|Year : 2015 | Volume
| Issue : 1 | Page : 114-118
Improved dynamic CT angiography visualization by flow territory masking
Søren Christensen1, Bruce Campbell2, Maarten G Lansberg3, Jacqui Hislop-Jambrich4, Stephen Davis2, Patricia Desmond1, Mark Parsons5
1 Department of Radiology, University of Melbourne, Melbourne, Victoria, Australia
2 Department of Neurology, University of Melbourne, Melbourne, Victoria, Australia
3 Department of Neurology, Stanford Stroke Center, Stanford University, California, USA
4 Clinical Applications Research Centre, North Ryde, Sydney, New South Wales, Australia
5 Department of Neurology, Hunter Medical Research Institute, John Hunter Hospital, University of Newcastle, Newcastle, New South Wales, Australia
|Date of Submission||27-Mar-2015|
|Date of Acceptance||15-Jun-2015|
|Date of Web Publication||30-Sep-2015|
Stanford Stroke Center, 780 Welch Road, Suite 350, Stanford, California - 94305
Source of Support: None, Conflict of Interest: None
Backgound and Purpose: Computerized tomography (CT) perfusion (or CTP) source images from CT scanners with small detector widths can be used to create a dynamic CT angiogram (CTA) similar to digital subtraction angiography (DSA). Because CTP studies use a single intravenous injection, all arterial territories enhance simultaneously, and individual arterial territories [i.e., anterior cerebral artery (ACA), middle cerebral artery (MCA), and posterior cerebral artery (PCA)] cannot be delineated. This limits the ability to assess collateral flow patterns on dynamic CTAs. The aim of this study was to devise and test a postprocessing method to selectively color-label the major arterial territories on dynamic CTA.
Materials and Methods: We identified 22 acute-stroke patients who underwent CTP on a 320-slice CT scanner within 6 h from symptom onset. For each case, two investigators independently generated an arterial territory map from CTP bolus arrival maps using a semiautomated method. The volumes of the arterial territories were calculated for each map and the average relative difference between these volumes was calculated for each case as a measure of interrater agreement. Arterial territory maps were superimposed on the dynamic CTA to create a vessel-selective dynamic CTA with color-coding of the main arterial territories. Two experts rated the arterial territory maps and the color-coded CTAs for consistency with expected arterial territories on a 3-point scale (excellent, moderate, poor).
Results: Arterial territory maps were generated for all 22 patients. The median difference in arterial territory volumes between investigators was 2.2% [interquartile range (IQR) 0.6-8.5%]. Based on expert review, the arterial territory maps and the vessel-selective dynamic CTAs showed excellent consistency with the expected arterial territories in 18 of 22 patients, moderate consistency in 2 patients, and poor consistency in another 2 patients.
Conclusion: Using a novel postprocessing technique, arterial territory maps and dynamic CTAs with vessel-selective color-coding can be derived from a standard CTP scan. These maps may be used to noninvasively assess collateral flow in patients with acute stroke.
Keywords: Angiography, cerebrovascular disease/stroke, collateral flow, computerized tomography (CT)
|How to cite this article:|
Christensen S, Campbell B, Lansberg MG, Hislop-Jambrich J, Davis S, Desmond P, Parsons M. Improved dynamic CT angiography visualization by flow territory masking. Brain Circ 2015;1:114-8
|How to cite this URL:|
Christensen S, Campbell B, Lansberg MG, Hislop-Jambrich J, Davis S, Desmond P, Parsons M. Improved dynamic CT angiography visualization by flow territory masking. Brain Circ [serial online] 2015 [cited 2020 Sep 26];1:114-8. Available from: http://www.braincirculation.org/text.asp?2015/1/1/114/164991
| Introduction|| |
The progression of infarction in an acute stroke is highly dependent on collateral flow. ,,, Digital subtraction angiography (DSA) is the gold standard for collateral grading, but this technique is not practical for the majority of stroke patients as it involves an intraarterial contrast injection. The advent of modern multidetector computerized tomography (CT) systems with large spatial coverage and small detector widths has made it possible to generate dynamic CT angiography (CTA) visualizations that can be used to assess collateral flow.
