Late gadolinium enhancement dispersion for predicting malignant arrhythmic events in patient with non-ischaemic dilated cardiomyopathy

Abstract

Aims

Arrhythmic risk stratification in patients with non-ischaemic dilated cardiomyopathy (DCM) remains challenging. The LGE-dispersion mapping is a novel method for the quantification of tissue heterogeneity through the Global Dispersion Score (GDS). We sought to evaluate the usefulness of GDS in arrhythmic risk stratification of DCM patients.

Methods and results

Consecutive non-ischaemic DCM patients underwent cardiac magnetic resonance imaging. GDS was calculated in LGE images. During a follow-up of 3.3 years (2 to 6 years), the combined endpoint of sudden cardiac death and appropriate implantable cardioverter-defibrillator intervention was considered. The final population included 510 patients (mean age was 56 ± 15 years). Left ventricular ejection fraction (LVEF) was >35% in 241 patients (47%). LGE was present in 225 patients (45%). Median extent of LGE was 12% of LV mass [interquartile range (IQR) 6–20%]. Among patients with positive LGE, GDS was 0.14 (IQR 0.08–0.20). During follow-up 81 patients had malignant ventricular arrhythmias (8 SCD, 73 appropriate ICD interventions). At Kaplan–Meier analysis, patients with GDS > 0.10 had worse prognosis than those with lower values of GDS (P < 0.0001). At multivariate analysis, GDS > 0.10 (HR 2.9, 95% CI: 1.7–5, P = 0.0002) was an independent predictor of events. The prognostic value of GDS was confirmed in subgroups of patients with LVEF ≤ 35% and >35%.

Conclusion

GDS is a useful marker to identify DCM patients at higher risk for malignant arrhythmic events regardless of LVEF and extent of LGE.

Graphical Abstract

Prognostic role of GDS in non-ischaemic DCM. The upper panel shows two examples of global dispersion mapping of patients with non-ischaemic DCM: (A) a case of DCM with cardiac arrest during the follow-up presenting a low extent of LGE (the contoured original LGE images in left panel, in middle panel the two-colour parametric map) but a high GDS (0.15, right panel); (B) a case of DCM without events during the follow-up but showing a greater extent of LGE (22% of LV mass, left and middle panels) but a lower GDS (right panel) than the previous patient. In the lower panel shows the Kaplan–Meier survival curve analysis comparing patients with GDS ≤ or >0.10. The graph demonstrated that patients with DCM with GDS >0.10 had greater probability of MVA than others (log-rank P < 0.0001).

Open in new tabDownload slide

Introduction

Non-ischaemic dilated cardiomyopathy (DCM) is a common heart muscle disorder (affecting 1 in 2500 adults), often genetic, defined by the presence of left ventricular or biventricular dilatation and systolic dysfunction in the absence of abnormal loading conditions or coronary artery disease (CAD).¹ Sudden cardiac death (SCD) accounts for 30% of all deaths in patients with DCM.² Arrhythmic stratification in DCM patients remains extremely challenging and currently relies on the assessment of left ventricular ejection fraction (LVEF), which is the primary determinant for implantable cardioverter-defibrillator (ICD) implantation.³

Although LVEF is an important prognostic factor in DCM, many patients with significantly impaired LVEF may still be at low risk for SCD. Moreover, most patients who experience SCD do not have severely reduced LVEF.^4,5 LV systolic function, assessed by EF, does not always correlate with the myocardial substrate required for ventricular tachycardia (VT) or ventricular fibrillation (VF). Instead, myocardial fibrosis has been recognized as an anatomical substrate for malignant ventricular arrhythmias (MVA) and SCD.⁶

Cardiac magnetic resonance (CMR) plays a crucial role in the diagnosis of DCM and in the non-invasive detection and quantification of myocardial fibrosis using the late gadolinium enhancement (LGE) technique.⁷ LGE is strongly and independently associated with MVA or SCD and is an incremental predictor of mortality in patients with DCM.^8,9 The LGE pattern also has important prognostic implications, regardless of LVEF. Mid-wall septal and ring-like LGE patterns are considered markers of greater risk for MVA than other presentations.

