Quantifying “Normalized” Regional Left Ventricular Contractile Function in Ischemic Coronary Artery Disease

Abstract

Objective

When significant coronary lesions are identified by angiography, regional left ventricular (LV) contractile function often plays a role in determining candidacy for revascularization. To improve upon current subjective and non-quantitative metrics of regional LV function, we tested a z-score “normalization” of regional strain information quantified from clinically acquired high-resolution LV geometric datasets.

Methods

Test subjects (n=120) underwent cardiac MRI with multiple 3D strain parameters calculated from tissue tag-plane displacement data. Sixty normal volunteers contributed strain parameter data at each of 15,300 LV grid points to form a normal human strain database. Point-specific database comparisons were then made in 60 patients with documented coronary artery disease (CAD) by angiography. Patient-specific color-coded 3D LV maps of z-score normalized contractile function were generated.

Results

Upon blinded clinical review, 55% (33/60) of the CAD patients had significant regional contractile abnormalities by one of three “gold standard” criteria 1) Q waves on ECG; 2) infarct on radionuclide SPECT; or 3) akinesia or dyskinesia on echocardiography. Consistency between all gold standard metrics was found in only 19% (6/31) of CAD patients who had at least two available metrics. Blinded MRI-based multi-parametric strain z-score localization of contractile abnormalities was accurate in 89% (ECG), 97% (SPECT), and 95% (echocardiography).

Conclusions

Non-subjective normalization of regional LV contractile function by z-score calculation from a normal human strain database can localize and quantitatively display regional wall motion abnormalities in CAD patients. This high-resolution localization of regional wall motion abnormalities may help improve the accuracy of therapeutic intervention in CAD patients.

Introduction

Patient-specific laboratory blood test values fuel many of the clinical algorithms that are commonly utilized in the daily practice of evidence-based medicine. They are most often interpreted by reference to large normal human databases with “normal” ranges being defined by standard deviation (SD) from normal mean values. Therapeutic intervention is often triggered by patient-specific test values that fall greater than 2 standard deviations from the mean. The comparison of an individual value to a normal database mean is known as “normalization.” This process allows for non-subjective interpretation of laboratory values that is easily incorporated into decision-making in diagnostic and therapeutic clinical algorithms. Decision-making in cardiac diagnostic and therapeutic algorithms unfortunately stands in stark contrast to this universally accepted medical paradigm.

Many factors determine branch-point decisions in clinical algorithms when coronary angiography demonstrates a significantly obstructive coronary arterial lesion. One such factor is the contractile function of the left ventricular region supplied by the diseased coronary artery.^{1, 2} Clinicians most often make related decisions based upon typewritten characterizations of LV contractile function where “normalization” (comparison to the normal mean) is most commonly by observer visualization of non-quantified wall motion imaging such as echocardiography. This non-quantitative, observer-dependent normalization has been repeatedly demonstrated to be inconsistent across time, observer, and imaging technique.^3–8

Heterogeneous LV contractile injury is expected in patients with coronary artery disease, with the distribution being dependent upon specific coronary arterial atherosclerotic involvement. Somewhat surprisingly, our lab and others have also shown a consistent pattern of heterogeneous contractile injury in patients with dilated cardiomyopathy^9–11 and aortic insufficiency.¹² In other words, not only is normal human left ventricular strain heterogeneous,¹³ but most LV strain injury also occurs in a heterogeneous distribution. These studies establish the critical necessity of utilizing normalized strain metrics (by way of comparison to a normal human strain database^{9, 14}) in the analysis of regional myocardial contractile function.

It is our hypothesis that the methodologies to quantify, sub-regionally localize, and database-normalize cardiac contractile data using high-resolution imaging are now available for application in the clinical setting. We further hypothesize that this normalized information is most applicable in patients with CAD and can be presented in an easily assimilated visual format that displays high-density, regionally varying contractile information utilizing intuitive color-coded mapping over patient-specific three-dimensional LV geometry.

Methods

Patient Characteristics

The Human Research Protection Office at Washington University (St. Louis, MO) approved this study, and all subjects gave informed written consent. No sex-based or racial/ethnic-based exclusions were present during patient recruitment.

