OUP user menu

Computer-assisted myocardial blush quantification after percutaneous coronary angioplasty for acute myocardial infarction: a substudy from the TAPAS trial

Mathijs Vogelzang, Pieter J. Vlaar, Tone Svilaas, Diny Amo, Maarten W.N. Nijsten, Felix Zijlstra
DOI: http://dx.doi.org/10.1093/eurheartj/ehn542 594-599 First published online: 24 January 2009

Abstract

Aims Myocardial reperfusion after acute myocardial infarction can be angiographically assessed by the myocardial blush grade (MBG) or TIMI Perfusion Grade. These scores are based on subjective human judgement and lead to a score of four categories. A more operator-independent way of scoring myocardial perfusion may facilitate research in this area.

Methods and results We designed the ‘Quantitative Blush Evaluator’ (QuBE), a computer program which calculates a score for myocardial perfusion. This program will be freely available as open source software. The inter-observer concordance was 97.7%. We calculated values on prospectively collected angiograms in patients with acute ST-elevation myocardial infarction from the TAPAS trial. Quantitative blush evaluator values could be assessed on 790 out of 980 collected angiograms (81%). The QuBE score correlated significantly with MBG as determined by a core lab. The QuBE score predicted complete ST-elevation resolution, low enzyme levels, and 1 year survival (all P < 0.001). Quantitative blush evaluator value was an independent predictor of mortality at 1 year [OR 0.40 (0.17–0.90), P = 0.02].

Conclusion The QuBE program provides a practical, freely available computer-assisted assessment of myocardial perfusion. The QuBE score provides a useful surrogate endpoint in trials of therapies aimed at improving myocardial reperfusion.

Keywords
  • Myocardial infarction
  • Coronary angiography
  • Blush
  • Myocardial perfusion
  • Quantification

Introduction

Mechanical reperfusion is the preferred treatment modality for acute ST-elevation myocardial infarction (STEMI).1 Despite adequate restoration of epicardial coronary flow, the incidence of adverse outcomes remain high in certain subgroups. Impaired myocardial reperfusion is an important determinant of adverse outcome in patients with restored epicardial flow.2,3 The myocardial blush grade (MBG) has been devised for the visual assessment of myocardial reperfusion after primary percutaneous coronary intervention (PCI), and this score is an independent predictor for adverse outcome.2,4 Similarly, the TIMI myocardial perfusion grade has been designed for patients receiving thrombolytic therapy.5

Although the visual assessment of myocardial reperfusion has been successfully used for risk stratification and as surrogate endpoint in clinical trials, important drawbacks complicate the widespread use of this measure. The visual assessment requires an experienced observer, and is associated with marked variability and an intrinsic limited reproducibility.6 Recently, myocardial blush was quantified by computer in a small cohort of patients with STEMI and predicted left ventricular ejection fraction as measured by echocardiography.7 We hypothesized that computer analysis of myocardial reperfusion on the digital coronary angiogram should be feasible in routine practice and should predict adverse clinical outcome after STEMI. In addition to being less observer-dependent, computer-assisted analysis may yield a more fine-grained assessment of reperfusion, facilitating comparisons between patients and patient groups. We tested this hypothesis in patients enrolled in the TAPAS trial.8,9

Methods

Quantitative blush evaluation program

We developed a computer program for the analysis of digital coronary angiograms in-house, named QuBE (Quantitative Blush Evaluator). This program loads coronary angiograms in standard DICOM format. From the different runs taken during one procedure, the operator selects the angiogram to use for assessment. On this angiogram, the operator indicates a polygonal shape that contains the distal infarct-related area. Figure 1A shows a typical screenshot of the program. When the operator has drawn the polygonal area the program calculates the QuBE value as follows: For each frame, a possible translation offset is determined compared with the other frames to correct for panning motions. This offset is calculated by trying all possible motions with the change in x-pixel offset ranging from −10 to 10 and the change in y-pixel offset also ranging from −10 to 10. For each change, the correlation between the old and the new frame is calculated, and the shift with the best correlation between the frames is taken. The polygon is shifted in each frame according to the calculated movement offset. When a angiographic run is longer than 10 s, only the first 10 s are analysed. In each frame of the run, a background mask is composed by taking a median filter. This background mask contains all large-scale structures, such as the diaphragm and large blood vessels, but no smaller-sized structures. This mask is subtracted from the original frame, so that only smaller-sized structures remain. All pixels in the polygon are subdivided in blocks of 5 by 5 pixels. For each block, the average of the darkest few pixels is taken (i.e. the pixels which have the most contrast compared with the background). The value of a single frame is calculated as the average of the best 50% of pixel blocks. As a value is calculated for each single frame in the angiogram, there is no need to specifically indicate end-diastolic frames. The values for all frames provide a curve as shown in the right part of Figure 1A. The MBG primarily aims to quantify the increase in greyvalue over time.2,4 The grading of TIMI myocardial perfusion grade includes the amount of washout of contrast from the myocardium.5 We therefore let the QuBE value reflect both the filling and emptying phase of the vessels, by summing the maximum increase in greyvalue and the maximum decrease after that.