However, myocardial scar detected as LGE, regardless of its pattern or global extent, can have different presentations, distributions, and signal intensities, which have not previously been evaluated in DCM. LGE-dispersion mapping (LGE-DM) is a novel method of analyzing LGE that allows calculation of the Global Dispersion Score (GDS), a quantitative marker of signal heterogeneity, dispersion, and irregularity of myocardial scar.¹⁰ GDS has provided better risk stratification in patients with hypertrophic cardiomyopathy (HCM) and a low-to-intermediate 5-year risk of SCD.¹⁰

The aim of this study is to propose the use of GDS, calculated by LGE-dispersion mapping, to assess arrhythmic risk and prognosis in patients with DCM.

Methods

Patients

In this multicenter study, 510 consecutive DCM patients were prospectively enrolled between 2008 and 2020 and clinically followed up. CMR with LGE was performed in all 510 patients at the time of enrolment. LGE images were then retrospectively analyzed for the evaluation of GDS. Based on the European Society of Cardiology (ESC) criteria,¹ the diagnosis of DCM was made using the following criteria: (i) LV or biventricular systolic dysfunction (LVEF <45%) and dilatation not explained by abnormal loading conditions or CAD; (ii) LV or biventricular global systolic dysfunction (LVEF <45%) without dilatation, also not explained by abnormal loading conditions or CAD. LV dilatation was defined by a left ventricular end-diastolic volume indexed to body surface area (EDVi) > 2 SD above the normal range.¹¹

Exclusion criteria were (i) contraindication to CMR; (ii) glomerular filtration rate <30 mL/min (contraindication to gadolinium-based contrast agent); (iii) NYHA class IV; (iv) presence of CAD on coronary angiography and/or coronary CT angiography; (v) factors that could a priori compromise LGE image quality (including pacemaker, ICD, and loop recorder). After enrolment, 10 patients were excluded due to suboptimal LGE image quality. The final study population included 500 patients. Patients were evaluated either as inpatients or outpatients. Prior to the CMR scan, clinical information was collected. This observational and retrospective study received approval from the institutional internal review board and complies with the guidelines of the Declaration of Helsinki. Informed consent was obtained from all patients prior to enrolment.

CMR acquisition protocol

All the CMR scans were performed using 1.5 T whole-body CMR machines with a dedicated cardiac coil. According to the protocols recommended by the Society for Cardiovascular Magnetic Resonance, a short-axis breath-hold balanced steady-state free-precession image was used to evaluate global LV function, acquired with the following parameters: 30 phases, slice thickness 8 mm, no gap, 8 views per segment, 35–40 cm field of view, phase field of view 1, matrix 224 × 224, reconstruction matrix 256 × 256, 45° flip angle, repetition time 4 ms and echo time 2 ms. LGE images were acquired about 10 min after the administration of 0.5 molar gadolinium contrast agent (0.2 mmol/kg) in short and long-axis views, using an inversion recovery T1-weighted gradient-echo sequence, acquired with the following parameters: field of view 35 to 40 mm, slice thickness 8 mm, no gap, repetition time 3 to 5 ms, echo time 1 to 3, a flip angle of 20°, matrix 224 × 224, and reconstruction matrix 256 × 256. Using a TI-scout, the appropriate inversion time is identified to null normal myocardium.

Post-processing image analysis

Three certified CMR readers with Level III accreditation from the European Association of Cardiovascular Imaging performed offline, blinded evaluations of the CMR images. Functional parameters and LV mass were measured by analyzing short-axis cine images using a commercially available software package (cvi42, Circle International Corporation, Canada), as recommended. In the short-axis images, maximum LV end-diastolic wall thickness was also measured.

The extent of LGE and LGE dispersion was assessed using a previously validated, homemade software tool, the LGE Dispersion Tool.¹⁰ This software is available for research purposes by contacting the corresponding author via email.