Dynamic CTA has several potential advantages over DSA, which is a projection-based technique that involves catheter-based contrast injection directly into the cerebral arteries. Dynamic CTA is less invasive, is quicker, yields a volumetric acquisition [three-dimensional (3D) time resolved, also referred to as four-dimensional (4D)] that allows visualization of arterial filling from any angle, whereas DSA visualizations are limited to the angle of acquisition. Finally, dynamic CTA does not alter arterial flow patterns, whereas the intraarterial injections used during DSA do. 
Dynamic CTA, however, also has limitations. As opposed to vessel-selective enhancement with DSA, the broad intravenous injection of the dynamic CTA leads to enhancement of the entire cerebral arterial tree. In addition, dynamic CTAs are acquired with a relatively slow frame rate (typically one frame every 1-3 s), whereas DSAs are acquired with multiple frames per second. Consequently, it is difficult on dynamic CTA to determine the major arterial supply of distal branches and to assess the direction of flow in the cerebral arteries.  This makes collateral flow assessment more difficult on dynamic CTA compared to DSA. In this study, we aimed to address this limitation by developing and testing a novel postprocessing method to mimic selective vessel injections of the main arterial branches, the anterior cerebral artery (ACA), middle cerebral artery (MCA), and posterior cerebral artery (PCA) on a dynamic CTA.
| Materials and Methods|| |
Patients and imaging protocol
We retrospectively identified 22 acute-stroke patients who underwent bolus CT perfusion (CTP) imaging within 6 h from symptom onset between June 2010 and February 2012. The sample included patients with internal carotid artery (ICA) and MCA lesions as well as cases with no major occlusion in order to assess the method across a range of collateral supply and normal supply patterns. Patients were imaged using the Aquilion One 320-row CT (Toshiba Medical Systems, Nasu, Japan),  which provides 16 cm of z-axis coverage with 0.5-mm slices and in plane resolution of 0.43 mm. The CTP protocol was the standard manufacturer stroke protocol, acquiring 19 frames with a variable sampling rate covering either 52 s or 63 s of the bolus passage. The images were 3D-motion-corrected using Autoreg (MINC tools) and used at the native resolution for rendering of a maximum intensity projection (MIP) of the dynamic CTA but downsampled from the native 512 × 512 × 320 reconstruction to a matrix size of 128 × 128 × 80, yielding a resolution of 1.7 mm, 1.7 mm, 2.0 mm along x, y, and z respectively. The downsampling is performed using volume averaging in order to increase the bolus contrast-to-noise ratio (CNR, defined as Hounsfield unit increase to noise ratio), which is required for the segmentation algorithm to work. The Institutional Ethics Committee approved the study, and all patients gave written, informed consent.
Territory labeling algorithm
We implemented and employed a novel arterial territory segmentation algorithm to label each of the major arterial territories based on user-provided seed regions using MATLAB R2013b (MathWorks, Natick, MA, USA). This segmentation algorithm exploits the fact that bolus arrival time is a continuous function across the imaged volume, and uses this information to track the bolus from proximal seed points (just distal to the circle of Willis) into the periphery of the vasculature.  The method is semiautomated. An operator selects five seeds on the bolus arrival map to mark the major arterial territories: one MCA seed in each hemisphere, one PCA seed in each hemisphere, and a common seed for the ACAs, which run too close together to resolve separately given the smoothing that is necessary for the algorithm to work. The seeding procedure is an interactive operation that takes place on a map of estimated bolus arrival times generated using a previously published method.  The steps involved in generating the color-coded arterial territory maps and dynamic CTAs are shown in [Figure 1], [Figure 2] and [Figure 3]. The dynamic CTAs are rendered as MIPs across the 19 frames and color-coded by arterial territory. The color-coded MIPs can either display all arterial territories simultaneously in different colors [Figure 4] or allow for the "vessel-selective" display of one or more arterial territories selectively to increase conspicuity [Figure 5].