Figure 1

The Quantitative Blush Evaluator program. (A) A typical screenshot in which an operator has indicated the region of interest on a coronary angiogram of a right coronary artery. The curve representing the quantified value in all frames is shown to the right. The Quantitative Blush Evaluator value equals the maximum increment (from frame 8 to 24) plus the maximum decrement after that (frame 24 to 92). (B) Agreement of Quantitative Blush Evaluator scores of 30 angiograms assessed twice by the same operator and twice by different operators with the concordance coefficient.

Figure 1B shows the agreement of the method when scored twice by a single operator (intra-observer variability), or when scored by two different operators (inter-observer variability). Scoring of a QuBE value typically took 1 min per patient.

The QuBE program will be provided as open source software on the site http://qube.sf.net after publication. Distribution as open source software makes the program free to use, redistribute, and modify.

Patients

We included patients who were enrolled in the TAPAS trial.810 The TAPAS trial was a trial in which STEMI patients were randomly assigned to undergo thrombus aspiration during PCI or a conventional PCI. The trial aimed at including typical STEMI patients, and therefore had wide inclusion criteria and few exclusion criteria. Patients were recruited from January 2005 to December 2006. Details of the study protocol, patient selection, baseline characteristics, pharmacological and interventional therapies have been described.8,10 In the current analysis, we included all patients who underwent PCI who had an angiogram on which the corelab had been able to visually assess the MBG.

TIMI risk scores11 were calculated for all patients.

Quantitative blush evaluator values and myocardial blush grading

Quantitative blush evaluator values were scored on angiograms made at the end of the primary angioplasty procedure and following administration of nitroglycerin. All angiograms were filmed at 12.5 frames per second in a single plane catheterization lab (Siemens AG, Germany). The QuBE value was determined for all included patients by one observer. TIMI flow and the MBG were assessed by an independent observer at a corelab (Cordinamo, Wezep, The Netherlands).

We included all baseline variables for all patients in our study, and used these to assess which variables at baseline correlate with the calculated QuBE value. We hypothesized that good perfusion would be correlated to age, infarct location, ischaemic time, and patency (TIMI flow) before and after intervention, as found in earlier studies.2,4 Because the quantification uses absolute intensities, we hypothesized that patients with a high body mass index might have lower QuBE values.

Outcome measures

The study protocol included analysis of the ECG both before and after the intervention. The sum of ST-elevation over all leads was determined on the initial ECG. Full ST-elevation resolution was defined as a residual elevation of <30% on the post-procedural ECG.

Enzyme release was routinely measured during the stay at the coronary care unit after the primary PCI. We calculated the area under the serum creatine kinase curve over the first 24 h post-PCI. No curve was calculated for patients having <2 measurements, or only measurements with a timespan <16 h. This mainly occurred in patients referred early to one of the six hospitals in our catchment area with no PCI facility.

Clinical follow-up was obtained by written interviews and telephone interviews. To assess the added value of QuBE, its association with outcome measures was analysed both in the complete patient group and in the subgroup of patients in which adequate reperfusion was achieved. Adequate reperfusion has previously been defined as TIMI flow 3 post-PCI and MBG 2/3.4

Statistical analysis

Statistics are expressed as median (interquartile range, IQR). Groups were compared with the Mann–Whitney U test for continuous or ordinal variables and the χ2 or Fisher’s exact test for categorical data when appropriate. To allow group comparisons in Table 1, patients were classified according to the tertile of their QuBE value. Inter- and intraobserver variances were determined by selecting 30 random angiograms to be graded by two different observers and by one observer on different days. Lin’s concordance coefficient was calculated to quantify agreement. Analysis of which of the factors listed in Table 1 correlate with QuBE score was performed by linear regression. The linearity assumption for continuous variables was assessed using residual plots. Each variable was tested separately and all variables with a univariable P-value smaller than 0.1 were included in the multivariable model. Predictiveness of QuBE for 1 year mortality was tested in a binary logistic regression model. Besides QuBE score, TIMI risk score, MBG, and randomization allocation were included in the model. A two-tailed P-value lower than 0.05 was considered statistically significant. All statistics were performed by the R software package (version 2.5.1). The authors had full access to the data and take responsibility for its integrity. All authors have read and agree to the manuscript as written.

