Non-contrast cardiovascular magnetic resonance detection of myocardial fibrosis in Duchenne muscular dystrophy

Duchenne muscular dystrophy (DMD) leads to progressive skeletal and cardiac myopathy and affects up to 1 in 4700 live male births [1]. Disease progression is variable, but all boys will have cardiovascular manifestations by 18 years of age [2]. There is no cure for DMD. While cardiovascular therapeutic options are limited, early therapy with angiotensin converting enzyme inhibitors and/or aldosterone inhibitors has demonstrated delayed mortality [3, 4]. Some boys with DMD have earlier onset of cardiovascular disease or more rapid progression, requiring either earlier initiation or more rapid intensification of therapy, resulting in early heart failure and death.

Cardiovascular imaging is necessary to detect both early onset and rapid progression of cardiomyopathy. While transthoracic echocardiography has been the mainstay of cardiovascular evaluation in DMD, the field has increasingly shifted to cardiovascular magnetic resonance (CMR) [5]. This is partially because CMR provides superior image quality, a significant advantage in a patient population with challenging acoustic windows [6]. CMR also allows for assessment of fibrosis with late gadolinium enhancement (LGE), which can precede left ventricular (LV) dysfunction and which predicts morbidity and mortality in multiple disease processes, including DMD [7‐11]. Clinical decision making is often based on LGE in boys with DMD. Tandon and colleagues have shown that LV ejection fraction (LVEF) declines once LGE is present but not before [11]. Thus, LGE can be an early sign of impending LV dysfunction. In addition, given the increased mortality associated with LGE, a more rapid progression of LGE in the setting of stable or mild changes in LVEF is treated with more aggressive medical therapy. The benefit of LGE is the information it provides in addition to LVEF, especially in a patient with stable ventricular function. However, LGE requires administration of gadolinium contrast agents. While these agents have an excellent safety profile in patients without renal disease, recent reports suggest that renal disease may be under-recognized in older DMD patients [12], coupled with growing concern over possible long-term brain deposition after repeated scans has raised concerns about long-term safety in children and young adults [8, 13].

Recently, mapping of the longitudinal time constant, T1, has gained popularity for myocardial tissue characterization. Native T1 maps can be performed without contrast administration, while extracellular volume (ECV) maps require a combination of native T1 and post-contrast T1 maps as well as a recent hematocrit [14‐16]. Native T1 times are elevated in the presence of myocardial fibrosis, and correlates with LGE in other disease processes [17‐20]. We previously demonstrated that native T1 is elevated in DMD patients compared with healthy subjects [21]. Myocardial strain has also been used to determine the presence or absence of LGE [22]. We hypothesized that native T1 and/or circumferential (ε_cc) and longitudinal strain (ε_ls) can be used as surrogates for LGE in DMD subjects, allowing for less frequent gadolinium contrast administration.

This study presents the largest cohort of CMRs in DMD patients evaluating ε_cc, ε_ls, and native T1. The primary findings of this study are: (1) modeling using pre-contrast sequences modestly predicts presence and severity of LGE for individual CMRs; (2) strain appears more predictive than native T1 in most models; (3) strain, but not native T1 or initial global LGE, predicts change in severity of LGE estimated by GSS (for feature tracking) and FWHM (for myocardial tagging); (4) Non-contrast sequences are not a substitute for LGE imaging in DMD. While strain was statistically predictive of most outcome variables, the performance of strain was only modest, and native T1 was not predictive in most models. The performance of most non-contrast models evaluated did not meet thresholds for routine clinical use, emphasizing the importance of LGE in the DMD population.

