Three-dimensional assessment of coronary high-intensity plaques with T1-weighted cardiovascular magnetic resonance imaging to predict periprocedural myocardial injury after elective percutaneous coronary intervention.

Background Periprocedural myocardial injury (pMI) is a common complication of elective percutaneous coronary intervention (PCI) that reduces some of the beneficial effects of coronary revascularization and impacts the risk of cardiovascular events. We developed a 3-dimensional volumetric cardiovascular magnetic resonance (CMR) method to evaluate coronary high intensity plaques and investigated their association with pMI after elective PCI. Methods Between October 2012 and October 2016, 141 patients with stable coronary artery disease underwent T1-weighted CMR imaging before PCI. A conventional 2-dimensional CMR plaque-to-myocardial signal intensity ratio (2D-PMR) and the newly developed 3-dimensional integral of PMR (3Di-PMR) were measured. 3Di-PMR was determined as the sum of PMRs above a threshold of > 1.0 for voxels in a target plaque. pMI was defined as high-sensitivity cardiac troponin T > 0.07 ng/mL. Results pMI following PCI was observed in 46 patients (33%). 3Di-PMR was significantly higher in patients with pMI than those without pMI. The optimal 3Di-PMR cutoff value for predicting pMI was 51 PMR*mm3 and the area under the receiver operating characteristic curve (0.753) was significantly greater than that for 2D-PMR (0.683, P = 0.015). 3Di-PMR was positively correlated with lipid volume (r = 0.449, P < 0.001) based on intravascular ultrasound. Stepwise multivariable analysis showed that 3Di-PMR ≥ 51 PMR*mm3 and the presence of a side branch at the PCI target lesion site were significant predictors of pMI (odds ratio [OR], 11.9; 95% confidence interval [CI], 4.6–30.4, P < 0.001; and OR, 4.14; 95% CI, 1.6–11.1, P = 0.005, respectively). Conclusions 3Di-PMR coronary assessment facilitates risk stratification for pMI after elective PCI. Trial registration retrospectively registered.


Introduction
Percutaneous coronary intervention (PCI) has become the most common procedure for coronary revascularization in patients with both stable and unstable coronary artery disease (CAD). Although technical advances in PCI have resulted in a safer therapeutic procedure, 5-30% of patients undergoing elective PCI demonstrate evidence of periprocedural myocardial injury (pMI) from the procedure itself [1,2]. The extent of pMI as determined by evaluation of creatine phosphokinase MB isoenzyme, high-sensitivity troponin T (hs-cTnT), and troponin I is significant enough to be prognostically important [1,3]. Therefore, pMI might reduce some of the beneficial effects of coronary revascularization.
Coronary high-intensity plaques (HIPs) with a high plaque-to-myocardial signal intensity ratio (PMR) on noncontrast T1-weighted (T1w) cardiovascular magnetic resonance (CMR) imaging are associated with future coronary events and PCI-related pMI [4][5][6][7]. However, current HIP-PMR evaluation is solely based on signal intensity without consideration of plaque volume, which yields a 2dimensional PMR (2D-PMR) based on a coronary plaque as a region of interest (ROI) [4,5]. Since larger plaque volume is associated with cardiovascular events such as acute myocardial infarction, which includes pMI [8,9], more accurate, quantitative assessment is needed to predict cardiovascular events.
Herein, we sought to examine whether 3-dimensional integral (3Di) volumetric assessment of coronary HIPs on non-contrast T1w imaging would have better predictive value for pMI after elective PCI compared with the current 2D-PMR in patients with stable CAD.

