A Blooming correction technique for improved vasa vasorum detection using an ultra-high-resolution photon-counting detector CT

Abstract

Proliferation of vasa vasorum, the microvasculature within artery walls, is an early marker of atherosclerosis. Detection of subtle changes in the spatial density of vasa vasorum using contrast-enhanced CT is challenging due to the limited spatial resolution and blooming effects. We report a forward model-based blooming correction technique to improve vasa vasorum detection in a porcine model imaged using an ultra-high resolution photon-counting detector CT. Six weeks preceding the CT study the animal received autologous blood injections in its left carotid artery to stimulate vasa vasorum proliferation within the arterial wall (right carotid served as control). The forward model predicted radial extent and magnitude of the luminal blooming affecting the wall signal by using prior data acquired with a vessel phantom of known dimensions. The predicted contamination from blooming was then subtracted from the original wall signal measurement to recover the obscured vasa vasorum signal. Attenuation measurements made on a testing vessel phantom before and after blooming corrections revealed a reduction in mean squared error by ~99.9% when compared to the ground truth. Applying corrections to contrast-enhanced carotid arteries from in vivo scan data demonstrated consistent reductions of blooming contamination within the vessel walls. An unpaired student t-test applied to measurements from the uncorrected porcine scan data revealed no significant difference between the vessel walls (p=0.26). However, after employing blooming correction, the mean enhancement was significantly greater in the injured vessel wall (p=0.0006).

1. INTRODUCTION

Angiogenesis and neovascularization of the micro-vessels (vasa vasorum) within arterial walls precedes atherosclerosis¹. Measuring changes in vasa vasorum density within the arterial wall could enable early diagnosis of atherosclerosis prior to fatal consequences such as stenosis, thrombosis and stroke. Conventional x-ray computed tomography (CT) angiography, which focuses on imaging vasculature using intravenous contrast agent, has been investigated for use in such applications¹. However, blooming artifacts resulting from the hyper-attenuating contrast media flowing through the vessel lumen hinders the detection of subtle signals originating from the vasa vasorum within the arterial wall. The finite sized CT detector pixels used during scan acquisition results in partial volume averaging at the interface between the arterial lumen and wall^2,3. This partial volume averaging is what inevitably causes blooming, which contributes to a significant artificial enhancement in voxels neighboring the vessel lumens, and can be observed as a faint halo of enhancement extending outward from the lumens². Compared to the luminal contrast enhancement, signals originating from enhanced vasa vasorum are relatively weak, having been discovered to be only about 10–15% of the total luminal enhancement⁴. In addition to this weak signal, vasa vasorum are also very small (diameters ranging from 5–150 um⁴) and reside within the outer portion of a very thin vessel wall (thickness around 30% of lumen radius⁴). Considering their small size, proximity to the lumen, and weak signal strength, it is evident that luminal blooming effects can completely obscure the measurement of vasa vasorum in CT images². In previous work presented by Rajendran et al.² an image deconvolution technique was used to improve the CT number (attenuation value in Hounsfield Units) accuracy in the arterial walls of the porcine model scanned with a whole-body photon counting detector (PCD) CT system. This deconvolution technique was developed for applications involving CT images scanned using the Macro mode of detector configuration (effective pixel pitch: 0.5 mm × 0.5 mm)⁵. However, this deconvolution technique was not discovered to yield a similarly significant benefit when applied to images resulting from scans using the ultra-high resolution (UHR) mode of detector configuration (effective pixel pitch: 0.25 mm × 0.25 mm)⁵.

In this study we proposed to correct for the artificial enhancement in arterial walls, due to blooming from the enhanced arterial lumens, using a forward model to reduce the CT number overestimation in voxels near the lumen-wall boundary. Our model quantifies vasa vasorum density within a vessel wall by averaging the CT number among voxels residing within the arterial wall. This experimental approach was first evaluated using an in-house vessel phantom which included both an enhanced lumen and arterial wall. Finally, we demonstrated the model’s signal recovery performance in an in vivo vascular perfusion scan of a swine model of enhanced vasa vasorum density⁴.

2. METHODS

2.1. CT acquisition and image reconstruction

The imaging system selected was a research whole-body PCD-CT scanner (Somatom CounT, Siemens Healthcare, Forchheim, Germany) with a cadmium telluride semiconductor^5,6. The UHR mode was chosen for all scans performed⁵. The iodinated intravascular contrast agent used in this study was Omnipaque^® 350 (GE Healthcare, Inc.). A vascular CT imaging protocol was used for all scans reported in this study. The scans were performed using the following parameters: 140 kV, 341 mAs, and energy threshold settings of 30 and 70 keV. Images were reconstructed using an 80 mm × 80 mm FOV, 1024×1024 matrix size, 1 mm slice thickness (yielding voxel size of 0.0781mm × 0.0781mm × 1 mm), and an iterative reconstruction algorithm with a sharp quantitative kernel (Q65f, SAFIRE).

