Magnetic Resonance Imaging of Arrhythmogenic Right Ventricular DysplasiaSensitivity, Specificity, and Observer Variability of Fat Detection Versus Functional Analysis of the Right Ventricle
Author + information
- Harikrishna Tandri, MD⁎,
- Ernesto Castillo, MD†,
- Victor A. Ferrari, MD‡,
- Khurram Nasir, MD⁎,
- Darshan Dalal, MD⁎,
- Chandra Bomma, MD⁎,
- Hugh Calkins, MD⁎ and
- David A. Bluemke, MD, PhD†,⁎ (dbluemke{at}jhmi.edu)
- ↵⁎Reprint requests and correspondence:
Dr. David A. Bluemke, MRI, Room 143 (Nelson Basement), Johns Hopkins Hospital, 600, N. Wolfe Street, Baltimore, Maryland 21287.
Abstract
Objectives The purpose of this study was to determine interobserver agreement for interpretation of magnetic resonance imaging (MRI) examinations of arrhythmogenic right ventricular dysplasia (ARVD) and to determine sensitivity and specificity of fat detection versus functional parameters measured by MRI.
Background The interobserver variability of MRI and the relative importance of different MRI parameters (fat detection, regional and global right ventricular [RV] function) for ARVD diagnosis is unknown.
Methods Two experienced observers blinded to the clinical history independently analyzed MRI datasets obtained from 40 patients evaluated for ARVD. Twenty normal subjects underwent MRI and served as control subjects. The MRI scans were performed according to a standard protocol on a 1.5-T scanner. The observers reported on fat infiltration, global and regional RV function, myocardial thinning, and chamber dilatation qualitatively. The RV volumes were measured on the cine sequences.
Results Interobserver kappa scores for fat infiltration, global and regional RV function, wall thinning, and RV outflow dilatation were 0.74, 0.94, 0.89, 0.93, and 0.93, respectively. Correlation coefficients between observers for RV end-diastolic volume, end-systolic volume, and ejection fraction were 0.93, 0.94, and 0.95, respectively (p < 0.001). Fifteen patients were diagnosed with ARVD using Task Force criteria. Sensitivity of fat infiltration, RV enlargement, and regional RV dysfunction for diagnosing ARVD was 84%, 68%, and 78%, and specificity was 79%, 96%, and 94%, respectively.
Conclusions Qualitative assessment of RV structure and function is highly reproducible for experienced observers. Among the qualitative parameters, fat infiltration is less reproducible and lacks specificity compared with RV kinetic abnormalities.
Arrhythmogenic right ventricular dysplasia (ARVD) is a cardiomyopathy of unknown etiology characterized by fibrofatty replacement of the right ventricular myocardium (1–3). The diagnosis of ARVD is based on the presence of major and minor criteria encompassing structural, histologic, electrocardiographic, arrhythmic, and genetic factors proposed by ARVD Task Force in 1994 (4). Demonstration of abnormalities in right ventricular structure and function constitutes one of the diagnostic criteria for ARVD. Magnetic resonance imaging (MRI) allows multiplanar evaluation of the right ventricle (RV), enabling accurate morphologic and functional assessment without any geometric assumptions (5). Intramyocardial fat accumulation is a pathologic hallmark of ARVD, and MRI has excellent tissue characterization capability, particularly for fatty tissue (6–8). The ability to provide tissue characterization as well as to visualize RV function makes MRI suitable for follow-up of patients and to study the progression of disease.
Despite its potential, MRI can be very problematic because the performance of MRI scans requires training and expertise, and image interpretation requires considerable experience (9,10). Bluemke et al. (11) reported interobserver reliability of MRI findings in ARVD among expert readers using static images (spin-echo images) performed on conventional scanners. They concluded that identification of fat signal is less reliable and reproducible compared with morphologic abnormalities of the RV. An important limitation of their study was that the images were acquired using nonuniform protocols, and the image quality was considered good in <40% of the examinations. We hypothesized that the reproducibility could be significantly improved by optimizing image quality and by using a uniform imaging protocol. The aim of the present study was to define the level of interobserver reproducibility in reporting of MRI in the diagnosis of ARVD and to determine the sensitivity and specificity of various MRI parameters used in the diagnosis of ARVD.