Briefly, endocardial and epicardial contours were manually traced on each LGE short-axis image. A region of interest (ROI) was selected in an area of myocardium without LGE. In this ROI, the average signal intensity (SI) and standard deviation (SD) were measured. Myocardial pixels were considered hyper-enhanced if their SI exceeded the mean SI of the ROI by more than 6 SDs. Across the entire LV myocardium, the proportion of hyper-enhanced pixels was calculated and expressed as a percentage of total LV mass to determine the LGE extent.

A two-colour parametric map was then generated from each LGE image (Figure 1), where normal myocardium was shown in blue and enhanced myocardium in yellow. Dispersion mapping was automatically generated from the analysis of these two-colour parametric maps, as previously described.¹⁰ Briefly, a local measure of heterogeneity was computed by analyzing a 3 × 3 pixel grid centred on each pixel. A dispersion score, ranging from 0 to 8, was assigned to each pixel: a score of 0 was given when all 8 surrounding pixels had the same classification (normal or enhanced) as the central pixel: a score of 8 was assigned when all surrounding pixels had a different classification from the central pixel (Figure 1).

This calculation was performed for each enhanced pixel of the LV myocardium. The GDS was then calculated as the average score of all enhanced pixels. A dispersion map was generated for each LGE image, assigning a specific colour to each dispersion score at the pixel level. Examples of dispersion maps are shown in the Graphical Abstract.

An inter- and intra-observer reproducibility analysis for GDS quantification was performed in a previous study,¹⁰ demonstrating an intraclass correlation coefficient of 0.96 (95% CI: 0.90–0.99) and 0.98 (95% CI: 0.96–0.99), respectively.

Clinical follow-up

After the CMR examinations, all patients underwent follow-up. Clinical data were collected through a questionnaire completed during routine outpatient medical visits or by contacting the patient’s family or general practitioner. In the event of a clinical episode, complete documentation was obtained from the patient, a family member, or the referring physician. Follow-up was updated annually. Documented MVA included: SCD, resuscitated cardiac arrest, appropriate ICD shock or anti-tachycardia pacing (ATP), and sustained VT. ICD interventions were considered appropriate if they were triggered by life-threatening arrhythmias, such as VT exceeding the programmed ICD cut-off (≥12 intervals at >180 beats/min), or VF. Sustained VT was defined as lasting ≥30 s at a rate of ≥100 beats/min. The referring cardiologist interrogated the ICD to confirm the appropriateness of the shock or ATP, and to verify the presence of self-limiting episodes of supraventricular tachycardia (SVT). ECG Holter monitoring was performed every 6 months. A panel of three expert investigators reviewed and confirmed the occurrence of MVA events.

Statistical analysis

A Kolmogorov-Smirnov test was used to test each variable for normal distribution. Normally distributed variables were shown as mean ± SD, while non-normally distributed variables were shown as median (25–75th). Depending on the situation, the Fisher exact test or the χ² test was used to compare categorical variables. The Wilcoxon non-parametric test or the t-test, depending on the situation, was used to compare continuous variables. Bonferroni correction was applied where necessary. A maximally selected rank statistical analysis was used to define the optimal cut-off of the GDS for survival analysis using the maxstat package of R software. The log-rank test was used to create and compare longitudinal curves between groups using the Kaplan–Meier time-to-event approach. Univariate and multivariable Cox regression analysis with competing risk analysis was used to explore the impact of each significant variable in univariate analysis to predict the occurrence of hard cardiac endpoints. The Harrell-C statistic was used in multivariable models. A P < 0.05 was considered statistically significant.