|Figure 1: Flow diagram of the processing steps involved in creating color-coded dynamic CTAs. CTP source data from the scanner (DICOM format) is processed on a workstation to generate bolus delay maps. An operator manually "seeds" the major vascular territories by identifying the proximal ACA, MCA, and PCA on the bolus delay map (A, also see Figure 2). A fully automated algorithm then generates arterial territory maps by growing the seeds to label entire arterial territories (B, also see Figure 3)|
Click here to view
|Figure 2: Manual seeding of arterial territories. The seeding tool allows the operator to visualize the region of tissue with a bolus arrival delay below an adjustable threshold. A-C are bolus delay maps of the same patient at increasing bolus delay thresholds. These images illustrate the propagation of the contrast bolus from proximal cerebral arteries into tissue|
Click here to view
|Figure 3: Graphical presentation of generating the arterial territory map and the color-coded dynamic CTA. a) Arterial territory map. A fully automated region-growing algorithm generates arterial territory maps by growing each of the five arterial seeds (see Figure 2) in 3D space until a color-coded map of entire arterial territories is created b) Conventional dynamic CTA without color coding c) Color-coded dynamic CTA. The arterial territory map (panel A) and the conventional dynamic CTA (panel B) are superimposed to create a color-coded dynamic CTA|
Click here to view
|Figure 4: Patient with a right MCA occlusion and prominent ACA collaterals. Panel A shows the conventional dynamic CTA without color-coding. Panel B shows a color-coded dynamic CTA (ACA = Red, MCA = Blue). The right MCA is occluded, which is evident from the absence of blue branches in the right hemisphere. The ACA supplies collateral fl ow to the right MCA territory via leptomeningeal vessels, which is evidenced by the red branches over the right cerebral convexity|
Click here to view
|Figure 5: Patient with a high-grade left MCA stenosis and prominent PCA collaterals. Panel A: ATM of this patient with a high-grade left MCA stenosis demonstrates that the territory supplied by the stenotic left MCA (small MCA territory in left hemisphere, blue) is markedly reduced compared to the right MCA (typical size of MCA territory in right hemisphere, blue). Panel B: The color-coded dynamic CTA of the same patient shows a reduction in the left MCA branches (blue) that is compensated for by an increase in left PCA branches (green) and, to a lesser extent, left ACA branches (red)|
Click here to view
Qualitative reproducibility assessment of arterial territory segmentation maps
The topography of the arterial territories depends on the user-defined seed points: different seed points can result in changes to the arterial territories. To examine the reproducibility of the technique, the seeding procedure was performed by two independent operators (SC and BC). This yielded two arterial territory maps for each case. Interrater volumetric agreement was quantified as the average relative difference between the volumes of the arterial territories on these two maps.
where i = left ACA, right ACA, left MCA, right MCA, left PCA, and right PCA; Volume_Operator1 is the volume of the arterial territory based on the arterial territory map generated by operator 1; and Volume_Operator2 is the volume of the arterial territory based on the arterial territory map generated by operator 2.
Interrater spatial agreement was assessed by a visual comparison of the arterial territory maps of the two operators.
Qualitative assessment of arterial territory maps
Two experts (P.D., neuroradiologist and B.C., stroke neurologist) performed a review of the arterial territory segmentation maps to determine if they were biologically consistent with hemodynamic findings on perfusion maps and conventional CTA. The experts' review was structured as follows:
- Determine site of arterial occlusion on a conventional dynamic CTA without color coding;
- Determine the region of hypoperfusion on conventional time to peak (TTP) and cerebral blood volume (CBV) maps;
- Determine if the arterial territory segmentation map and the color-coded dynamic CTA are consistent with the arterial territories that are expected based on the findings from 1 and 2 above.
Specifically, a classic arterial topography was expected in hemispheres without an arterial occlusion, and an altered topography was expected in hemispheres with an arterial occlusion (reduction of the territory of the occluded artery and expansion of the territories of neighboring patent arteries). Consistency between the experts' expected arterial topography and the observed topography on the arterial territory map and the CTA with vessel-selective color-coding was rated on a 3-point scale (excellent, moderate, poor).
| Results|| |
The two independent investigators seeded the five major arteries (a single ACA, bilateral MCA, and bilateral PCA seeds), thereby generating arterial territory maps and dynamic CTAs with vessel-selective color-coding in all 22 cases. The median volumetric difference of the arterial territories between investigators was 2.2% [interquartile range (IQR) 0.6-8.5%] and the mean difference was 7.8%. The volumetric difference was less than 5% in 15 of the 22 cases. The maximum volumetric difference was 48%, which occurred in a patient with motion-induced streak artifact. The minimum difference was 0%, indicating complete pixel-by-pixel agreement of the arterial territories between investigators. This occurred in five cases.
