Author + information
- Published online August 15, 2018.
- Katherine C. Wu, MD∗ ()
- ↵∗Address for correspondence:
Dr. Katherine C. Wu, Johns Hopkins Medical Institutions, Division of Cardiology, Blalock 536, 600 North Wolfe Street, Baltimore, Maryland 21287.
Advances in medical therapy for heart failure have significantly reduced mortality, particularly for sudden arrhythmic cardiac death (SCD) (1). A recent meta-analysis of primary prevention implantable cardioverter-defibrillator (ICD) trials reported an overall annualized mortality rate of 5.4% in patients with nonischemic cardiomyopathy (NICM) (2), and a post hoc analysis of the DANISH (Danish Study to Assess the Efficacy of Implantable Cardioverter Defibrillators [ICD] in Patients With Non-Ischemic Systolic Heart Failure on Mortality) trial suggested a relatively low incidence rate of SCD of 1.6 to 1.8 per 100 patient-years (3). Hence, the relative value of primary prevention ICDs in NICM has been questioned in light of the low event rates and projected large number needed to treat. This underscores the critical need for enhanced risk stratification methods.
The pathophysiology of ventricular arrhythmic susceptibility in NICM involves replacement fibrosis that is often patchy, intermingled with normal myocytes, and distributed in irregular patterns (4). Cardiac magnetic resonance imaging (CMR) with late gadolinium enhancement (LGE) detects regions of focal and patchy fibrosis in NICM whose presence and extent have been correlated with arrhythmic outcomes (4). From a meta-analysis of 34 studies including more than 4,500 patients, the presence of fibrosis detected by CMR with LGE compared with its absence was associated with a significantly increased risk for ventricular arrhythmic events (odds ratio: 4.52; 95% confidence interval: 3.41 to 5.99) (5). However, published studies also show that LGE on CMR predicts heart failure events and all-cause mortality, competing risks that are not treatable by ICDs and that warrant other management strategies. Hence, CMR LGE alone as a predictor may not be sufficiently sensitive, and improved imaging metrics are needed.
In this issue of iJACC, Muthalaly et al. (6) describe the novel application to CMR with LGE of an imaging processing metric termed image entropy. Entropy of a grayscale image, such as that depicted by magnetic resonance imaging, is a scalar statistical measure of randomness that can be used to characterize the distribution and dispersion of signal intensity amplitudes within an image. An example from photo editing of a low-entropy image would be one that encompasses a large swath of black sky with little contrast and pixels with the same or similar signal intensity values. High-entropy images would include numerous stars of varying brightness scattered within the black sky. The CMR LGE method for scar assessment currently depends on detecting large relative signal intensity differences between hyperenhanced, focally scarred myocardium and normal myocardium with low signal intensity. It cannot distinguish between homogeneously dark, normal myocardium with a narrow range of low signal intensity values and myocardium with grayish, intermediate-signal intensity values that may span a wide distribution and reflect diffusely distributed fibrosis. Diffuse scarring is often seen in NICM and can be a potential substrate for arrhythmogenesis. Hence, an index of the complexity of the pixel intensity distributions within a CMR image may provide important incremental information for SCD risk stratification.
The Muthalaly et al. (6) study comprised a small registry-based observational cohort of 130 predominantly male (87%) patients with dilated NICM (mean left ventricular [LV] ejection fraction 29.4 ± 13.5%) undergoing CMR with LGE prior to insertion of primary prevention ICDs and followed for a median of 3.2 years. There were high rates of contemporary guideline-directed pharmacological therapy, and 33% also received cardiac resynchronization therapy devices. The primary endpoint (n = 18) comprised arrhythmic events, defined as appropriate ICD shocks (n = 11) as well as antitachycardia pacing, sustained ventricular arrhythmias, or SCD not resuscitated by the device (n = 1). Overall incidence of the arrhythmic endpoint was 4 events per 100 person-years. The secondary outcome (n = 17) included end-stage heart failure events (cardiac death, heart transplantation, or ventricular assist device placement). CMR LGE was present in the majority (57%) with small average scar sizes (median 2.0 g; interquartile range: 5.96 g), as is typically seen in such cohorts. Scar size was defined as dense scar quantified by the full width half maximum method.