Results

The final population included 510 patients with non-ischaemic DCM (292 males; 57%) and a mean age of 56 ± 15 years. The general characteristics of the population are shown in Table 1. A family history of SCD was recorded in 89 patients (18%). The average LVEF was 35% ± 11; LVEF was >35% in 241 patients (47%). At the time of the CMR scan 256 patients (50%) were in NYHA class I, 196 (39%) in NYHA class II, and 58 (11%) in NYHA class III. The median NT-pro-BNP level was 1150 (466–2276) pg/mL. Left bundle branch block was present in 104 patients (20%). On 24-h Holter monitoring, 148 patients (29%) showed 7 (NSVT), and103 (20%) had >1000 premature ventricular contractions (PVCs)/24 h. Sixty patients (12%) were on antiarrhythmic therapy (mostly amiodarone) at the time of enrolment. CMR was performed a median of 7 months (IQR 2–17) after the initial diagnosis. LGE was present in 234 patients (45%), with a median extent of 12% (IQR 6–20%) of LV mass. A mid-wall septal LGE pattern was observed in 116 patients (22%), a ring-like pattern in 69 (13%), and a sub-epicardial pattern in 49 (10%). The median GDS was 0 (IQR 0–0.12). Among patients with positive LGE, the median GDS was 0.14 (IQR 0.08–0.20).

Follow-up in the whole population

Clinical follow-up was conducted over a median of 3.3 years (IQR 2–6). During this period, an ICD was implanted in 239 patients (47%), of whom 122 also received cardiac resynchronisation therapy. During the follow-up, 81 patients experienced MVA: 8 cases of SCD, 27 appropriate ICD shocks for VF/VT, and 46 appropriate ATP interventions. In addition, 53 patients died due to heart failure unrelated to arrhythmic complications. The characteristics of patients with and without MVA are reported in Table 2. Patients with MVA were significantly older (P < 0.01), more frequently in NYHA class > I (P = 0.01), and more often had episodes of non-sustained ventricular tachycardia (NSVT).

Patients who experienced events had significantly higher LV end-diastolic volume index (LVEDVi) (P < 0.001), lower LVEF (P < 0.001), more frequent presence of LGE (P < 0.001), greater extent of LGE (P < 0.001) and more often a mid-wall septal or ring-like LGE distribution (P < 0.001). Events occurred in 36 out of 116 patients with mid-wall septal LGE, and 23 out of 69 with ring-like LGE (P = 0.82). Finally, patients with events had significantly higher GDS values than those without events: median 0.15 (IQR 0.05–0.20) vs. 0 (IQR 0–0.09) (P < 0.001).

Kaplan–Meier analysis showed that patients with positive LGE had a significantly higher probability of MVA compared with those without (Figure 2, left panel). Moreover, patients with mid-wall septal and/or ring-like LGE patterns had a worse prognosis than those with other LGE patterns (Figure 2, right panel).

Using maximally selected rank statistics, the optimal cut-off for GDS in predicting MVA was identified as >0.10 (Figure 3). Differences among three subgroups of patients with (a) negative LGE (GDS = 0), (b) positive LGE and GDS ≤ 0.10, and (c) positive LGE and GDS > 0.1 are reported in Table 3. Compared with those with GDS ≤ 0.10, patients with GDS > 0.10 were younger, more often on diuretic therapy, had greater LV mass, and more frequently showed a sub-epicardial LGE pattern. In comparison to patients with GDS = 0, those with GDS > 0.10 were more frequently male, more often on diuretics, more commonly experienced NSVT, and showed signs of more advanced disease on CMR, including higher LVEDVi and LV mass, lower LVEF, and greater LGE extent.

Kaplan–Meier analysis demonstrated that patients with GDS > 0.10 had a significantly higher risk of MVA compared with those with lower GDS values (P < 0.0001) in the overall population (Graphical Abstract). The prognostic value of GDS remained significant even when considering only patients with mid-wall septal or ring-like LGE (Figure 4). As shown in Table 4, among patients with mid-wall septal or ring-like LGE, the prevalence of MVA increased from 17% in those with GDS ≤ 0.10 to 40% in those with GDS > 0.10 (P = 0.0016).