Native T1 mapping has shown promise as a tool for detecting myocardial fibrosis in dilated cardiomyopathy as assessed by LGE and histologically [33]. Given the regionality of fibrosis observed in DMD cardiomyopathy, the utility of this sequence may be in selected segments rather than as a global predictor. Apical slices are susceptible to errors from partial volume averaging and also have less fibrosis than basal and mid-LV slices in DMD boys. Global values were thus calculated excluding apical slices. Recently, Olivieri and colleagues found that native T1 and ECV in the lateral wall were increased in DMD boys with LGE compared to those without LGE, though there was no significant difference in the septum [34]. They also found that the saturation recovery single shot acquisition (SASHA) technique was better at discriminating between disease states (control, DMD no LGE, and DMD with LGE) than MOLLI; however, modeling suggested that MOLLI could also distinguish between DMD boys with and without LGE. Studies suggest that native T1 values of myocardium and, to a lesser extent blood pool, remain relatively stable in healthy adults [35] and those with arrhythmias [36]. However, systematic longitudinal assessment of native T1 reproducibility has not been well characterized in dilated cardiomyopathies. There are many factors that may have affected the association of T1 maps with LGE in the DMD patient population. Fatty infiltration has been well-described in skeletal muscle and may be present with progressive myocardial disease. This could lower native T1– thus, T1 values may depend on the ratio of fibrosis to fat in LGE. Indeed, a recent study suggested native T1 decreases with worsening LGE in DMD boys, which the authors hypothesized was secondary to fatty infiltration [37]. DMD boys also may have lower image quality as they age, as many DMD boys have progressive respiratory weakness, limiting ability for adequate breath-holding. This limitation should be corrected in most cases with the motion correction and we excluded patients with poor map quality. Positioning and image quality may lead to scan-to-scan variation in native T1 maps. This may lead to inaccurate estimation of the change in native T1. Finally, while LGE demonstrates replacement fibrosis, native T1 is increased with both replacement and diffuse fibrosis; in addition, native T1 detects abnormalities in both myocytes and the extracellular matrix. Taken together, these factors may contribute to the poor performance of our models when using native T1 as one of the predictors for global severity as assessed by LGE.

By contrast, strain is a more functional assessment and areas demonstrating abnormal strain patterns secondary to fibrotic changes would be expected to progress as the pathological process progresses. Good correlation with ε_cc and myocardial fibrosis has been previously demonstrated based on histopathology and CMR assessment with LGE and ECV in adult patients with aortic stenosis [38] and hypertrophic cardiomyopathies [39]. Our data suggest that this holds true for dilated cardiomyopathy in DMD boys, with strain predicting both change in presence/absence and change in LGE extent in most of our models. Of note, while strain predicts change in extent of LGE, model performance is suboptimal (Fig. 6). Further, univariate analysis demonstrates that LVEF is at least as good as our multivariate models with native T1 and strain in predicting presence or absence of LGE in DMD patients with normal function (Additional file 1: Figure S1). While LVEF alone predicts the presence of LGE with high specificity (although less robust sensitivity), this is not as useful in clinical practice as the population of interest is DMD patients with normal or stable LVEF and progression of LGE. Interestingly, our model performs better in terms of predictive power for all DMD boys compared to the subset with normal LVEF (≥ 55%) and our models have better negative likelihood ratios than LVEF alone (Table 3). This is consistent with studies in adult populations with chronic heart failure, where addition of strain to LVEF had modest but significant improvements in AUCs [40]. This suggests a very limited role for strain, where one could consider using it only in those DMD boys who have an absolute contraindication to gadolinium or in whom intravenous access cannot be obtained. Given the suboptimal results of most models, the authors cannot recommend routine use of non-contrast CMR in place of LGE in patients without a contraindication to gadolinium. Despite the safety concerns associated with contrast, DMD subjects will likely continue to require conservative administration of gadolinium contrast for appropriate clinical management.

It is important to note the segmental distribution of the predictive models using native T1 and strain. Unlike many other forms of cardiomyopathy, the fibrosis seen in DMD dilated cardiomyopathy tends to be patchy and found in the subepicardial region, often in one or more segments of the lateral free wall. Segmental differences in native T1 and strain are more likely to be predictive in areas where LGE is usually found in DMD patients. Thus, there may be some benefit to monitoring for changes in segmental parameters even in patients with normal LVEF and no previous LGE. Strain calculated using either feature tracking or myocardial tagging was predictive of LGE presence and extent with similar efficacy. It should be noted that HARP analyzes mid-myocardial strain while feature tracking analyzes endocardial strain. There is debate about which one is more relevant clinically, but we suspect that the difference is a result of the underlying software differences. Apical LGE was better predicted by ε_cc-tag, however, this likely is of less clinical importance given that the apex in DMD is relatively spared until late in the disease. It is still unclear what outcome measures are best used for DMD cardiomyopathy. LVEF and LGE have been the standard because clinicians have the most experience with their use and interpretation [11]. It is possible that more advanced methods, such as native T1, ECV, and strain, will supplant LVEF and LGE for assessment of DMD cardiomyopathy severity and prediction of progression, but this study was not designed to determine the optimal CMR biomarkers, only whether non-contrast sequences can predict LGE.