The extent of LV entropy was only weakly correlated with the presence and extent of scar (r = 0.20, p < 0.006 for scar size), suggesting that this metric could provide unique information about the CMR signal intensity characteristics. The investigators did indeed find that higher entropy values were independently associated with arrhythmic outcomes and were incrementally predictive in a multivariate model that included scar mass. Moreover, higher LV entropy values appeared to be specific for ventricular arrhythmia and were not associated with end-stage heart failure outcomes or all-cause mortality. Because the LV entropy metric requires only planimetry of the LV endocardium and epicardium, it requires much less user input than scar assessment and quantification and is more automated. Hence, intraobserver and interobserver variability are minimized, as is image analysis time, which are all strengths.
Although these preliminary results are encouraging, a number of study limitations require further investigation. Foremost, the current approach is 1-dimensional and “calculates the entropy by aggregating all slices of ventricular myocardium that are demarcated by the observer and drawing the entropy value from that distribution.” In so doing, the 2- and 3-dimensional spatial locations and patterns of the variable signal intensity values are not preserved. Thus, the calculated entropy values do not adequately reflect the spatial complexity of fibrosis architecture, which is recognized as an essential component of propensity for ventricular arrhythmogenesis and electrophysiologic re-entry (4).
Other major limitations include the low event rate, with only 18 total arrhythmic events, limiting the robustness of the multivariate analyses, although the investigators incorporated propensity scoring to reduce overfitting. Furthermore, inclusion of antitachycardia pacing therapy as an arrhythmic outcome may overestimate the true event rate because many such events may have been self-terminating without the ICD. The incremental value of including entropy in a prediction model may also be overemphasized by the use of the net reclassification index, which can be misleading (7). Other statistical measures of the extent of the improvement in prediction performance warrant evaluation (7). Scanner field strengths and vendors were mixed in this study (GE 1.5-T and Siemens 3.0-T machines were used), with a quantifiable 2.00 greater entropy value seen in patients scanned with the 3.0-T compared with the 1.5-T scanner. As acknowledged by the investigators, the impact on LV entropy values of various imaging acquisition factors needs to be better defined, as does the range of values in normal individuals.
Improvements in the prevention and management of coronary artery disease have resulted in a decline in ischemic cardiomyopathy incidence and an increased proportion of patients with NICM who are candidates for ICDs (8). With intensification of guideline-directed medical therapy and the success of cardiac resynchronization therapy in stimulating LV reverse modeling and improved LV ejection fraction, the incidence of arrhythmic outcomes is declining but still tangible and preventable by ICDs, however resource intensive such devices may be. This behooves us to identify robust risk predictors that can differentiate the relative risk for SCD against competing causes of death (heart failure and noncardiac) in these patients with NICM. Although CMR has provided a much-needed window to the LV structural and myocardial abnormalities that predispose to arrhythmias, it remains underused clinically for a number of reasons, including labor-intensive image analysis. Novel imaging metrics that require minimal user input and analysis time, have high reproducibility, and are specific to arrhythmic outcomes are particularly valuable. If validated, CMR entropy may fulfill these needs. As Carl Jung stated, “In all chaos there is a cosmos, in all disorder a secret order.” Perhaps CMR entropy can bring order to the disorder of SCD risk prediction.
↵∗ Editorials published in JACC: Cardiovascular Imaging reflect the views of the authors and do not necessarily represent the views of iJACC or the American College of Cardiology.
Dr. Wu is supported by National Institutes of Health grant R01HL132181 (“Cardiac MR to Improve Clinical Risk Prediction in Defibrillator Patients”) and was the principal investigator for National Institutes of Health grant R01HL103812 (“Left Ventricular [LV] Structural Predictors of Sudden Cardiac Death [SCD]”).
- 2018 American College of Cardiology Foundation
- Shen L.,
- Jhund P.S.,
- Petrie M.C.,
- et al.
- Shun-Shin M.J.,
- Zheng S.L.,
- Cole G.D.,
- Howard J.P.,
- Whinnett Z.I.,
- Francis D.P.
- Elming M.B.,
- Nielsen J.C.,
- Haarbo J.,
- et al.
- Wu K.C.
- ↵Becker MAJ, Cornel JH, van de Ven PM, van Rossum AC, Allaart CP, Germans T. The prognostic value of late gadolinium-enhanced cardiac magnetic resonance imaging in nonischemic dilated cardiomyopathy: a review and meta-analysis. J Am Coll Cardiol Img 2018 Apr 18 [E-pub ahead of print]
- Muthalaly R.G.,
- Kwong R.Y.,
- John R.M.,
- et al.