Methods
The study population included 40 patients who were evaluated for possible ARVD at the Johns Hopkins ARVD center. Each of these 40 patients presented with left bundle branch block morphology ventricular arrhythmias. All of these patients underwent initial noninvasive testing including an electrocardiogram, signal-averaged electrocardiogram, echocardiogram, exercise testing, and MRI. Invasive testing was performed at the discretion of the cardiologist and included electrophysiologic study, RV angiography, and endomyocardial biopsy. Diagnosis of ARVD was made excluding the results of MRI. Twenty normal healthy volunteers who underwent MRI according to the same protocol served as control subjects.
MRI protocol
All the datasets were obtained on a 1.5-T scanner (CV/i, General Electric Medical Systems, Waukesha, Wisconsin) at the Johns Hopkins Hospital and included both fast spin-echo (FSE) and gradient-echo sequences. Fat-suppressed and nonfat-suppressed (FSE) sequences were acquired in the axial and short-axis planes with breath hold double-inversion recovery blood suppression pulses. The repetition time was 1 or 2 R-R intervals, and the time to excitation (TE) was 10 ms. The slice thickness was 5 mm and slice gap 5 mm. The matrix and field of view were 256 × 256 and 24 cm, respectively. Gradient echo sequences were acquired in the axial and short-axis planes using breath-hold steady-state free precession (SSFP) imaging. The flip angle was 40°, and TE was set to minimum. For SSFP imaging, the slice thickness was 8 mm with a slice gap of 2 mm. The matrix and field of view were 256 × 160 and 36 cm, respectively. A phased array cardiac coil was used for all the studies.
The MRI datasets were made anonymous and were identifiable only by a unique ID number. The MRI datasets were transfered to an Advantage Windows workstation (General Electric Medical Systems) for analysis. Quantitative analysis was performed using the software program MASS (Medis, Leiden, the Netherlands). This software was used by the readers to view images using standardized window width and level settings. The same software was also used for measurement of right ventricular and right atrial chamber size. Ventricular volume measurements were performed on short-axis gradient-echo images. The first image after the R-wave trigger represented the end-diastolic image. End-systolic image was defined visually as the one with the smallest ventricular cavity size.
Two experienced cardiac MRI readers were asked to assess images for: 1) fat infiltration (intramyocardial high T1 signal); 2) abnormal wall thinning (<2 mm RV wall thickness); 3) RV outflow dilation; 4) RV enlargement; and 5) RV regional wall motion abnormality. The readers had 11 years (Johns Hopkins Hospital) and 11 years (University of Pennsylvania) of experience in evaluating cardiac MRI. Both of the centers (Johns Hopkins Hospital and University of Pennsylvania) perform around 100 exams per year particularly for the diagnosis of ARVD. For qualitative reporting of intramyocardial fat, the RV was divided into 3 areas on the FSE axial images (RV free wall, RV apex, and RV outflow tract) (Fig. 1).The presence or absence of fat infiltration was reported in each of these individual areas. The RV wall thinning was also evaluated on the FSE images. The diagnosis of ARVD was made according to the ARVD Task Force recommendations if the patient had either 2 major criteria, 1 major and 2 minor criteria, or 4 minor criteria (4). Intramyocardial fat was defined as a partial or transmural hyperintense signal within the RV wall on T1-weighted black blood image. Wall thinning was defined as focal and abrupt reduction in wall thickness to <2 mm with adjacent areas of normal wall thickness. Both RV outflow tract (RVOT) and left ventricular outflow tract were assessed on axial black blood images. The RVOT was evaluated at the level of the aortic valve, and aortic outflow was assessed 1 cm below the aortic valve. The RVOT was considered to be enlarged if it was larger that the aortic outflow. An example of an enlarged RVOT is shown in Figure 2.Functional analyses of the RV were performed in a binary fashion. The RV was considered to be enlarged if it was equal to or larger than the left ventricle in diastole, on the apical 4-chamber diastolic cine images. The RV global and regional functions were qualitatively evaluated on the cine images. The RV enlargement and regional function were assessed on the gradient-echo images. Observers were blinded to all clinical histories.