Expert review of the arterial territory maps and the vessel-specific dynamic CTAs showed excellent consistency with the expected arterial territories in 18 of the 22 cases (82%), moderate consistency in 2 cases, and poor consistency in 2 cases [Table 1]. The cases with poor consistency resulted from the following issues: in one patient with a high-grade PCA stenosis, the arterial territory map showed complete absence of the PCA territory because the PCA could not be seeded, while expert reviewers identified a PCA territory; In another patient with a proximal MCA occlusion, the arterial territory map demonstrated a small MCA territory in the region of the caudate, while expert reviewers classified this region as most likely supplied by the ACA. Moderate consistency was due to difficulty in seeding the ACA as a result of poor-quality source data in one case and overestimation of the PCA territory on the arterial territory map in the other case.
| Discussion|| |
This study demonstrates that it is feasible to generate arterial territory maps and dynamic CTAs with vessel-specific color-coding from images that are acquired during a standard CTP scan. The color-coded dynamic CTAs emulate conventional DSA and have, therefore, great promise as a noninvasive technique for collateral grading. The postprocessing method used to generate the maps and CTAs is semiautomatic and only requires an operator to seed the proximal arterial segments. The arterial territories maps and color-coded CTAs are robust against variability in seed selection between operators and they are biologically plausible based on expert review.
In most cases, there was excellent agreement between the arterial territory maps generated by two independent operators. This demonstrates that the algorithm is capable of defining arterial territories with acceptable reproducibility. The territory maps should be reviewed alongside the perfusion maps and the CTA, as this can help identify cases where the technique failed.
In a minority of cases, the arterial territory maps differed between operators, which was principally related to the seeding of an incorrect vessel.
Future enhancements to the postprocessing technique may address this problem. In the current implementation, the seeds are placed without use of the dynamic CTA in a region superior to the circle of Willis. Implementation of a method that enables seeding directly on the dynamic CTA would allow a more precise anatomical localization within the proximal cerebral vessels. Future implementation of a fully automated seeding procedure will further reduce variability but we see a standardized manual seeding procedure on the dynamic CTA as the first step to improve the performance of the algorithm.
Another shortcoming of the current algorithm is that it can only be applied to CT slices that are superior to the circle of Willis [Figure 5]. The segmentation algorithm is a threshold-connected type algorithm that is applied to tracer arrival delay maps estimated from a curve fitting procedure described previously.  In its current form, the segmentation algorithm requires that the seeds are not interconnected by vasculature proximal to the seeds. Because the proximal arterial vessels are all connected via the circle of Willis, these connections need to be removed for the algorithm to operate correctly. The simplest way to remove the connections is to exclude the circle of Willis in its entirety and focus on the territory superior to the circle of Willis. Future development of a more elegant removal of the proximal vessel connections could overcome this limitation and would allow the generation of arterial territory maps covering the entire imaging volume.
We assessed the algorithm based on an expert review of the arterial territory maps and the color-coded dynamic CTAs. The algorithm has no anatomical constraints and is, therefore, free to grow the small, initial seeds in any direction. Despite the highly unconstrained nature of the algorithm, it converges to what appears to be plausible arterial territories. Specifically, expert review demonstrated that the topography of the arterial territories in unaffected hemispheres is strikingly similar between patients and that the topography of arterial territories in hemispheres with arterial occlusions is generally consistent with the topography that is expected based on knowledge of the site of the arterial occlusion and the extend of the perfusion deficit on conventional CTP maps. While the results of the expert review were encouraging, validation with a true gold standard of vascular territory mapping would have been preferred. However, such a gold standard does not, to our knowledge, exist. Arterial spin labeling (ASL) has arterial territory labeling capability, but is itself not validated. Moreover, ASL has inherent problems with the characterization of delays in excess of 3-4 s, which are very frequently seen in acute stroke.  DSA also has arterial territory mapping capability, but it cannot generate a 3D rendering of the arterial territories and it involves an invasive procedure with multiple contrast injections, which was not obtained in this cohort.
In this context, we believe that the best way to validate this technique would be using back-to-back territory ASL mapping and CTP in a cohort of patients with moderate hemodynamic delays, such as those seen in patients with chronic stenosis. Such a cohort would provide samples of abnormal arterial distribution territories, which can be characterized by both ASL and CTP.