2.2. Development of blooming correction technique

The blooming correction technique is an algorithm based on a forward model and which was written in MATLAB (MATLAB ver. 2018b, The MathWorks Inc.). First, the model predicted the magnitude of blooming-induced artificial enhancement occurring within a given vessel’s wall. Second, the predicted value was subtracted from average wall enhancement calculated in the original image, leaving a residual enhancement value arising from the locally enhanced vasa vasorum. Accurate prediction of blooming artifacts within each individual voxel is challenging, however, a unique aspect of our study is that we are interested in the collective increase of arterial vasa vasorum density rather than the number of individual vessels. Thus, all measurements were based on the average CT numbers in annular regions of interest (ROI) generated within the arterial walls of the vessels.

The blooming predictions of the forward model were based upon the three most relevant parameters: (i) lumen diameter, (ii) lumen enhancement (CT number of the enhanced lumen), and (iii) radial distance of the ROI from the lumen/wall boundary. A lookup table (LUT) which could be used to make predictions based on these input parameters was generated using measurements of blooming magnitude obtained from the phantom data described herein. Three cylindrical inserts (2, 4, and 8 mm diameter) embedded within a 10 cm solid water phantom were used to emulate contrast infused vessels of various sizes. The solid water phantom was additionally wrapped in soft-tissue equivalent gel sheets to increase its overall diameter to 30 cm. The phantom was then scanned twice, with the three inserts filled with iodine contrast of 21 mg/ml (Figure 1) and 35 mg/ml. The selected 21 and 35 mg/ml iodine concentrations corresponded to luminal enhancements of approximately 500 and 1000 HU respectively. Within the resulting phantom images, a series of 1 voxel thick concentric annular ROIs were defined beyond each luminal boundary and the mean CT number was calculated within each ROI.

Figure 1.

(Left) A PCD-CT image of the solid water phantom containing 21 mg/ml iodine contrast within its three inserts. An 11 mm line profile was generated across each insert in order to show the impact of blooming relative to the diameter of each insert. (Right) The CT numbers (Hounsfield Units) of the underlying voxels were plotted along the length of each respective line. The line profile for the 2 mm insert (red), 4 mm insert (magenta) and 8 mm insert (blue) were each compared to a ground truth plot (black) matching their respective physical geometries. In the absence of blooming we would anticipate the measured and ground truth lines to be superimposed.

The forward model for blooming correction was applied to a given vessel in the following sequence: (i) the enhanced vessel lumen’s mean CT number was calculated within a central circular ROI. (ii) Then the mean CT number of the unenhanced background tissue was calculated. (iii) The lumen was then segmented based on a half-maximal, threshold-based method, using the CT numbers obtained in steps 1 and 2 as described by:

$$T=\frac{T_{\mathit{\text{Lumen}}}+T_{\mathit{\text{Background}}}}{2},$$ (1)

where T is the half-maximal threshold, T_Lumen is the mean CT number measured within the lumen, and T_Background is the mean CT number measured from within the tissue surrounding the vessel. (iv) Successive 1 voxel thick dilations from the lumen boundary were used to define concentric annular ROIs within the vessel wall. (v) The radial distance between each annular ROI and the lumen was calculated. (vi) The blooming magnitude within a given ROI was determined through the pre-built LUT with input values of lumen size, lumen enhancement and radial distance determined in previous steps; with appropriate interpolation used as needed. (vii) The CT number corresponding to the luminal blooming, occurring within the arterial wall ROI, was subtracted away from the ROI’s original mean CT number, as described by:

$$C{T}_{\mathit{\text{Target}}}=C{T}_{\mathit{\text{Original}}}-C{T}_{\mathit{\text{Blooming}}},$$ (2)

where CT_Original is the mean CT number originally calculated from the annular ROI, CT_Blooming is the estimated attenuation due to blooming from the nearby lumen, and CT_Target is the mean CT number of the wall ROI in the absence of attenuation due to blooming.