(A)Axial black blood image of the right ventricle. The right ventricle was divided into free wall (black arrow)and apex (white arrow)by a line drawn in the midanterior wall. (B)Axial black blood image of the outflow tract (white arrow).
Axial black blood image at the level of the aortic valve. The image on the leftshows the right ventricular outflow tract (RVOT) in a healthy volunteer, which is comparable to the size of the aorta. The image on the rightshows a grossly enlarged RVOT from a patient with arrhythmogenic right ventricular dysplasia.
Statistical analysis
Data were analyzed using STATA version 7 (Stata Corp., College Station, Texas). Agreement between observers for qualitative variables was characterized by unweighted kappa values. Kappa scores between 0.41 and 0.6 indicate moderate, 0.61 to 0.8 good, and above 0.8 very good agreement. Significance testing was done using a z-score with a p value of ≤0.05 considered statistically significant. Sensitivity and specificity of MRI variables for both readers were generated, and average sensitivity and specificity for diagnosis of ARVD were determined. Inter-reader agreement for a quantitative variable was assessed using correlation coefficient and limits of agreement using Bland Altman plots.
Results
The study population included MRI datasets from 40 patients (19 men and 21 women) and 20 control subjects (9 men and 11 women). Baseline characteristics of the study population are shown in Table 1.The image quality was considered adequate for reporting in all of the 40 patients for all the series. The final diagnosis of ARVD was made according to the Task Force criteria (Table 2).For the purposes of this study, MRI results were not used to make the final diagnosis of ARVD being present or absent.
Baseline Characteristics of Study and Control Subjects
Task Force Criteria for the Diagnosis of Arrhythmogenic Right Ventricular Dysplasia/Cardiomyopathy
Of the 40 patients with ventricular arrhythmias, 15 had a final diagnosis of ARVD. The characteristics of patients with ARVD are shown in Table 3.Twenty-five patients were diagnosed with idiopathic RVOT ventricular tachycardia. The diagnosis was based on lack of abnormalities on surface electrocardiogram (ECG), signal-averaged ECG, and absence of structural heart disease on echocardiography (Table 1). Of these, 21 patients underwent electrophysiologic testing. Sustained monomorphic ventricular tachycardia (VT) was induced in 2 patients and nonsustained VT induced in 6 patients. The VT was induced by ventricular overdrive pacing within a range of cycle lengths consistent with a triggered or automatic mechanism. The remaining 13 patients had only induced frequent premature ventricular contractions (PVC). Isoproterenol was required for VT/PVC induction in 12 patients. The morphology of the VT was left bundle inferior axis consistent with an origin in the RVOT in each of these patients. Seven of the 8 patients with inducible VT underwent successful catheter ablation with resolution of the arrhythmia. The remainder elected for medical therapy with beta-blockers.
Clinical Characteristics of Patients With Arrhythmogenic Right Ventricular Dysplasia
Qualitative parameters
The results of the qualitative analysis are summarized in Table 4.The incidence of intramyocardial hyperintense MRI signal intensity indicative of fat infiltration of RV was 4%, 34%, and 84% for control subjects, idiopathic RVOT VT, and ARVD, respectively (control subjects vs. idiopathic VT: p = NS; idiopathic VT/control subjects vs. ARVD: p < 0.01). The inter-reader agreement for fat infiltration was good (88%) with a kappa score of 0.74. Wall thinning was reported in 1 patient with ARVD, and inter-reader agreement was 100%. Right ventricular enlargement was reported in 4%, 0%, and 68% of control subjects, idiopathic RVOT VT, and ARVD patients, respectively. The agreement was 95% with a kappa score of 0.84. Regional wall motion abnormalities were reported in 4%, 0%, and 78% of control subjects, idiopathic RVOT VT, and ARVD patients, respectively, with excellent inter-reader agreement (kappa = 0.90). Right ventricular outflow tract enlargement was reported in 4%, 0%, and 66% of control subjects, idiopathic RVOT VT, and ARVD patients, respectively. The agreement was 93% with a kappa score of 0.78.