Emphasizing the importance of limiting contrast exposure, a recent study demonstrated no difference in medical management seen with traditional versus conservative contrast administration in DMD patients [41]. Of note, the conservative contrast group still received contrast 84% of the time, suggesting that conservative contrast administration may be just the first step in decreasing cumulative contrast load in DMD. While our institution has been using conservative gadolinium administration for the past 3 years, non-contrast sequences, particularly ε_ls and Ɛ_cc, have potential to help in further reducing contrast administration in this patient population.

Given the substantial mobility challenges and other morbidities, such as contractures and severe scoliosis, seen in DMD patients, reduction in scanner time is an important consideration for patient comfort as well as patient compliance for breath holding in order to obtain the best quality studies. Based on our institutional data, the average total scan time was 39.6 min, with 10 min attributed to post-contrast imaging from sequences that would not have been used if contrast was not administered. Thus, eliminating contrast administration would allow for sub-30 min studies in most cases, which is an important consideration for this patient population in particular, but may also be important in centers were scanner access is limited.

This study is limited by the fact that only 33% (18 of 55 patients) had a change in GSS ≥ 1; a larger change may have improved model performance. Additionally, 70% of patients had LGE severity score of at least 1 on their initial scan. This limited our ability to do subset analysis with sufficient power comparing the severity score 0 group alone with those ≥ 1. As discussed above, native T1 may not be completely linearly related to LGE extent and this may bias the data toward non-significance when there actually is a detectable difference for larger changes in LGE severity. Additionally, the variation in native T1 values is inversely related to flip angle, with lower flip angles having higher standard deviations between measurements [36]. We chose to use MOLLI due to its wide availability, which potentially make our findings more generalizable. However, it is possible that other methods of T1 mapping, such as SASHA, would provide a better prediction of LGE. There were some patients that were excluded because we were not able to obtain IV access in order to administer contrast and a very small minority had to stop early due to poor endurance or discomfort. These patients were not included if they did not complete the protocol; as this was a small minority of patients, we do not believe this led to any bias in the results. It should be noted that any patient with a heart rate > 90 bpm used the high heart rate version of the T1 map. In our data set, 72% of the CMRs were performed in the setting of heart rates > 90 bpm. While this sequence increases the voxel size by decreasing the matrix size, this correction significantly decreases the sensitivity of MOLLI to heart rates at longer T1 times. Our prior work demonstrated no significant effect of HR on T1 times using these parameters [21]. Further investigation with a larger, multicenter cohort may allow for better resolution of detectible differences in native T1 in patients who have small changes in the extent of LGE.

Most of the patients assessed were on steroids or cardiovascular medications. While medications may temper the progression of LGE, which could certainly increase the number of patients with no change in LGE on longitudinal assessment, therapies are unlikely to substantially change the predictive potential of these sequences. Despite these therapies, there does appear to be a relationship with mortality. It is important to assess prediction in a clinically relevant setting, and all of these medications represent standard therapies at most institutions.

Finally, another limitation of this analysis is that the models will require validation secondary to the large number of multiple comparisons.

In this cohort of DMD boys, models using pre-contrast sequences are predictive of presence and severity of LGE for a particular CMR using both feature tracking and myocardial tagging. For longitudinal assessment for individual patients, ε_cc-tag was superior for identification of change in severity by FWHM and ε_ls was superior for change in GSS. Although these models were predictive, they were only modestly so, suggesting that non-contrast studies cannot currently replace LGE imaging. Given the critical clinical data provided by LGE, we suggest that DMD subjects should continue to receive gadolinium contrast using a conservative protocol for clinical management. Non-contrast imaging should only be used to estimate LGE when contrast cannot be administered (renal failure or no intravenous access) or if total time must be decreased due to physical limitations.