Myocardial Segmentation of Area at Risk Based on Coronary Computed Tomography Angiography and Voronoi Diagram in Comparison with Magnetic Resonance Perfusion Imaging

Purpose: To assess the clinical feasibility of automated segmentation of the myocardial area at risk (MAAR) using coronary computed tomography angiography (CT-MAAR), as compared to stress magnetic resonance myocardial perfusion imaging (MR-MPI). Materials and Methods: Thirty patients who underwent coronary computed tomography angiography (CTA) and stress MR-MPI were retrospectively evaluated. The myocardial territory of the left ventricle (LV) distal to coronary artery stenosis (≥50% or ≥70% stenosis on coronary CTA) was three-dimensionally quantified using a Voronoi diagram. The ratio of all stenosis-related territories to the LV volume was defined as CT-MAAR (%-LV volume). The proportion of segments with perfusion defects in stress MR-MPI to the total of 16 segments (range: 0% 100%; with a 6.3%-interval scale) was defined as the reference. Correlation was assessed using Spearman’s test. The capability of CT-MAAR to predict the ischemic burden was assessed. Results: Stress MR-MPI depicted a median ischemic burden of 25.2% (range: 18.9% 44.1%) in 30 patients without myocardial infarction. When CTA stenosis criteria of ≥50% (n = 30) and ≥70% (n = 27) were applied to estimate CT-MAAR, the median CT-MAAR values were 48.2% (31.6% 64.3%) and 32.5% (23.7% 51.9%), respectively. The correlations between the CT-MAAR values and the MR-based ischemic burden were significant (0.73 and 0.97 for ≥50% and ≥70% stenosis, respectively). CTMAAR predicted the MR-based ischemic burden within ±1 segment of %-LV (6.3%) in 40% (12/30) of patients with ≥50% stenosis, and in 81.5% (22/27) of How to cite this paper: Fukuyama, N., Kido, T., Kurata, A., Tanabe, Y., Kido, T., Yokoi, T., Ogawa, R., Nishiyama, H., Uetani, T. and Mochizuki, T. (2017) Myocardial Segmentation of Area at Risk Based on Coronary Computed Tomography Angiography and Voronoi Diagram in Comparison with Magnetic Resonance Perfusion Imaging. Open Journal of Radiology, 7, 922. https://doi.org/10.4236/ojrad.2017.71002 Received: January 24, 2017 Accepted: February 26, 2017 Published: March 1, 2017 Copyright © 2017 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY 4.0). http://creativecommons.org/licenses/by/4.0/

patients with ≥70% stenosis. Conclusions: Comprehensive assessment of resting coronary CTA combined with Voronoi diagram-based myocardial segmentation may help predict the myocardial ischemic burden in patients with severe coronary CTA stenosis.

Keywords
Computed Tomography, Ischemia, Myocardial Area at Risk

Introduction
Coronary computed tomography angiography (CTA) is widely used in clinical practice for assessing obstructive coronary artery disease (CAD), because of its high sensitivity and negative predictive value [1] [2] [3]. However, because of the limited spatial resolution, further evaluation, such as invasive coronary angiography (ICA) or stress myocardial perfusion imaging (MPI), is often required in the diagnostic workflow of coronary artery disease (CAD) [4]. Stress MPI, using single-photon-emission computed tomography (SPECT) and magnetic resonance imaging (MR), can non-invasively detect and quantify ischemic myocardium as a standard reference [5] [6], and is useful as the ischemic ratio of the left ventricular (LV) myocardium provides important information for diagnostic and therapeutic purposes in patients with CAD [7]. Furthermore, the Voronoi diagram is a type of centerline method that divides space (i.e., volume) by seeded points or lines [8], and is utilized in industry and geography. In diagnostic imaging, a previous study has shown the usefulness of this algorithm for liver segmentation, based on CT portal venography [9], and a recent study has reported that ICA-based stenosis-related CT myocardial territory correlates with the SPECT-based myocardial area at risk (MAAR) [10]. Coronary CTA stenosis-related CT myocardial territory is an assumption of the maximum MAAR that is obtained from a resting coronary CTA dataset. Stress MR-MPI has a high spatial resolution and yields better diagnostic performance in assessing multi-vessel disease in comparison with SPECT-MPI [11] [12]. Thus, this study aimed to assess the clinical feasibility of applying automated segmentation of MAAR using coronary CTA, as compared to stress MR-MPI.

Study Design
This retrospective observational study was approved by the local ethics committee (clinical research 1509002). The need for informed consent was waived because of the retrospective nature of the study. From June 2009 to March 2016, we retrospectively collected the records of 91 consecutive patients who underwent coronary CTA and stress MR-MPI within an interval of less than 6 months.
All patient information was protected in compliance with the guidelines pro-vided by the institutional review board. The flow chart for patient selection is shown in Figure 1. Patients with myocardial ischemia diagnosed by stress MR-MPI were excluded if they fulfilled the following criteria: 1) no invasive coronary angiography (ICA) available; 2) previous history of revascularization therapy, such as coronary artery bypass grafting (CABG) or percutaneous coronary intervention (PCI); 3) heart valve disease; 4) cardiomyopathy; 5) total coronary artery occlusion confirmed byICA and 6) poor image quality of coronary CTA.