Incidence and Interobserver Agreement for the Qualitative Magnetic Resonance Variables
Quantitative parameters
Right ventricular end-diastolic volume (RVEDV) and RV ejection fraction (RVEF) was significantly different in ARVD compared with idiopathic VT and control subjects (Fig. 3).There was an excellent correlation between the observers in quantitatively reporting the ventricular volumes and ejection fraction, reflected by correlation coefficients of 0.93, 0.94, and 0.95 for RVEDV, RV end-systolic volume, and RVEF, respectively (p < 0.001 for all). The Bland-Altman plots showing the limits of agreement are shown in Figure 4.
(A)Right ventricular end-diastolic volume was significantly greater in patients with arrhythmogenic right ventricular dysplasia (ARVD) compared with idiopathic ventricular tachycardia (VT) and control subjects. No differences were noted in the left ventricular end-diastolic volumes. (B)Right ventricular ejection fraction was significantly lower in patients with ARVD compared with idiopathic VT and control subjects. No differences were noted in the left ventricular ejection fraction.
Bland-Altman analysis showing excellent agreement between the two observers for right ventricular end-diastolic volume (RVEDV) (top, mean difference 7.9 ± 3.1 ml; upper and lower limits of agreement −26 ml and 42 ml, respectively) and right ventricular (RV) ejection fraction (bottom, mean difference −0.02 ± 0.04; upper and lower limits of agreement −0.17 and 0.1, respectively). Limits of agreement are displayed as ±2 SD.
Sensitivity and specificity
Fat infiltration of the RV had a sensitivity of 84% and specificity of 79% with a positive predictive value of 80%. Regional wall motion abnormality of the RV had a sensitivity of 75% and specificity of 97% with a positive predictive value of 90%. Among the variables assessed, fat infiltration was the most sensitive and least specific in diagnosis of ARVD.
Discussion
The growing use of MRI in the diagnosis of ARVD is based on its excellent demonstration of RV morphology and function. Magnetic resonance imaging is uniquely suited to provide tissue characterization by demonstrating fat infiltration, which is the pathologic hallmark of ARVD. Recent studies have highlighted the role of MRI in detecting fibrosis in the RV myocardium in patients with ARVD (12). The increasing use of MRI in ARVD is based on the premise that the MRI findings are reliable and reproducible. Newer and faster imaging techniques, breath-hold imaging, and use of surface coils to improve signal-to-noise ratio lead to high-quality images of the RV, which are believed to allow reproducible and reliable interpretation. A recent ex vivo study demonstrated potential limitations of certain fast imaging techniques for ARVD (13), demonstrating the need for human in vivo determination of the reliability of MRI for this condition.
The present study has several important findings. First, the inter-reader reproducibility for qualitative assessment of MRI variables in the diagnosis of ARVD is excellent among experienced observers. The kappa value for intramyocardial fat infiltration was 0.74, which is significantly higher than that reported by Bluemke et al. (11). That study was multicenter and used non-breath hold conventional spin imaging; a method not currently used at most sites owing to low image quality with resulting frequent motion artifacts. Our findings are similar to the single-center results reported by Abbara et al. (14), who used conventional spin-echo images with fat suppression and demonstrated good agreement for intramyocardial fat detection. However, the utility of fat-suppression was not separately assessed in our study. Similarly, inter-reader agreement was excellent in our study for other qualitative variables, such as RV dilation, regional wall motion abnormalities, and RV outflow dilation.
We also found that intramyocardial fat infiltration is the most sensitive (84%) and the least specific (79%) for diagnosis of ARVD. Intramyocardial fat was also the least reproducible parameter among all the qualitative variables (kappa = 0.74). There are several possible reasons for this. The normal presence of epicardial and pericardial fat makes identification of true intramyocardial fat difficult. Some areas, such as the subtricuspid region, are not easily distinguished from the atrioventricular sulcus, which is rich in fat (15). Furthermore, intramyocardial fat is frequently seen in normal hearts in the anteroapical regions (16). The RV free wall is only 3 to 5 mm thick, and the spatial resolution is often unsatisfactory to reliably comment on fat infiltration, particularly in patients with arrhythmia and resulting motion artifacts on MRI. To differentiate fatty infiltration from the normal epicardial fat requires considerable experience in MRI interpretation and may be impossible if motion artifacts due to cardiac arrhythmia are present.