Sixty patients with coronary artery disease (defined as having at least one >70% lesion on coronary angiography) were enrolled in the study and underwent cardiac MRI with tissue tagging. An additional sixty healthy volunteers with no historical, physical, or clinical test evidence for any kind of heart disease also underwent cardiac MRI and contributed complete strain parameter information to a normal human strain database. Demographic data are presented in Table 1.

Table 1.

Characteristics of patients with CAD

Characteristic	Patients with CAD	Healthy Volunteers
Age (y)	60.7 ± 9.1	33.1 ± 10.8
Female sex	20% (12/60)	53 (32/60)
Diabetes	28% (14/50)	0%
Hypertension	84% (42/50)	0%
COPD	6% (3/50)	0%
Chronic Renal Insufficiency	32% 16/50	0%
Atrial Fibrillation	16% (8/50)	0%
NYHA Class I	54% (27/50)	100% (60/60)
NYHA Class II	34% (17/50)	0%
NYHA Class III	12% (6/50)	0%
Previous CABG	32% (16/50)	0%
Echocardiographic Ejection Fraction (%)	42 ± 12%	NA

Cardiac Magnetic Resonance Imaging

All imaging studies were carried out using a 1.5T scanner (Siemens, Erlangen, Germany). ECG-gated short-axis tagged MR images were acquired in multiple parallel planes starting at the level of the mitral valve and extending to the apex of the LV. Additionally, long-axis tagged images were acquired in four radially oriented planes separated by 45° and intersecting at the approximate center of the LV cavity. In each imaging plane, a spatial modulation of magnetization radiofrequency tissue-tagging preparation was applied, followed by a 2D balanced steady-state free precession cine image acquisition.^{15, 16} This process creates markers on the myocardium that deform with the tissue, which allows for quantification of wall motion. Typical imaging parameters were: repetition time 32.4ms, echo time 1.52ms, field of view 350×350mm, flip angle 20°, tag spacing 8mm and slice thickness 8mm.

Strain Analysis

Strain measurements were obtained for all study participants using an established and validated method developed at our institution.^17–20 Clinical validation of this methodology has demonstrated resistance to inter-observer variability.^{21, 22} Briefly, endocardial and epicardial wall boundaries were manually identified on each of the tagged images used in the study. Tag lines were tracked on each of the images semi-automatically utilizing an active contour approach.¹⁷

Three-dimensional displacements were computed for the systolic interval by tracking the intramyocardial intersection points of the three tag surfaces. The end-diastolic wall boundaries were used to create a finite element model of the LV. Using the anterior and posterior attachment points of the RV with the LV as landmarks, a standardized 18-element mesh was created for each model. A least squares fitting of the displacement data was used to obtain a continuous description of displacement for the entire model. Using the results of this fitting, circumferential strain, longitudinal strain, and the minimum principal strain angle were computed over a standardized grid of 15,300 points for the entire LV. The minimum principal strain angle is the angle between the plane formed by the circumferential and longitudinal axes and the axis of minimum principal strain.²³ In the LV, this angle quantifies the orientation of maximal contraction. The finite element software package StressCheck (ESRD, Inc., St. Louis, MO) was used to obtain the strain measurements for this study.

MRI-based Multi-parametric Strain Z-score Analysis

A total of 120 test subjects were subjected to full strain analysis following cardiac MRI with radiofrequency tissue tagging. In all individuals, 3D LV systolic strain components were calculated from tag surface intersection displacement data over a standardized grid of 15,300 LV points. A normal human strain database was constructed using strain measurements from 60 normal volunteers. At each LV grid point, and for each strain metric, measurements from the healthy volunteers were used to generate a normal mean and standard deviation at that particular point. This database of normal strain values allows for patient-specific, point-specific, and parameter-specific comparisons and z-score generation. A z-score is a statistical measurement that relates an individual score to a mean in a group of scores. The value of the z-score is directly proportional to its variation from the mean. A value of 0 means the score is the same as the mean and a value of 1 is 1 standard deviation above the mean.

A second group of 60 patients with known CAD underwent cardiac MRI, and z-scores for 3 strain parameters at each of the 15,300 LV grid-points were calculated by reference to the normal human strain database. The z-scores for 3 strain metrics at each point were then averaged to generate a single multi-parametric strain z-score value at each of the 15,300 points.