2.3. Vessel phantom validation

An in-house vessel phantom with known physical dimensions was used to provide a preliminary evaluation of the blooming correction technique’s accuracy. The phantom was composed of a 6 mm diameter plastic straw (the vessel’s lumen) and a 0.5 mm thick layer of attenuating tape wrapped around the outside of the straw (emulating enhanced vasa vasorum). This vessel phantom was scanned twice, the first time it was filled with deionized water (offering only the pure vasa vasorum signal), the second time filled with 22 mg/ml of iodine contrast (providing a blooming corrupted signal). Prior to both scans, the vessel phantom was submerged in a 30 cm water tank. The vessel wall in the resulting CT images from both scans were segmented and the annular ROI mean CT numbers were calculated based on the approach described in Sec. 2.2 above. Additionally, blooming correction was applied to the iodine infused image set, which were then identically reanalyzed. Measurements from the iodine infused contrast scan, with and without the blooming correction method, were compared alongside the reference standard from the water scan.

2.4. Animal model scan application

The blooming correction technique’s performance was then evaluated using data collected from the PCD-CT scans of the porcine model for enhanced vasa vasorum density. This animal study was similar to those presented in previous work by our group^2,4 and was approved by the Mayo Clinic Institutional Animal Use and Care Committee. Enhanced vasa vasorum density was induced by injecting ~0.1 mL of autologous blood into the arterial wall of the left common carotid artery of the anesthetized (Telazol/Ketamine/Xylazine) animal. After a six week recovery period the animal was anaesthetized again prior to scanning the 6–7 cm long neck region which encompassed the injections. Scans were acquired with the PCD-CT system using the settings noted in Sec. 2.1. During these scans the animals received an injection of intravascular iodinated contrast agent (40 ml injected at 10 ml/s), which was followed by an initial 2 s delay, prior to the acquisition of twenty identical “time-point scans” (one every 3 s)⁴. Both of the carotid arteries were imaged during the successive twenty scans such that during analysis the right carotid artery could serve as a control to the injured left artery.

The following preprocessing was applied to the CT images before conducting data analysis on the carotid arteries: (i) to reduce the influence of image noise, an image based multi-energy non-local means (MENLM) noise reduction technique was applied to the reconstructed images⁷. (ii) For cases where the enhanced jugular veins were located too near to the carotid arteries, they were manually masked out of the images. (iii) Prior to calculating the mean CT number of the annular wall ROIs, extravascular lipid tissue located beyond the vessel’s adventitia layer, was removed from the ROIs via CT number thresholding.

After image preprocessing, blooming correction was performed on the right carotid artery. The luminal segmentation of the vessel and the generation of seven concentric annular ROIs follow the same strategy noted in Sec. 2.2. The mean attenuation values acquired from the blooming corrected ROIs were then compared with the values obtained from the original images. A supplemental evaluation was performed thereafter to illuminate any differences in vessel wall enhancement which might exist between the injured and control vessels. Thus, the blooming correction approach taken for the right carotid artery was then repeated for the left carotid artery. The unpaired student t-test was applied to determine if there was a significant difference in the mean CT numbers from the wall ROIs of the two vessels being compared (p<0.05 was considered as statistically significant).

3. Results

3.1. Blooming correction in phantom scan

The initial evaluation of the blooming correction technique was performed on CT images obtained from the vessel phantom scans described in Sec. 2.3. A total of seven concentric annular ROIs were generated around the contrast infused lumen in images from the original iodine scan (Figure 2(A)). These same seven ROI were superimposed onto the images of the water scan (Figure 2(B)) such that they could be used for determination of the vessel’s “ground truth” wall enhancement (free of luminal contrast blooming). A mean enhancement value was calculated from the voxels contained within each annular ROI. The resulting CT numbers were used to compare the vessel wall enhancement observed in the following images: (i) the iodine infused lumen scan, (ii) the water infused lumen scan and (iii) the iodine infused lumen scan after blooming correction. The unique blooming contamination occurring within each ROI was subtracted from its mean CT number as described in Sec 2.2. Wall enhancement measurements were graphically displayed to facilitate a visual comparison of this data (Figure 2(C)). It can be noted that the wall enhancement curve representing the iodine infused lumen scan becomes very similar to the ground truth curve after blooming correction was applied. Wall enhancement signal accuracy from the original and corrected iodine scan data were compared, with respect to the ground truth water scan data, using the mean squared error (MSE). The wall signal from the original iodine scan produced an MSE of 9801.5. The wall signal from the blooming corrected iodine scan produced a MSE of 10.7. The resulting ~99.9% reduction in MSE post correction further supports the improvement in signal accuracy noted in Figure 2(C).

Figure 2.

(A) A CT image of the vessel phantom containing 22 mg/ml Iodine solution. The seven concentric annular ROIs were superimposed onto the image in blue. (B) A CT image of the vessel phantom containing deionized water. The same seven ROI are superimposed onto the image in blue. The enhanced layer of tape appears to reside between the 4^th and 7^th ROI. (C) Effect of Iodine blooming correction on the CT numbers measured in the ROI of the vessel phantom images. The corrected scan is nearly an exact match with the water scan.