View this table:
Table 1

Patient characteristics

1st QuBE tertile2nd QuBE tertile3rd QuBE tertileP-value
n261260269
QuBE value (median, range)7.5 (1.3–10.2)13.0 (10.2–15.4)18.7 (15.5–36.4)
Age67 (55–76)61 (54–71.25)61 (52–69)<0.0001
Male sex, %68 (177/261)74 (192/260)71 (191/269)0.17
Ischaemic time (min)202 (144–346)187 (130–266)187 (130–245)0.0002
Body mass index26.9 (24.7–30.3)26.4 (24.2–28.7)26.0 (24.2–28.1)0.006
Heart rate80 (69–93)80 (65–90)77 (63–87)0.02
Systolic blood pressure127 (111–145)125 (108–145)128 (110–144)0.45
Diastolic blood pressure75 (64–86)72 (65–85)75 (63–83)0.34
Culprit vessel, %<0.0001
 RCA24 (62/261)42 (108/260)54 (145/269)
 LAD44 (116/261)44 (115/260)38 (101/269)
 Cx30% (77/261)14 (36/260)8 (21/269)
TIMI flow pre-PCI, %0.005
 0/167 (174/260)58 (150/257)52 (136/264)
 216 (41/260)19 (49/257)22 (58/264)
 317 (45/260)23 (58/257)27 (70/264)
Stent placement, %89 (224/251)96 (236/245)95 (240/253)0.004
GP IIb/IIIa inhibitor treatment, %90 (236/261)92 (240/260)95 (256/269)0.004
TIMI flow post-PCI, %<0.0001
 0/14.6 (12/261)1.5 (4/260)0.4 (1/269)
 218 (47/261)13 (34/260)7.1 (19/269)
 377 (202/261)85 (222/260)93 (249/269)
  • Values are represented as median (IQR) or percentage, unless otherwise indicated.

  • MI, myocardial infarction; QuBE, Quantitative Blush Evaluator; CABG, coronary artery bypass grafting; PCI, percutaneous coronary intervention; RCA, right coronary artery; LAD, left anterior descending artery; Cx, circumflex artery.

Results

Figure 2 shows a flow diagram of patients in this study. Out of 1071 included patients in the TAPAS trial, 980 met the inclusion criteria of a performed PCI, and MBG scored by the core lab. Of these patients, a number of angiograms (190, 19%) were not assessable for the reasons given in Figure 2. The flexible identification of a polygonal area allowed the operator to circumvent overlapping arteries, and therefore only eight angiograms could not be used due to overlap of the region of interest with non-infarcted blood vessels. The proportion (95% confidence interval) of successful scored angiograms was 80.6% (78.1–83.1%). Therefore 790 patients were included in this substudy. None of the variables listed in Table 1 nor any outcome variable was significantly different between the group of 790 patients with a QuBE score and the group of 190 patients who were excluded because of unsuitable angiograms.

Figure 2

Flow diagram of analysed patients and angiograms. Reasons of exclusion before and after Quantitative Blush Evaluator evaluation are shown. MBG, myocardial blush grade; IRA, infarct related artery.

Table 1 shows baseline patient characteristics according to tertiles of QuBE values. In a multivariable linear regression model including the variables with a P-value < 0.10, age, body mass index, heart rate, and TIMI flow both pre- and post-PCI remained as significant factors predicting the QuBE value. These factors accounted for 11% variance in QuBE value according to the r-value of the multivariable regression.

Quantitative blush evaluator and 1 year mortality

One-year all-cause mortality was available in 790/790 (100%) of patients. Total 1 year mortality was 46/790 (5.8%). Of these, 36 were cardiac deaths. Figure 3 shows distribution of QuBE values and mortality. In our study group, 596/748 (75.4%) of patients met criteria for optimal reperfusion (TIMI 3 flow combined with MBG 2 or 3). Within this group, the QuBE value still had discriminating power with respect to 1 year mortality (Figure 3B).

Figure 3

One-year mortality and Quantitative Blush Evaluator value. Mortality is shown for survivors and non-survivors. (A) All patients are analysed and (B) Only the patients with optimal reperfusion (TIMI flow 3 and blush grade 2 or 3). MBG, myocardial blush grade.