In univariate Cox regression analysis, the following variables were associated with the occurrence of MVA (Table 5): age, male sex, NYHA class > I, NSVT, LVEDVi, LVEF, LV Mass, RVEF, positive LGE, the extent of LGE, mid-wall septal LGE and GDS (threshold >0.10).

In multivariate Cox regression with competing risk analysis (Table 6), the following were identified as independent predictors of MVA: age (HR 1.03, 95% CI: 1.01–1.05, P = 0.02), NSVT (HR 2.4, 95% CI: 1.4–4.2, P = 0.001), LV mass (HR 1.01, 95% CI: 1.01–1.02, P = 0.04), mid-wall septal/ring-like LGE (HR 3.3, 95% CI: 1.8–5.8, P = 0.0002) and GDS >0.10 (HR 2.9, 95%: CI 1.7–5, P = 0.0002).

In Figure 5, the incremental predictive value of GDS > 0.10 for MVA is illustrated by stepwise model inclusion, progressively adding LV mass, NSVT, mid-wall/ring-like LGE, and age.

Patients with LVEF ≤35%

In patients with LVEF ≤ 35%, a total of 60 MVA events occurred during follow-up. Kaplan–Meier analysis showed that within this subgroup, patients with GDS > 0.10 had a significantly worse prognosis compared with those with lower GDS values (log-rank P = 0.0001, Figure 6).

As shown in Table 7, univariate regression analysis for predicting events in this subgroup identified the following variables as significantly associated with outcomes: age, LVEF, LV mass, positive LGE, mid-wall septal/ring-like LGE and GDS > 0.10. These parameters were included in the multivariate regression analysis (Table 8), which demonstrated that GDS > 0.10 was the only independent predictor of events in patients with LVEF ≤ 35% (HR 19.4; 95% CI: 6.9–42; P = 0.0009).

Patients with LVEF > 35%

In patients with LVEF > 35%, 21 MVA events occurred. Kaplan–Meier analysis showed that, within this subgroup, patients with GDS > 0.10 had a significantly worse prognosis compared with those with lower GDS values (log-rank P = 0.0002, Figure 6).

A further univariate regression analysis was conducted in this subgroup (Table 8), identifying the following variables as significantly associated with events: age, positive LGE, mid-wall septal LGE, sub-epicardial LGE and GDS > 0.10. Due to the limited number of events in this subgroup, four different bivariate models were constructed (Table 9), each including GDS > 0.10 and one of the other variables significantly associated with events in the univariate analysis. As shown in the Table 10, GDS > 0.10 remained a significant predictor of events in all bivariate models.

The 10-year risk of MVA for the overall population and for subgroups with LVEF ≤ 35% and > 35%, stratified by LGE positivity, LGE pattern, and GDS, is presented in Table 11.

Discussion

The main results of the present study can be summarized as follows: (i) in non-ischaemic DCM, GDS is a novel prognostic marker that allows the identification of a subgroup of patients at higher risk of ventricular arrhythmias; (ii) GDS has incremental prognostic value beyond the presence of LGE, and in particular, the combination of mid-wall septal/ring-like LGE and GDS > 0.10 is associated with a 5-year MVA risk of 39%; (iii) GDS was the only independent predictor of MVA in the subgroup of patients with LVEF ≤ 35%, and it was also significantly associated with MVA in the LVEF > 35% subgroup across all bivariate models.

The 2022 AHA/ACC/HFSA guidelines for the management of heart failure recommend ICD implantation in patients with LVEF ≤ 35% and NYHA class II or III symptoms, on chronic guideline-directed medical therapy, with a reasonable expectation of survival for more than one year.¹¹ ICD implantation is also recommended for patients with LVEF ≤ 45% who carry arrhythmogenic genotypes.

A similar approach is adopted in the 2023 ESC guidelines for the management of cardiomyopathies, which recommend ICD implantation for primary prevention in patients with non-ischaemic DCM, LVEF ≤ 35%, and symptomatic heart failure despite at least three months of optimized medical therapy.¹² These guidelines also suggest considering ICD implantation in patients with LVEF > 35% who present with high-risk genotypes and/or additional risk factors such as syncope or positive LGE on CMR.