Our study was performed using a single protocol on a single system. Translation to other full-brain coverage systems is likely to require modification of the imaging protocols to achieve sampling rates and CNRs similar to those produced by our protocols.
Selective arterial territory visualization on dynamic CTAs is an entirely new technique that may have multiple clinical applications. One potential application is the ability to rate collateral flow. Larger clinical studies should compare collateral grading on color-coded dynamic CTAs to collateral grading on DSA and should examine if the color-coded dynamic CTAs provide improved diagnostic and/or prognostic performance over standard CTA visualizations.
| Conclusion|| |
We have presented a novel postprocessing technique that segments the arterial territories and enables color-coding of the arterial territories on the dynamic CTA. This technique holds promise for more easily interpretable visualization of dynamic CTA scans and more accurate collateral flow assessments.
Financial support and sponsorship
Toshiba Medical, Nasu, Japan has assisted with funding for this study.
Conflicts of interest
Søren Christensen is a co-inventor of a patent of an algorithm used in this study, owned by University of Melbourne. Jacqui Hislop is an employee of Toshiba Medical, Nasu, Japan.
| References|| |
Bang OY, Saver JL, Buck BH, Alger JR, Starkman S, Ovbiagele B, et al
. UCLA Collateral Investigators. Impact of collateral flow on tissue fate in acute ischaemic stroke. J Neurol Neurosurg Psychiatry 2008;79:625-9.
Bang OY, Saver JL, Kim SJ, Kim GM, Chung CS, Ovbiagele B, et al
. Collateral flow predicts response to endovascular therapy for acute ischemic stroke. Stroke 2011;42:693-9.
Liebeskind DS. Collateral circulation. Stroke 2003;34:2279-84.
Miteff F, Levi CR, Bateman GA, Spratt N, McElduff P, Parsons MW. The independent predictive utility of computed tomography angiographic collateral status in acute ischaemic stroke. Brain 2009;132:2231-8.
Menon BK, O'Brien B, Bivard A, Spratt NJ, Demchuk AM, Miteff F, et al
. Assessment of leptomeningeal collaterals using dynamic CT angiography in patients with acute ischemic stroke. J Cereb Blood Flow Metab 2013;33:365-71.
Brouwer PA, Bosman T, van Walderveen MA, Krings T, Leroux AA, Willems PW. Dynamic 320-section ct angiography in cranial arteriovenous shunting lesions. AJNR Am J Neuroradiol 2010;31:767-70.
Orrison WW Jr, Snyder KV, Hopkins LN, Roach CJ, Ringdahl EN, Nazir R, et al
. Whole-brain dynamic CT angiography and perfusion imaging. Clin Radiol 2011;66:566-74.
Christensen S, Calamante F, Hjort N, Wu O, Blankholm AD, Desmond P, et al
. Inferring origin of vascular supply from tracer arrival timing patterns using bolus tracking MRI. J Magn Reson Imaging 2008;27:1371-81.
Wang DJ, Alger JR, Qiao JX, Gunther M, Pope WB, Saver JL, et al
. UCLA Stroke Investigators. Multi-delay multi-parametric arterial spin-labeled perfusion MRI in acute ischemic stroke - comparison with dynamic susceptibility contrast enhanced perfusion imaging. Neuroimage Clin 2013;3:1-7.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5]
|This article has been cited by|
||Imaging selection for acute stroke intervention
| ||Bruce CV Campbell,Mark W Parsons |
| ||International Journal of Stroke. 2018; 13(6): 554 |
|[Pubmed] | [DOI]|
||Vascular territorial segmentation and volumetric blood flow measurement using dynamic contrast enhanced magnetic resonance angiography of the brain
| ||Oren Geri,Shelly I Shiran,Jonathan Roth,Moran Artzi,Liat Ben-Sira,Dafna Ben Bashat |
| ||Journal of Cerebral Blood Flow & Metabolism. 2017; 37(10): 3446 |
|[Pubmed] | [DOI]|
||Multimodal CT techniques for cerebrovascular and hemodynamic evaluation of ischemic stroke: occlusion, collaterals, and perfusion
| ||Baixue Jia,Fabien Scalzo,Elijah Agbayani,Graham W. Woolf,Liping Liu,Zhongrong Miao,David S. Liebeskind |
| ||Expert Review of Neurotherapeutics. 2016; : 1 |
|[Pubmed] | [DOI]|