In multivariable binary logistic analysis including the TIMI risk score, randomization allocation, QuBE value, and MBG, only TIMI risk score and the QuBE value remained as independent predictors of mortality at 1 year (OR per 10 point QuBE increase 0.397, 95% CI 0.174–0.903, P = 0.02). When this analysis was performed with cardiac mortality instead of total mortality, the results were similar.

Quantitative blush evaluator and myocardial blush grade, ECG, and enzyme parameters

Myocardial blush grade was available for analysis in all patients. Both ECGs needed for analysis of ST resolution were available in 737/790 patients (93%), and the 24 h area under the curve of serum creatine kinase measurements could be assessed in 578/790 patients (73%). Figure 4 shows patterns of MBGs, ST resolution, and enzyme levels per tertile of QuBE value. All three parameters showed a strong correlation to the QuBE score (all P < 0.0001). Linear regression of QuBE values with the area under the curve of serum creatine kinase showed a significant correlation (P < 0.0001). This correlation remained significant in the subset of patients with optimal reperfusion defined as TIMI flow 3 post-PCI in combination with a MBG 2 or 3.

Figure 4

Blush, ECG, and enzyme parameters. Visual myocardial blush grade (MBG), full ST resolution and myocardial enzyme levels (CK-AUC 24 h: area under the curve of creatine kinase) and their association with Quantitative Blush Evaluator scores. MBG 0/1, no or minimal myocardial blush or contrast density; 2, moderate blush or contrast density; 3, normal blush (comparable with a non-infarct related artery).

Discussion

In this study, we developed and evaluated QuBE, a computer program for quantification of myocardial perfusion after PCI in STEMI patients. We found that perfusion values as quantified by QuBE were associated with high MBG scored at a corelab, ST-segment elevation resolution, smaller infarct sizes as measured by enzyme release, and survival at 1 year. Current reperfusion strategies are generally successful at achieving epicardial reperfusion. However, mortality rates after adequate infarct-related vessel reperfusion (TIMI flow 3) are still high in subgroups with inadequate myocardial reperfusion. In particular, loss of ventricular function with subsequent heart failure remains an important problem.2,4,12 For the development of new therapies it is important to be able to identify high-risk subgroups, and to implement appropriate surrogate endpoints to measure the desired effect.

The QuBE method provides a practical and efficient method to quantify myocardial perfusion. Although advanced techniques such as perfusion MRI may give a more exact assessment of perfusion,13 these methods require additional diagnostic procedures, and are therefore not applicable in large scale trials. The QuBE value only requires one specific angiogram taken at the end of the angioplasty procedure, and the result can be made immediately readily available. We were able to quantify perfusion in the majority (81%) of scored angiograms from a recently completed randomized controlled trial, despite the fact that they were not specifically made for this goal. From the ongoing ADAPT trial,14 we know that with more effort by interventional cardiologists to instruct the patient to hold his/her breath and not pan while shooting the blush sequence, the percentage of adequately scorable angiograms further increases to >90%. Furthermore, the fine-grained QuBE assessment allows more accurate measurements. Therefore, the ADAPT trial includes two paired blush sequences, before and after administrating intracoronary adenosine.14 The visually scored blush value of 0 to 3 would lack adequate precision to make a paired assessment possible. The increased power of the QuBE value has the potential to identify beneficial interventions in patient cohorts with limited size.

A number of limitations must be mentioned. Although we found good reproducibility of QuBE values, we have not assessed the differences that come from the specific acquisition machine and method used. At the moment, we do not know how parameters such as the amount of infused contrast, the exact quantity of nitrate administered, and the heart rate of the patient influence the QuBE value. These parameters have previously been found to potentially confound the TIMI frame count, a measure for epicardial flow.15

As QuBE will be released as open source, groups other than ours are free to experiment, adjust, and expand the QuBE program. We hope that future research will define the extent to which QuBE values are comparable between catheterization laboratories. We suggest that additional sites implementing QuBE scoring should start with scoring a cohort of 50–100 patients as calibration. When these values are significantly different from the ones from this study, QuBE values can be normalized using the mean and standard deviation, for instance to values ranging from 0 to 100. We also hypothesize that further analysis may reveal additional parameters that can be derived from the perfusion score profiles as shown in Figure 1.

In summary, the QuBE method provides a practical and feasible way to quantify myocardial perfusion after PCI, and may therefore be useful as risk indicator in clinical practice or surrogate endpoint in clinical trials.

Funding

This study had no external funding sources.

Conflict of interest: none declared.

Footnotes

  • This is a substudy of the TAPAS trial, Dutch trial register ID NTR914, ISRCTN 16716833.

References

View Abstract