Blinded clinicians with no knowledge of the MRI-based multi-parametric strain results performed an extensive review of all available medical records for every CAD patient included in our study population. The goal in this clinical review was to identify myocardial regions that had sustained myocardial infarction by echocardiography, SPECT, or ECG, since this would guarantee a left ventricular region with significant contractile dysfunction by which the accuracy of our strain methodology could be tested. Using gold standard definitions (Q-waves on ECG, infarction on radionuclide SPECT, and/or akinesis or dyskinesis on echo), in each CAD patient, regional wall motion abnormalities were identified. Once this data was compiled, the z-scores were examined in each of the regions identified by the clinical review. If the average z-score was >1.5 for the affected region, then for that particular gold-standard imaging modality, the multiparametric strain analysis was considered accurate. Accurate identification of affected regions was assessed for each imaging modality and compared.

Results

Multi-parametric strain information was successfully obtained in all 60 CAD patients with post-processing display of “normalized” micro-regional contractile function on patient-specific LV geometry. The display of “normalized” micro-regional contractile function by intuitive z-score color contour mapping for all 60 CAD patients is represented in Figure 1. For ease of display, only a single view of each patient-specific, rotatable, semi-transparent 3D model is shown in this illustration. Further, in order to display the functionality and applicability of this methodology in a paradigm with which most clinicians are familiar, we scaled the color contouring based upon thresholds similar to those utilized in the clinical integration of patient blood laboratory test values. Specifically, <1 standard deviation from the normal mean was colored blue; 1 – 2 SD, yellow; and > 2 SD, red. For comparison, a normal volunteer with normal contractile function would appear diffusely blue (Figure 2). The average sub-regional z-score data from the 60 CAD patients is displayed in Table 2 and suggests that we will more accurately define clinically relevant thresholds with the accrual of more clinical data. This patient-specific normalized micro-regional contractile function is mapped (Figure 1) over patient-specific 3D LV geometry as defined by individual patient MR-based epicardial and endocardial wall boundaries.

Figure 1.

The display of “normalized” micro-regional contractile function by intuitive z-score color contour mapping (<1 standard deviation = blue; 1–2 = yellow; >2 = red) over patient-specific 3D LV geometry for all 60 CAD patients. The resolution of each MRI image is 72 pixels/inch while the 3D color contour maps are 300 pixels/inch.

Figure 2.

The display of “normalized” micro-regional contractile function by z-score color contour mapping in a normal, healthy volunteer (left) is contrasted against a patient (right) with severe regional contractile impairment secondary to myocardial infarction (<1 standard deviation = blue; 1–2 = yellow; >2 = red).

Table 2.

Regional average MPS z-scores from 60 patients

	PS	AS	A	AL	PL	P
Base	1.13±1.01	1.43±1.04	1.07±0.72	0.88±0.80	1.05±1.07	1.14±1.08
Mid-ventricle	1.39±1.35	1.34±1.22	1.20±0.95	1.07±0.87	1.45±1.15	1.42±1.07
Apex	0.97±0.96	0.87±0.88	0.79±0.75	0.77±0.81	0.97±0.90	0.96±0.80

^*PS = posteroseptal, AS = anteroseptal, A = apical, AL = anterolateral, PL = posterolateral, P = posterior

A blinded chart review of the 60 patients with known CAD revealed that 55% (33/60) of these patients had evidence of regional wall motion abnormalities as determined by current clinical gold standard diagnostic criteria (Q-waves on ECG, infarction on radionuclide SPECT, and/or akinesis or dyskinesis on echo). In this population, abnormalities were found in the anterior wall in 66% (22/33) of patients, anterolateral wall in 39% (13/33) of patients, anteroseptal in 58% (19/33), and posteroseptal, posterior, and posterolateral in 88% (29/33) of patients. In the majority of patients, the so-called “gold standard” criteria disagreed with each other regarding the presence or absence of wall motion abnormalities. In fact, consistency between all gold standard metrics was found in only 19% (6/31) of the CAD patients where at least two gold standard metrics were available.