Three randomized controlled trials, Sudden Cardiac Death in Heart Failure Trial, DEFINITE (Defibrillators in Non-ischaemic Cardiomyopathy Treatment Evaluation), and DANISH (The Danish Study to Assess the Efficacy of ICDs in Patients with Non-ischaemic Systolic Heart Failure on Mortality), have reported discordant results regarding the impact of ICD implantation in non-ischaemic DCM.^4,13,14 As a result, new risk stratification criteria are still being sought to more accurately identify patients who would benefit most from ICD therapy.

Multiple aspects of LGE can be evaluated for the prognostic assessment of non-ischaemic DCM, including the presence/absence of LGE, its extent, the distribution pattern, the presence and extent of a grey zone, and new features such as dispersion mapping. In non-ischaemic DCM, the prognostic value of LGE for predicting SCD and MVA has been demonstrated in several prospective studies^7–9 and summarized in two meta-analyses.^8,15

In the meta-analysis by Becker et al.¹⁵ which included data from 34 studies involving a total of 4554 patients, the presence of LGE was associated with an odds ratio of 4.52 (95% CI: 3.41–5.99) for MVA.¹⁵ Similar findings were reported in the meta-analysis by Di Marco et al., which also demonstrated that the prognostic value of LGE was stronger in patients with LVEF > 35% than in those with LVEF ≤ 35%.⁸

The quantification of LGE remains limited by the lack of a standardized approach. Several methods have been employed, including visual identification based on manual change of SI thresholds with fixed cut-offs of 2 to 6 standard deviations (SD) above the mean SI of remote (non-enhanced) myocardium, the Full-Width Half-Maximum (FWHM) method and the Rayleigh curve method, among others.^16,17 The 2 SD method has been largely abandoned, as it significantly overestimates LGE extent because normally null myocardium does not follow a Gaussian distribution.¹⁷ The FWHM method is considered particularly accurate for quantifying ischaemic scars and in myocarditis, but it is difficult to apply to the small, patchy scars typically seen in DCM, as it requires placing a ROI in enhanced myocardium.¹⁶ The Rayleigh curve method is the only technique with a strong physics-based rationale, but it is highly complex, requiring adjustment of the threshold formula based on the configuration and number of channels of the receiver coil.¹⁷ As a result, the 6 SD methods are often considered a pragmatic compromise, offering a reasonable balance between robustness and simplicity for LGE quantification in cardiomyopathy.

Although there is substantial evidence supporting the prognostic role of LGE quantification in ischaemic heart disease and HCM, where an LGE extent >15% has been incorporated into the guidelines as a marker of arrhythmic risk, in non-ischaemic DCM, the literature presents conflicting results.

In a cohort of DCM patients, Gulati et al.⁹ found that both LGE presence (HR 4.61) and extent (HR 1.1) were independent predictors of SCD or aborted SCD. In contrast, the majority of other studies in non-ischaemic DCM did not find a significant association between LGE extent and SCD or MVA.^18–22

Several studies^19,20,23,24 have attempted to identify a more accurate LGE extent cut-off for predicting major arrhythmic events in DCM using ROC analysis. Despite differences in quantification techniques, these studies consistently reported a common cut-off of approximately >5% LGE extent. However, in all these studies, it was the presence of LGE, rather than its extent, that emerged as an independent predictor of events in multivariate analysis. Overall, the relationship between LGE extent and arrhythmic risk in non-ischaemic DCM remains unclear. In studies where a significant association between LGE extent and events was reported, the HR for LGE presence was typically much higher than for extent. Moreover, the >5% threshold corresponds to an extent smaller than a single myocardial segment in the standard 17-segment model, and may essentially function as a cut-off for identifying true LGE positivity, helping to exclude false positives, such as signals from connective tissue at right ventricular insertion points, septal perforator branches, or imaging artefacts.