Three-dimensional plaque analysis with CMR
On CMR analysis, the coronary vasculature tree was subdivided into 8 segments [12]. In brief, the right coronary artery was analyzed in 3 segments (segments 1, 2, 3). The left coronary artery was analyzed in 4 segments that comprised the left anterior descending artery (segments 6, 7) and left circumflex artery (segments 11, 13). Since lesions in the left main trunk, both in the left anterior descending artery and left circumflex artery (segment 5), are associated with a large ischemic burden, PCI for left main stenosis is indicated only in patients with a low SYNTAX score and non-multivessel disease [13], and PCI for left main lesions may be susceptible to operator bias. Therefore we excluded left main lesions in the present analysis. For segment identification, segments were pre-defined according to the distance from the vessel's origin [5,6,12]. The location of a coronary plaque was determined by carefully comparing either CTA or CMR angiogram or invasive coronary arteriography images during PCI using fiduciary points (e.g. side branches and vessel bends). Once a coronary plaque had been confirmed with either CTA or coronary CMR angiogram, the corresponding areas on coronary T1w images were carefully matched using the surrounding cardiac and chest wall structures [11]. Fifty-seven patients underwent coronary CTA and co-registration between coronary T1w imaging and CTA (Additional file 1: Supplemental Methods and Table S1).
2D-PMR was assessed as previously described [4][5][6]11]. For 3D analysis of coronary plaques on non-contrast T1w imaging, OsiriX MD software (version 8.0.2, Pixmeo, Geneva, Switzerland) was used. The 3D integral of the plaqueto-myocardium signal intensity ratio (3Di-PMR) was defined as the integral of voxel volume (0.55 mm × 0.60 mm × 1.0 mm = 0.33 mm 3 ) multiplied by its PMR value > 1.0 from a coronary plaque (Fig. 1, Additional file 1: Figure S1, Figure  S2, Figure S3, and Additional file 2: Step 1, Videos S1, Additional file 3: Step 2, Videos S2, Additional file 4: Step 3, Videos S3). The algorithm for plaque segmentation was based on a region-growing technique, a function built into the OsiriX MD software consisting of the following steps ( Figure 2 and Additional file 2: Step 1, Videos S1, Additional file 3: Step 2, Videos S2, Additional file 4: Step 3, Videos S3) [14]. First, an elliptical ROI of ≥1.0 cm 2 in area was drawn around the myocardium near a coronary plaque. The mean signal intensity of the ROI was defined as the lower threshold of the signal intensity of the coronary plaque. Next, entire voxels of a coronary plaque above the threshold PMR of 1.0 were segmented automatically within contiguous slices to calculate both the integral signal intensity and voxel volume (3D region growing: Step1, Additional file 1: Figure  S1 and Additional file 2: Step 1, Video S1) [14]. Segments with a PMR above the threshold that were not analyzed by the automatic 3D segmentation, which was defined as those with expansion of the ROI beyond the boundaries of the vessel, were divided into 2 groups according to the course of the coronary vessel: those running along axial slices (Step 2, segments 1, 3, 6, 7) and those running perpendicular to axial slices (Step 3, segments 2, 11, 13). These segments were then analyzed using the methods described below.
Since the mean diameter of proximal coronary segments in Japanese subjects is 4-5 mm [15], we used a slice thickness of 1 mm to cover the entirety of each vessel. Therefore, in the Step 2 method (for segments 1, 3, 6, 7), 3Di-PMR was determined by performing automatic contiguous 2D segmentation of each coronary plaque above the PMR threshold, in not more than 5 contiguous slices and without expansion beyond the boundaries of the vessel (Additional file 1: Figure  S2-a, S2-b and Additional file 3: Step 2, Video S2).
For segments 2, 11, 13, the lower segmentation threshold was increased from the signal intensity of the nearby myocardium in increments of 5 units until there was no longer expansion beyond the vessel boundaries (Additional file 1: Figure S3 and Additional file 4: Step 3, Video 3).

PCI
PCI was performed as previously reported [16]. pMI after elective PCI was defined as an increase in serum hs-cTnT levels to more than 5 times the upper limit of normal (0.07 ng/ Fig. 1 Principle behind 3-dimensional (3D) plaque assessment on T1-weighted imaging. Gray cubes represent voxels. α n represents the signal intensity of each voxel with higher signal intensity than that of nearby myocardium. Entire voxels of a coronary plaque that are above the signal intensity of nearby myocardium (plaque-to-myocardial signal intensity ratio > 1.0) were segmented within contiguous slices (surrounded by yellow dotted lines) to calculate the integral of signal intensity and voxel volume mL) at 24 h after PCI. Slow flow was defined as the Thrombolysis In Myocardial Infarction (TIMI) grade 0, 1, or 2 flow in the distal infarct-related artery despite the absence of occlusion or dissection at the treatment site. The presence of a side branch was defined as the presence of a vessel of ≥1.5 mm in diameter within the target lesion. Side branch occlusion was defined as TIMI grade 0, 1, or 2 flow during PCI.

Intravascular ultrasound image analysis
Intravascular ultrasound (IVUS) images were obtained during PCI and analyzed in 126 patients. IVUS was performed using a commercially available IVUS catheter (View It; Terumo, Tokyo, Japan) with 0.5 mm/sec autopullback. Quantitative and qualitative analysis were performed in a blinded manner as previously described [4,6,7]. Positive remodeling was defined as a remodeling index of > 1.05. Ultrasound attenuation was defined as IVUS images with backward signal attenuation of ≥180°behind the plaque without dense calcium. The presence of ultrasound attenuation was defined as longitudinal attenuation length of ≥5 mm [17]. For tissue characterization, IVUS data were analyzed using the manufacturer's default setting on the basis of previous data used to define a range of integrated backscatter values. Coronary plaques were classified into three categories: lipid pool (blue), fibrosis (green/ yellow), and calcification (red) [9].