Based upon our initial use of MRI-based multi-parametric strain analysis in a small test group of CAD patients with known regional nonviability,¹⁴ a sub-regional average z-score >1.5 was used as the threshold indicative of clinically significant contractile dysfunction. Of the patients in our complete study group who had wall motion abnormalities on EKG, SPECT, and Echo, the blinded MRI-based multi-parametric strain z-score localization of these contractile abnormalities using this z-score threshold was accurate in 89%, 97%, and 95%, respectively. An example of SPECT compared to the MRI-based, normalized stain analysis in a patient with CAD is represented in Figure 3.

Figure 3.

Comparison of a SPECT image (top) to MRI-based “normalized” micro-regional contractile function (bottom) in the same patient with CAD.

Discussion

MRI-based multi-parametric strain analysis provides a high-resolution quantification of micro-regional LV contractile function.^{14, 24} Patient-specific three-dimensional intramural LV point displacements obtained using MRI tissue tagging are used to quantify each of three strain parameters (circumferential magnitude, longitudinal magnitude, and minimum principal strain angle²³) at each of 15,300 LV grid-points. A z-score is calculated for each patient-specific raw strain parameter value at each grid point by its comparison to the normal mean and standard deviation (normal human strain database) for that specific strain parameter at that precise LV grid point. Thus to complete a single routine CAD patient study utilizing a standard 15,300 grid-point model requires 45,900 microprocessor-controlled database comparisons. This normalization process then allows multiple individually effective strain parameters (with variable data ranges) to be combined into more powerful composite multi-parametric strain indices.^{14, 25} These multi-parametric strain values, which represent a highly quantified normalization of point-specific LV contractile function, are then plotted over patient-specific LV geometry with color contour mapping to enhance intuitive interpretation.

To be clinically applicable, the capabilities of MRI-based multi-parametric strain analysis must be demonstrated in the actual patient populations and clinical settings in which it may have diagnostic or therapeutic value. We therefore report the results of its application in a study group of 60 patients with known ischemic coronary artery disease and, therefore, a high incidence of regional contractile dysfunction. This testing has confirmed not only its accuracy in detecting regional wall motion abnormalities, but also the applicability of this high-resolution, dynamic cardiac geometric data acquisition in the clinical setting frequented by patients with advanced atherosclerotic coronary occlusive disease. We have in fact already applied this same methodology effectively in multiple other patient populations including those with aortic valvular disease,^{12, 26} mitral valvular disease,²⁷ and non-ischemic dilated cardiomyopathy.⁹

The patient-specific LV contractile information supplied by this methodology is the product of complex post-processing of a substantial volume of quantified micro-regional myocardial point displacement data. The data resulting from the previously described 45,900 microprocessor-controlled database comparisons alone would overwhelm even the most ardent supporter of this analysis. In the past, it was precisely these scrolls of typewritten mathematical data that made this sort of complex micro-regional contractile analysis irrelevant to the clinician who had neither the time nor the expertise to decipher the information contained within.

For this complex micro-regional contractile information to be clinically relevant—even in the age of the electronic medical record (EMR)—it must therefore be presented in a 3D color contour visual format that allows easy and immediate clinical assimilation of “normalized” micro-regional LV contractile information over patient-specific LV geometry. It remains our further contention that the patient-specific regional LV contractile data that fuels cardiac clinical algorithms can be presented in a normalized z-score format similar to that which is already so familiar to physicians in the interpretation of patient-specific laboratory blood test values.

Traditionally, echocardiography most often supplies the dynamic LV geometrical input upon which clinical decisions in CAD patients are made. Observer-dependent visual differentiation into one of four contractile classifications (normal, hypokinetic, akinetic, dyskinetic) supplies only a gross normalization over six arbitrarily assigned geometric LV regions. Thus, unlike the quantified normalization commonly applied to laboratory blood testing, regional contractile function is qualitative, not normalized, and subject to inter-observer variation. Interpreter and temporal inconsistency is predictable since the assessment of regional contractile function is most often dependent upon the visual detection of 1 mm LV wall thickness changes in 8 mm thick walls that are moving in and out of the echocardiographic imaging plane with every breath and every left ventricular contraction.⁷ In contrast, since our methodology relies heavily upon a quantitative comparison of patient-specific strain to a fixed standard (normal human strain database), there is very little room for temporal inconsistency.^{21, 22}