The prognostic role of LGE distribution patterns in DCM has also been investigated. Most studies have found that the mid-wall pattern of LGE carries greater prognostic significance compared with the sub-epicardial pattern, particularly when the mid-wall of the interventricular septum is involved.^{9,18,23,25–27} Among patients with septal mid-wall involvement, those who also show enhancement of the lateral segments and other myocardial walls, forming the so-called ring-like pattern, appear to represent a higher-risk subgroup.^18,26,28

However, a few studies have reported opposing results, finding no significant association between LGE pattern and clinical events.^{20–22,24,26} These discrepancies may be explained by several factors, including low statistical power due to subdivision of patients into small subgroups, inconsistent definitions of LGE patterns across studies, and heterogeneity in the characteristics of the enrolled populations.

In our population, LGE was positive in 45% of DCM cases, a prevalence consistent with previous studies, where LGE prevalence ranged from 30% to 46%.^18,27,29 The presence of LGE has been associated with an increased risk of all-cause mortality, heart failure hospitalisation, and SCD in patients with DCM.³⁰

Consistent with prior findings, our study confirmed that mid-wall septal/ring-like LGE was an independent predictor of MVA in multivariate analysis. This result underscores the important prognostic value of LGE pattern and distribution, regardless of extent.¹⁸ Notably, the mid-wall septal/ring-like pattern has also shown prognostic relevance in other cardiac conditions. For example, the ITAMY study demonstrated that, in patients with acute myocarditis and preserved LVEF, LGE located in the mid-wall layer of the anteroseptal segment was associated with a worse prognosis than other LGE patterns.³¹

In addition to traditional LGE assessment, we evaluated a further characteristic using LGE-dispersion mapping, a texture analysis technique that assesses inter-pixel relationships, grey-level distribution, and other statistical properties that characterize image texture. LGE-dispersion mapping enables quantification of fibrosis heterogeneity and dispersion through calculation of the GDS.¹⁰ We previously evaluated GDS in a population with HCM, demonstrating its prognostic value in predicting MVA.¹⁰ In that study, LGE dispersion maps were generated using three-colour parametric maps, distinguishing among normal myocardium, hyper-enhanced myocardium, and mid-enhancement myocardium, as mild enhancement in HCM was considered a marker of increased arrhythmic risk.^32,33 In the present study, we adopted a dichotomous approach, categorising only normal and enhanced myocardium. This decision was because positive LGE is less common and generally less extensive in DCM compared with HCM. Furthermore, in DCM, myocardial wall thinning is often present, which makes the accurate characterisation of LGE more challenging. Despite these differences, the findings of this study indicate that GDS retains a crucial prognostic role even in the context of DCM.

Intrinsically, the GDS integrates multiple aspects:

1. Presence of LGE, as GDS = 0 in the absence of any enhancement.

2. Extent of LGE, since a greater number of enhanced pixels increases the probability of higher dispersion values.

3. Pattern of LGE, as mid-wall LGE, being surrounded by normal myocardium, tends to generate higher dispersion than sub-epicardial LGE.

4. Dispersion of LGE within the myocardium, which reflects LGE entropy and heterogeneity.

A high GDS value corresponds to irregularly shaped myocardial scars, which in DCM may be more arrhythmogenic than a single, large, homogeneous, and inert scar. From a practical point of view, GDS quantification is technically simple and can be integrated into routine clinical workflows. It can be performed simultaneously with standard LGE quantification as an additional automatic analysis, requiring no additional time. The only manual step for the operator is the tracing of endocardial and epicardial contours of the left ventricular myocardium, an operation that modern post-processing software can perform automatically using artificial intelligence algorithms.

Our results demonstrated that GDS > 0.10 was associated with a higher risk of major cardiac events, independent of LGE extent. Patients with GDS > 0.10 were significantly younger, more frequently on diuretic therapy, experienced more frequent episodes of NSVT, and exhibited signs of more advanced disease on CMR (higher LVEDVi, higher LV mass, and lower LVEF) compared with those with lower GDS values. These findings indicate that higher GDS scores are associated with worse clinical status. In multivariable analysis, GDS > 0.10 emerged as an independent predictor of MVA, alongside age, NSVT, LV mass, and mid-wall septal/ring-like LGE. GDS also provided additive prognostic value within the statistical model incorporating all these parameters.