3.2. Blooming correction in animal model scan application

The blooming correction technique was used to process the animal model data described in Sec. 2.4. After the CT images were reconstructed and preprocessed (Figure 3(A)), blooming corrections were performed on the segmented right carotid artery (Figure 3(B)). The radius of the segmented vessel lumen was discovered to be ~1.37 mm. The magnitude of luminal contrast blooming varied among the seven surrounding annular ROIs (Figure 3(C)). Three mean CT numbers were calculated with each ROI: (i) the mean CT value from the unenhanced baseline scan, (ii) the mean CT value from the contrast enhanced scan and (iii) the mean CT value from the blooming corrected contrast enhanced scan. This data is presented graphically in Figure 3(D). The large degree of luminal blooming contamination observed in the original enhancement curve is absent from the corrected enhancement curve. However, the corrected enhancement curve is not an exact match with the baseline curve between annular ROIs 1 and 5, which was different than the final result in Sec. 3.1. The subtle enhancement occurring between the first and fifth ROI of the blooming corrected curve may be due to the presence of enhanced vasa vasorum in the vessel wall. Assuming the vessel wall was 30% of the vessel lumen radius⁴, or about 0.41 mm thick, we would indeed expect any signal from enhanced vasa vasorum to appear within a wall region spanning these first five annular ROI (0.41 mm / 0.0781 mm/ROI = ~5 ROI).

Figure 3.

(A) CT image of the neck region from a porcine iodine perfusion angiogram. (B) Carotid artery at peak iodine enhancement and (C) the same image with the seven superimposed concentric annular ROIs. (D) Impact of blooming reduction technique applied to the right carotid artery. The original ROI attenuation values (red) are greatly over-enhanced (due to luminal blooming) relative to the blooming reduced values (blue). The small enhancement in the blooming corrected curve (between ROI 1 and 5) is due to enhanced vasa vasorum within the arterial wall.

3.3. Impact of blooming correction on vasa vasorum density measurement

The same animal model data described in Secs. 2.4 and 3.2 was again evaluated in this section. Signals originating within the enhanced vessel walls of both the left (injured) and right (control) carotid arteries are compared before and after blooming correction. A single mean CT number was used to represent the enhancement occurring within the carotid artery vessel wall for each contiguous CT slice evaluated (n=13). The mean CT number was calculated within a larger wall ROI, which included the first through fifth annular ROIs, and which corresponded to the vasa vasorum signal identified in Sec. 3.2. The wall ROI signal was also averaged over the time point scans which corresponded to the arterial contrast recirculation period (time point scan 10 to 15). The mean wall enhancements were normalized with the mean enhancement of their associated vessel lumen by calculating the wall-to-lumen enhancement ratio. This accounted for any differences in luminal enhancement magnitude which might exist between the injured and control arteries. Prior to blooming correction (Figure 4(A)), the CT slice enhancement ratios from the injured and control artery data appeared similar, and the unpaired t-test suggested that the difference in their mean was not statistically significant (p = 0.26). After blooming correction (Figure 4(B)), the CT slice enhancement ratios from the injured and control artery data appeared dissimilar, and the unpaired t-test revealed that a statistically significant difference existed between their means (p = 0.0006). This result suggests that the signal from the increased enhancement in the injured artery’s vasa vasorum (relative to the control) was only observable after the blooming contamination was removed.

Figure 4.

Left and Right carotid artery wall-to-lumen enhancement ratios (left carotid wall has sustained an injury to promote vasa vasorum proliferation and right carotid is control) measured within 13 consecutive CT slices from a pig scan before blooming correction (A) and after blooming correction (B).

4. Conclusions

We demonstrated the ability of a forward-model-based technique to correct for the artificial enhancement occurring within arterial walls due to blooming of luminal contrast. The phantom results demonstrated the accuracy of the blooming correction technique and its successful recovery of the underlying wall signal. The signal contamination due to contrast-related blooming was effectively reduced from measured wall mean CT numbers and the underlying signal was restored to levels matching those measured from the contrast-free (and thus blooming-free) scan. The animal scan results in Sec. 3.2 showed that the blooming reduction technique was able to recover the signals corresponding to enhanced vasa vasorum which would otherwise be obscured by luminal blooming. A significant difference in vasa vasorum density existed between injured and control arterial walls, however, this was only realizable after blooming correction. The proposed approach enables reliable detection of vasa vasorum, a biomarker of early atherosclerosis.