Several other alternatives to standard 2D echocardiography have been evaluated. Speckle-tracking echocardiography, which is based upon the spatial displacement of “speckles” generated by the interaction between ultrasound beams and myocardial fibers,²⁸ was designed to provide further information on global and regional contractile function. This modality tracks these speckles in 3 dimensions during the cardiac cycle allowing for more detailed information on myocardial contractility. Despite improvement in quantifying LV dysfunction, this imaging modality is still limited by limited imaging windows, low resolution image quality, and significant user-dependent inter-observer variability.²⁹ Three-dimensional echocardiography provides higher-resolution and more detailed images than conventional echocardiography, but is still limited by subjective interpretation, variable acoustic windows, and inter-observer variability.³⁰ In contrast, MRI-based multi-parametric strain z-score “normalization” is based upon fully 3-dimensional, high-resolution imaging evaluated in a highly quantified, non-subjective manner.

One of the obvious and long-awaited advantages of the EMR is the ability to present large volumes of complex mathematical patient data, such as that obtained from MRI-based multi-parametric strain z-score analysis, in an easily interpreted, intuitive visual format. The addition of quantified normalization to a visually intuitive format maximizes the efficient assimilation of this information, even by physicians without subspecialty training in cardiology, radiology, or cardiothoracic surgery. In fact, since the primary determination of “normal versus abnormal” is already automatically made by non-biased, repeatedly consistent computational methods and then presented in an intuitive format, the results can be easily interpreted not only by cardiac imaging subspecialists, but also by primary care physicians. In fact, the presentation of regional contractile data in an already normalized format allows micro-regions with contractile function falling outside of 1 or 2 standard deviations to be clearly delineated (yellow and red coloration in our images) to any interested party, including the patients themselves.

The presentation of complex medical information in such a format represents the zenith of the EMR, allowing even patients an improved understanding of their disease process. This is the clear future of medicine as patients are being asked to participate as informed consumers and assume more responsibility in their medical decision-making. The EMR provides the ideal mechanism to present and explain complex medical information—in this case, patient-specific micro-regional LV contractile information—in a format that can be readily understood by the patients themselves. As the EMR matures, this visual presentation of micro-regional contractile information will take full advantage of the obvious benefits the EMR offers. A laptop computer or touchpad is all that is necessary to spin a 3D LV model of the patient’s very own normalized contractile function projected upon an accurate model of their LV geometry. There is considerable potential for immediate impact in the clinical application of this methodology in directing therapeutic intervention in those patients with a regionally variable degree of ischemic contractile injury.

There are several important limitations of the proposed imaging modality. In particular, MRI can be expensive and post-processing of myocardial strain can be time-consuming. That being said, with advances in MR image acquisition technology and more automated data analysis, our current scan times are less than 30 minutes with associated data analysis being completed in less than 20 minutes. Another limitation of this methodology, which is shared with other measurements of regional myocardial contractility, is its limitation to quantifying only myocardial strain. Ventricular contractility takes into account strain (how well the wall is moving), stress (how much force the wall is exerting), and time (how quickly the wall is moving). Ideally, stress and strain would be concurrently quantified. Unfortunately, current methods for accurately quantifying regional myocardial stress in the clinical setting are severely limited. Lastly, the average age of our normal volunteers is significantly less than the CAD study group. Nonetheless, investigations into age-related differences in strain suggest that significant differences occur only at the extreme ends of normal population age range.³¹

This investigation has further confirmed the limited ability of current clinical gold standard contractile metrics to consistently and quantitatively localize regional LV contractile abnormalities. As a reasonable clinical alternative, MRI-based multi-parametric strain 3D z-score “normalization” of micro-regional contractile function by comparison of patient-specific raw strain values to a normal human strain database can accurately quantify, normalize, and localize LV contractile dysfunction on a micro-regional basis. With the widespread implementation of the EMR, this intuitive 3D color contour mapping of normalized regional contractile function on patient-specific 3D LV geometry has considerable potential to replace the current non-quantitative, type-written reports upon which critical clinical decisions are often based.