This additive role of GDS was further confirmed in the subgroup of patients with mid-wall septal/ring-like LGE, where the prevalence of MVA increased from 17% in those with low GDS to 40% in those with GDS > 0.10. This underscores the importance of scar heterogeneity as a relevant risk factor even within this already high-risk LGE pattern.

The prognostic value of GDS > 0.10 remained significant when stratifying patients by LVEF severity. Specifically, GDS was the only independent predictor of cardiac events in patients with LVEF ≤ 35%, and it remained a significant predictor across all bivariate models in the LVEF > 35% subgroup. However, interpretation in this latter group is limited by the low number of events, and larger studies are warranted to confirm the prognostic role of GDS in this population.

The prevalence of MVA in the present study (15%) was slightly higher than in previous reports. In the meta-analysis by Di Marco et al.⁸ which included a total of 2948 patients with DCM, ventricular arrhythmic events occurred in 12% of patients overall and in 21% of those with positive LGE. In that meta-analysis, 43% of patients had LGE, compared with 45% in our study. Moreover, the average follow-up in our study was longer, as only 5 out of 29 studies in the meta-analysis reported a follow-up duration >2 years.

Several limitations of this study should be acknowledged. First, the study population was of moderate size, and although patient enrolment was prospective, the analysis of GDS from LGE images was conducted retrospectively. Prospective studies are needed to confirm the prognostic value of GDS in DCM.

Second, GDS was measured in two dimensions, by analyzing the eight surrounding pixels around a central voxel. A three-dimensional analysis, incorporating pixels from adjacent slices, could have provided greater accuracy. However, we employed a two-dimensional LGE pulse sequence with an 8 mm slice thickness, which remains the most used technique for LGE assessment in clinical practice.

Third, T1-, T2-, and extracellular volume (ECV) mapping techniques were not used, as they were not available at our institutions during the initial enrolment period. Future studies could explore the relationship between LGE dispersion, T1 mapping, and ECV mapping.

Fourth, genetic testing was performed in only a minority of patients, as it was not a mandatory criterion for risk stratification in DCM at the time, and the current understanding of the genetic contribution to DCM had not yet fully emerged. Future research should investigate the relationship between high-risk genotypes and GDS.

Additionally, although we excluded patients with a diagnosis of prior myocarditis, we cannot completely rule out DCM secondary to subclinical myocarditis, as endomyocardial biopsy was not performed.

Finally, patient enrolment spanned 18 years, during which new heart failure therapies were introduced. While this may have influenced patient outcomes, therapeutic evolution during follow-up is a common limitation in long-term prognostic studies.

Conclusion

GDS is a quantitative marker of the signal heterogeneity, dispersion, and irregularity of myocardial fibrosis in DCM. GDS > 0.10 was the only independent predictor of MVA in patients with LVEF ≤35% and was a significant predictor of events in patients with LVEF >35%. GDS represents a valuable tool for identifying DCM patients at higher risk of major cardiac events, regardless of LVEF and LGE extent.

Clinical perspective

Competency in patient care and procedural skills

In patients with non-ischaemic DCM, those with a high GDS are at increased risk of malignant ventricular events. GDS provides additive prognostic value beyond that of NSVT, mid-wall septal/ring-like LGE, and left ventricular mass.

Translational outlook

Further studies are warranted to investigate the association between elevated GDS and the presence of pathogenic genetic mutations linked to high arrhythmic risk.

Funding

Institutional funds of G. Monasterio CNR-Tuscany Foundation, Pisa, Italy and of University of Pisa.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author. In particular, the post-processing software for LGE image analysis and for the quantification of the GDS is available upon request from the corresponding author.

REFERENCES

Author notes