Statistical analysis
Values are presented as medians (IQR). Values were compared using the Mann-Whitney U test or Kruskal-Wallis test for multiple comparisons followed by the Steel-Dwass test for post hoc analysis. Categorical baseline variables were compared using Fisher's exact test or the chisquared test as appropriate. 3Di-PMR and 2D-PMR cutoff values for the development of coronary events were determined with receiver operating characteristic (ROC) analysis. To identify risk factors for pMI, univariable and multivariable logistic regression models were constructed using 3Di-PMR, attenuation, positive remodeling, calcification, type B2/C lesion, and presence of a side branch.
Stepwise multivariable logistic regression with a P value of 0.10 for backward elimination was performed to select the best predictive model. All analyses were conducted using JMP (version 12, SAS Japan, Tokyo, Japan) and Stata, version 14 (StataCorp LP, College Station, Texas, USA). A P value less than 0.05 was considered statistically significant. Table 1 summarizes the baseline characteristics of the study patients. Of 141 patients, pMI was observed in 46 patients (33%). Both 3Di-PMR and 2D-PMR were significantly higher in patients with pMI than those without pMI (P < 0.001), although age, gender, coronary risk factors, medications, and pre-PCI hs-cTnT level were similar. As shown in Additional file 1: Figure S4-C, 3Di-PMR is strongly and positively correlated with 2D-PMR (r = 0.832, P < 0.001). Intraclass correlation coefficients with 95% confidence intervals (CIs) were calculated to assess intra-and interreader agreement regarding 3Di-PMR. The interval between initial analysis of 3Di-PMR of HIPs and secondary analysis was 3 months. The intra-and inter-reader intraclass correlation coefficients for 3Di-PMR were 0.947 (95% CI, 0.911-0.969) and 0.926 (95% CI, 0.876-0.957), respectively. All correlation coefficients for calculating 3Di-PMR were greater than 0.8, indicating good intra-and interobserver agreement.
As shown in Table 4, univariable Cox regression analysis showed that 3Di-PMR ≥ 51 PMR*mm 3 , attenuation detected on IVUS, and the presence of a side branch at the PCI target lesion site were associated with pMI. Multivariable analysis showed that 3Di-PMR ≥ 51 PMR*mm 3 and the presence of a side branch at the PCI target lesion site     5 Incidence of periprocedural myocardial injury (pMI) based on 3Di-PMR and 2D-PMR cutoff values. The red and blue bars represent patients with 3Di-PMR ≥ 51 PMR*mm 3 and < 51 PMR*mm 3 , respectively. P < 0.001 based on the chi-squared test.* P = 0.006 vs. 2D high 3D low group. † P < 0.001 vs. 2D low 3D low group, and P = 0.003 vs. 2D high 3D low group were significant predictors of pMI (odds ratio [OR], 11.9; 95% CI, 4.63-30.4; P < 0.001 and OR, 4.14; 95% CI, 1.55-11.1, P = 0.005; respectively). Finally, we compared between the 3Di-PMR and coronary CTA derived predictors of pMI previously described to be indicative of high-risk plaque among the 57 patients who underwent CTA (Additional file 1: Tables S1-S2). From ROC analysis, 3Di-PMR also had higher AUC