Stress Cardiac MR-MPI
We used a 3-T MR system (Achieva 3.0 T Quasar Dual; Philips Healthcare, Cleveland, OH, USA) equipped with a 32-element cardiac phased-array coil.
Using an established comprehensive cardiac MR protocol [13], stress and rest MR perfusion and late images were obtained. For the stress image, pharmacological hyperemia was induced with an adenosine triphosphate infusion (0.16 mg·min −1 •kg −1 ) for ≥3 min. Stress and resting MR perfusion images (for 35 s) were obtained from three short-axis images of the basal, mid-, and apical LV, using a two-dimensional (2D) T1 turbo field-echo sequence with k-space and time broad-use linear acquisition speed-up technique (k-t BLAST), after the injection of gadopentetate dimeglumine (0.05 mmol/kg; Magnevist; Schering, Berlin, Germany), followed by a 30-mL saline chaser, at an injection rate of 4 mL/s. Late images were obtained 10 min after the resting image using an inversionrecovery 3D T1 turbo field-echo sequence. The inversion time was individually determined using the Look-Locker sequence.

Invasive Coronary Angiography
ICA was performed following the standard institutional catheterization approach. Quantitative coronary analysis was performed by an independent cardiologist (T.U., 15 years of experience), who was blinded to other results, using commercially available software (CAAS5.9; Pie Medical Imaging, Maastricht, the Netherlands). Coronary artery stenosis ≥50% and ≥70% were considered as significant and obstructive CAD, respectively. When multiple stenoses were seen in two or more segments, the proximal stenosis was defined as the culprit stenosis.
All coronary artery segments and stenotic lesions (≥50% stenosis) were classified by the main coronary vessels: the left anterior descend-ing artery (LAD), including diagonal branches, the left circumflex artery (LCX), including the high lateral branch and ramus intermedius, and the right coronary artery (RCA), based on the American Heart Association guidelines [14].

MR-Based Ischemic Burden
The 16-segment model, excluding the apex, was applied [15]. The perfusion images were visually analyzed by two observers (T.K. and R.O., with 3 years and 8 years of experience in cardiac MR, respectively). These observers were blinded to all other data. If a perfusion defect in the LV myocardium was present in four or more cardiac cycles in the stress image, but not present in the resting image, it was considered to be a stress-induced perfusion defect [16]. When a segment with ≥50% of stress-induced perfusion defect was seen along the circumferential direction of the LV short axial image, corresponding to significant ICA stenosis by vessel-based analysis, it was defined as the ischemic segment in this study.
Discrepancies between the opinions of the first two observers were solved by a third senior observer (T. K., with 15 years' experience).In the late image, a segment with any late gadolinium enhancement was defined as a segment with myocardial infarction, irrespective of the presence or absence of myocardial ischemia. Patients with any segments showing late gadolinium enhancement were excluded from the study. The ratio of the number of segments with myocardial ischemia to the total of 16 segments was defined as the standard reference in this study (range: 0% -100% with a 6.3%-interval scale). The vesselbased analysis of the 16-segment model was based on a standardized myocardial segmentation [15].

Assessment of Coronary CTA
All CTA images were evaluated by two observers (Y.T. and T.Y., who had 4 years and 2 years of experience in cardiac CT, respectively) using a commercially available workstation (Aquarius intuition; TERARECON, Inc., Tokyo, Japan).
First, an overall assessment of image quality was performed at the subject level to assess misalignment of the coronary arteries between the slabs and discontinuity of coronary vessels due to motion artifacts and arrhythmia. If any of the aforementioned requirements were unsuitable, the patient was excluded from the study. Using a single dataset of coronary CTA with sufficient image quality, stenosis severity per segment was semiquantitatively assessed using the standard guidelines for reporting coronary CTA [17]: "non-CAD", 0% -49%; "moderate", 50% -69%; "severe", 70% -99%; and "occluded", 100%. The unassessable calcified segments were assumed to indicate severe stenosis. When multiple stenoses were found in the same coronary segment and vessel, the proximal stenosis was considered to be the culprit stenosis. Discrepancies were solved by consensus.
Coronary artery segmentation was applied, in addition to ICA.

CTA-Based MAAR
The same two observers evaluated the coronary CTA-based myocardial territory using dedicated software (TVA; TERARECON Inc. Tokyo, Japan). Post-processing of Voronoi diagram-based myocardial segmentation using coronary CTA was shown in Figure 2. In a series of Voronoi diagram-based myocardial segmentations based on a single CT dataset, the LV myocardium was automatically extracted using basic cardiac function analysis, and the coronary arteries were extracted in a semiautomatic manner. The software three-dimensionally integrated the two datasets and quantified the LV territories distal to any point on coronary CTA irrespective of coronary stenosis severity.
For this study, the stenosis-related CT myocardial territory was quantified at both levels of stenosis (≥50% and ≥70% on coronary CTA). The discrepancies of the stenosis-related CT myocardial territories were individually reviewed and solved by consensus. The two sums of all stenosis-related CT myocardial territories (i.e., the ratio of the LV volume at risk to the whole LV volume) were calculated using the two standards of the CTA stenosis (≥50% and ≥70%) as significant and obstructive CTA stenosis-based CT-MAAR, respectively.

