corrected images of (a) and (c). ABHC found the appropriate correction parameters for each kVp. The flow ratio of the low flow region to the high flow region was increased from 0.61 to 0.85 at 80 kVp and from 0.43 to 0.89 at 140 kVp. In both cases, ABHC would change interpretation from a false positive for myocardial ischemia to a true negative. In both cases ABHC could have changed a clinical decision. Parameters (a, b) were (0.12, −1.79 × 10⁻⁶) and (0.14, −8.23 × 10⁻⁶), at 80 and 140 kVp, respectively. [Color figure can be viewed at wileyonlinelibrary.com]