Discussion
The major finding of this study is that the presence of a coronary HIP with 3Di-PMR ≥ 51 PMR*mm 3 detected by non-contrast T1w imaging was a significant independent predictor of PCI-related pMI compared to conventional 2D-PMR and other IVUS-or CAG-derived predictors. Thus, noninvasive characterization of coronary plaques based on 3Di-PMR with CMR, which includes the concept of plaque burden extent, is clinically informative for risk stratification of patients with elective PCI.
Although PCI is an important coronary revascularization strategy, especially in patients with CAD, the effect of PCI on clinical outcomes is limited despite the development of contemporary PCI techniques and devices [18]. Thus, reducing the number of complications during PCI may improve the clinical outcomes of PCI. pMI is a complication of PCI whose mechanism is thought to be related to side branch occlusion, atherosclerotic embolism, or both during PCI and may be associated with cardiovascular events after PCI [2,3]. The present study shows that novel 3D evaluation of coronary plaques using noncontrast T1w imaging could facilitate the prediction of pMI after elective PCI. Previous reports have shown that characterizing coronary HIPs with non-contrast T1w imaging is a promising noninvasive method and a novel biomarker for identifying high-risk coronary plaques without any radiation or iodinated contrast exposure [4][5][6][7]. Currently, coronary plaques on T1w imaging are quantified using the signal intensity ratio between plaque and a reference point (i.e., nearby myocardium). This value, 2D-PMR, characterizes plaques without taking into account the concept of plaque volume. Matsumoto et al. reported a relationship between qualitative coronary PMR morphology (e.g., intrawall or intraluminal high-intensity signals) visualized with non-contrast T1w imaging and their clinical severity [19]. In addition, since larger coronary plaque volume was associated with future cardiovascular events [8,9,20], we have developed a novel technique for quantitative 3D plaque analysis with non-contrast T1w imaging ( Figs. 1 and 2). Indeed, as shown in Table 3, the present 3D coronary plaque assessment produces a more sensitive predictor of pMI after elective PCI than current 2D-PMR and indices derived from IVUS or coronary angiography. It may also be a better predictor of pMI than CTAderived factors (Additional file 1: Tables S1-S2), because CMR was reported to be superior to CTA in terms of characterizing noncalcified, atherosclerotic plaques in an experimental rabbit atherosclerotic model [21], and current 3D plaque assessment may involve both plaque characteristics and their volume. As shown in Figure 5, the prevalence of pMI was higher in patients with 3Di-PMR ≥ 51 PMR*mm 3 than in those with 3Di-PMR < 51 PMR*mm 3 , irrespective of their 2D-PMR value. Additional file 1: Table S3 also showed that patients with higher 3Di-PMR had positive vessel remodeling compared with patients with lower 3Di-PMR, and patients with higher 2D-PMR had longer attenuation length compared with patients with lower 2D-PMR. Because coronary plaque volume and composition play an important role in pMI after elective PCI, the present 3D evaluation using non-contrast T1w imaging for coronary atherosclerosis may improve the accuracy of predicting cardiac events. To decrease the incidence of pMI and improve clinical 3Di-PMR 3-dimensional integral of the plaque-to-myocardium signal intensity ratio, CI Confidence interval, OR Odds ratio, PMR*mm 3 the unit of 3Di-PMR; defined as the integral of voxel volume multiplied by its PMR value > 1.0 from a coronary artery outcomes, this noninvasive preoperative evaluation before scheduled PCI might facilitate risk stratification for subgroups at high risk for pMI, who might then benefit from intensive pharmacological approaches (e.g., statins and antiplatelet therapy) and usage of filter devices during elective PCI procedures [22][23][24][25]. As shown in Figure 3 and Table 2, total plaque volume and lipid-rich plaque as assessed by IVUS were significantly correlated with 3Di-PMR. In carotid atherosclerosis, HIP atherosclerotic lesions detected by noncontrast T1w imaging consist of complex atheromas with large necrotic cores and intraplaque hemorrhage [26,27]. Our previous case report showed that coronary emboli during PCI, detected as a coronary HIP on noncontrast T1w imaging before PCI, is composed of a large necrotic core with cholesterol crystals and thrombus [10]. Additionally, a histological analysis using formalinfixed post-mortem human hearts, which were imaged at 1.5 T with T1w imaging with fat suppression, showed that the majority of coronary HIPs on T1w imaging was reflect intraplaque hemorrhage [28]. Thus, coronary HIPs detected on T1w imaging could include complex atheromas with intraplaque hemorrhage. The precise relationship between plaque burden with intraplaque hemorrhage and pMI during PCI remains unclear. Further studies are needed to clarify the relationship among MRI findings, associated histological and molecular characteristics, and clinical events in coronary atherosclerosis.

Limitations
Our study has several limitations. The number of patients was small and there might have been selection bias related to how patients were chosen for CMR. Second, the algorithm for 3D plaque evaluation requires discussion. The Step 1 algorithm, which is based on automated 3D region growing for a coronary plaque with higher signal intensity than nearby myocardium, allowed for 3D quantification of coronary plaques in 128 of 141 patients (91%). The other 13 patients required additional steps. In particular, the Step 3 algorithm required an increment of the lower segmentation threshold for plaque segmentation until there was no longer expansion beyond the vessel boundaries. These additional steps could hamper this assessment in more complex lesions and have little impact on predicting clinical outcomes. However, there is currently no method to quantify plaque morphology on T1w imaging. Thus, further studies to develop plugin software or other algorithms for plaque segmentation are needed for 3D plaque quantification with CMR in coronary atherosclerosis. Third, we did not perform systemic analyses to compare the findings between coronary CMR and CTA, which could also predict pMI.

Conclusion
Three-dimensional assessment of coronary plaques reflecting both plaque volume and composition facilitates more accurate risk stratification and prediction of pMI after elective PCI in stable CAD patients.