Statistical Analysis
Categorical variables were expressed as proportions and continuous variables were Altman plotting expressed as the mean ± the standard deviation or as the median (interquartile range), as appropriate. With regard to coronary CTA stenosis severity and the stenosis-related CT myocardial territory for stenotic lesions ≥50%, the intra-and inter-observer reproducibility of the two operators were assessed using Spearman's test and Bland-Altman plotting.

Correlations between CTA-based MAAR and MR-based ischemic burden
were evaluated with Spearman's test. The ability of CTA-based MAAR to estimate the MR-based ischemic burden within ±6.3%, corresponding to a single segment of the LV (100/16 = 6.3) in the assessment of MR-based ischemia, was calculated. All analyses were performed using JMP version 11 (SAS Institute, Cary, NC, USA). For all analyses, p < 0.05 was significant.

Stenosis-Related CT Myocardial Territory
Seventy-seven coronary CTA lesions with a stenosis of ≥50% were assessed. Voronoi-based segmentation was successfully performed for all lesions. The respesentati ve for Compute cases are shown in Figure 3 and Figure 4. The intra-and   inter-observer agreements of the significant CTA-related myocardial territories were 0.99 and 0.99, respectively. The mean differences for intra-and inter-observer measurements of CT-based MAAR were 0.004% (95% confidence interval [CI], −0.053% to 0.045%) and −0.003 (95% CI, −0.030% to 0.024%), respectively.

Discussion
In this study, we showed that: 1) the reproducibility of the procedural steps for assessment of the CTA-based MAAR was good, and that 2) severe coronary CTA stenosis-based MAAR more accurately estimated the MR-based myocardial ischemic burden than did moderate stenosis.
Stress MR-MPI has been established as a useful diagnostic tool for detecting obstructive CAD [11]. The high spatial resolution of MR-MPI enables the delineation of subendocardial ischemia and myocardial infarction in patients with CAD [18]. Therefore, a comprehensive cardiac protocol involving the use of stress MR-MPI is helpful for diagnosing and managing CAD, even in patients with multivessel disease [19]. Stress MPI is essential for assessing the MAAR; therefore, many researchers have investigated stress-induced myocardial perfusion abnormalities using SPECT, MR, and CT [20] [21] [22].
By using a Voronoi diagram, CTA-based myocardial segmentation can indicate the theoretical maximum potential myocardial territory to a seeded point on the coronary CT angiogram. A recent study has demonstrated the high reprodu-cibility of CTA-based LV myocardial territory and the good correlation of CTbased MAAR with SPECT-based MAAR [10]. SPECT has been established as a clinical standard because of the substantial evidence supporting its use in the diagnosis and management of CAD [5] [20] [23]. However, some studies [11] [12] have discussed the potential disadvantages of SPECT as compared to MR-MPI, such as its lower spatial resolution and visualization of relative myocardial perfusion abnormalities.
Some researchers have reported clinical application of Voronoi diagrams in cardiovascular imaging [24] [25] [26]. Termeer et al. [25] utilized the Voronoi algorithm for estimating the MAAR on a 2D bull's eye map of the 17-segment model. Kang et al. [26] agreements of recently reported that the impact of coronary stenosis severity as assessed by intravascular ultrasound on the dependent myocardial territory is related to the severity of the invasive fractional flow reserve.
We considered that the CTA stenosis-based CT MAAR might contain some artificial post-processing.  [28]. The present study may show the additional value of this method for further examinations, such as stress MPI and invasive examination in the assessment of myocardial ischemia.
The present study has several limitations. First, the study population was relatively small. Second, stress MR-MPI as a reference was obtained with three cardiac short axial images, assessed using a conventional 16-segment model (excluding the apex), with a step-formed ordinal scale per 6.3%, independent from the morphological volume-based or the transmural extent-based evaluation. It is to be noted that these differences between CT-MAAR and stress MR-MPI might potential place a limitation on the evaluation of CT-MAAR in the study. Third, this study excluded patients with myocardial infarction because a myocardial scar may influence the quantification of CT-MAAR, by affecting the volumetric estimation of CT-MAAR, and the true extent of myocardial ischemia. Fourth, the CT-MAAR was estimated based on the anatomical location of coronary stenosis, independent of whether the lesion was diagnosed as hemodynamically significant. Fifth, the precision of the ischemic burden in stress MR-MPI was not objectively validated on anatomical territory that corresponded to coronary artery stenosis. Further studies with a large number of patients will be required to clarify the significance of these findings in clinical practice.

Conclusion
Voronoi diagram-based myocardial segmentation has the potential for predicting the MAAR in patients with obstructive coronary stenosis on CTA. The estimation of MAAR using resting coronary CTA may provide useful information in clinical decision-making.