Finally, we found that experienced observers had excellent agreement for quantitative assessment of the RV, thus validating the use of quantitative MRI in a clinical setting. Routine quantification of the RV may further decrease the interobserver variability in reporting clinical studies. The main prerequisite for quantitative evaluation is availability of age- and gender-corrected normative data (17). Although data are available on RV volumes and function, data on RV dimensions are lacking. The normative data from echocardiography is not readily adapted to MR imaging, because of differences in imaging planes (18).
Study limitations
Although we have shown that qualitative assessment of regional RV kinetic abnormalities has a high positive predictive value for diagnosing ARVD, a method for reliable quantitative assessment of regional RV function by MRI has not been assessed. One approach to this problem is myocardial tissue tagging (19). Fayad et al. (20) have previously described this approach in the RV. However, tissue tagging has not been found to be easily applied owing to the very thin RV free wall and resultant poor signal-to-noise ratio.
Clinical implications
Interpretation of MR examinations for ARVD is inherently qualitative. Typically, MRI readers judge the presence or absence of fatty infiltration, chamber dilatation, and RV dysfunction. Previously, little has been known regarding the relative value of any of these parameters in contributing to the final diagnosis or exclusion of ARVD. Our results show that identification of fat within the myocardium by MRI may not be specific and is in fact the least reproducible MRI parameter. Reliance on fat detected by MRI for ARVD diagnosis may lead to false positive MRI scans, which has serious consequences both for patients and for their families (21). Regional wall motion abnormalities limited to the RV are highly specific to the disease. The possibility of ARVD should be considered when fat infiltration occurs in conjunction with wall motion abnormalities and the diagnosis should be made using the Task Force criteria. Although intramyocardial fat detection on MRI is a well-accepted alternative for endomyocardial biopsy, it is still not a Task Force criterion for ARVD diagnosis. Caution should be exercised in interpreting the results, especially in the absence of wall motion abnormalities. It is hoped that the U.S. ARVD study, a National Institutes of Health-sponsored prospective multicenter study that is currently underway, will further clarify the role of MRI in ARVD. The study aims to enroll 100 newly diagnosed patients with ARVD and 500 first-degree relatives. All the subjects in the U.S. ARVD study are evaluated using a standardized MRI protocol similar to the protocol described in the present report.
In conclusion, the present study shows that interobserver agreement for qualitative and quantitative evaluation of MRI for ARVD is excellent among experienced observers. Among qualitative MRI variables, intramyocardial fat on MRI is less reliable compared with wall motion abnormalities. The RV volumes are significantly different in ARVD, and quantitative RV analyses may be useful in both the diagnosis and the follow-up of patients with ARVD.
Footnotes
The Johns Hopkins ARVD Program is supported by the Bogle Foundation, the Campanella family, the Wilmerding Endowment, and National Institutes of Health grant 1 UO1 HL65594-01A1. Dr. Ernesto Castillo is a research fellow in Radiology and is supported by a research grant from the Fundaciön Ramön Areces, Madrid, Spain.
- Abbreviations and Acronyms
- ARVD
- arrhythmogenic right ventricular dysplasia
- FSE
- fast spin-echo
- MRI
- magnetic resonance imaging
- RAD
- right atrial diameter
- RV
- right ventricle/ventricular
- RVEDD
- right ventricular end-diastolic diameter
- RVESD
- right ventricular end-systolic diameter
- RVOT
- right ventricular outflow tract
- SSFP
- steady-state free precession
- VT
- ventricular tachycardia
- Received January 27, 2006.
- Revision received July 6, 2006.
- Accepted July 12, 2006.
- American College of Cardiology Foundation
