Analysis of Three-Dimensional Scar Architecture and Conducting Channels by High-Resolution Contrast-Enhanced Cardiac Magnetic Resonance Imaging in Chagas Heart Disease

ABSTRACT Background: We aimed to describe the morphology of the border zone of viable myocardium surrounded by scarring in patients with Chagas heart disease and study their association with clinical events. Methods: Adult patients with Chagas heart disease (n=22; 55% females; 65.5 years, SD 10.1) were included. Patients underwent high-resolution contrast-enhanced cardiac magnetic resonance using myocardial delayed enhancement with postprocessing analysis to identify the core scar area and border zone channels number, mass, and length. The association between border zone channel parameters and the combined end-point (cardiovascular mortality or internal cardiac defibrillator implantation) was tested by multivariable Cox proportional hazard regression analyses. The significance level was set at 0.05. Data are presented as the mean (standard deviation [SD]) or median (interquartile range). Results: A total of 44 border zone channels (1[1-3] per patient) were identified. The border zone channel mass per patient was 1.25 (0.48-4.39) g, and the extension in layers of the border zone channels per patient was 2.4 (1.0-4.25). Most border zone channels were identified in the midwall location. Six patients presented the studied end-point during a mean follow-up of 4.9 years (SD 1.6). Border zone channel extension in layers was associated with the studied end-point independent from left ventricular ejection fraction or fibrosis mass (HR=2.03; 95% CI 1.15-3.60). Conclusions: High-resolution contrast-enhanced cardiac magnetic resonance can identify border zone channels in patients with Chagas heart disease. Moreover, border zone channel extension was independently associated with clinical events.


INTRODUCTION
Chagas disease (CD) still affects six to seven million people worldwide. Although this neglected disease occurs mainly in Latin America, it is increasingly recognized as a new public health challenge in non-endemic countries due to migration 1, 2 . Chronic Chagas heart disease (CHD) is characterized by chronic fibrosing myocarditis that leads to cardiac remodeling, heart failure (HF), arrhythmia, and cardioembolic stroke [3][4][5] .
Cardiac magnetic resonance imaging (MRI) permits the evaluation of the prognostic value of cardiac fibrosis in CHD. Indeed, cardiac fibrosis was reported to be associated with the occurrence of ventricular arrhythmias 6,7 , cardiovascular events 8 , and all-cause mortality 9 .

Santos JBF et al. | 3D Conducting Channels by ce-CMR in Chagas Disease
High-resolution contrast-enhanced cardiac MRI (ce-CMR) precisely differentiates the scar core and the morphology of the border zone (BZ) of viable myocardium surrounded by the scar core and/or the mitral annulus, and that connect two areas of normal myocardium 10 . These BZ channels have shown a close correlation with conducting channels on endocardial electroanatomic maps, allowing reentrant circuits 10 . In post-myocardial infarction patients, BZ size was associated with the inducibility of ventricular tachycardia (VT) and mortality 11 . BZ channel mass was an independent predictor of sustained VT in a retrospective case-control study among postmyocardial infarction patients 12 . Transmural BZ channels were also an independent predictor of post-ablation VT recurrence 13 . Furthermore, in ischemic and non-ischemic patients with left ventricular (LV) systolic dysfunction, BZ mass, and BZ channel mass were the strongest predictors of appropriate defibrillator therapy or sudden cardiac death 14 . In patients that implanted an internal cardiac defibrillator (ICD) for primary prevention, the identification of BZ channels or a scar mass larger than 10 g was associated with appropriate ICD therapy, in contrast with patients without those characteristics who presented a very low rate of appropriate therapies 15 .
Identifying BZ channels in CHD is important since sudden death is the primary mode of death in CHD 16 . ICD implantation is indicated as secondary prophylaxis in CHD for those who survive a sudden cardiac arrest or present sustained VT 17 . However, the clinical benefit of ICD in CHD is challenged by frequent appropriate shocks 18 that VT ablation procedures could minimize. The mapping of BZ channels by ce-CMR may increase the success of such procedures in patients with CHD. Therefore, our study aimed to evaluate if ce-CMR image postprocessing could identify BZ channels in patients with CHD and describe their morphology and association with clinical events.

Study Subjects
This paper is the continuation of a previous project on the study of cardiac MRI in patients with CHD with and without ventricular arrythmias 6 . The current longitudinal and prospective study evaluated a convenience sample of adult patients of both sexes with chronic CHD who underwent ce-CMR for research purposes with a diagnosis of cardiac fibrosis between October 2014 and March 2016. All patients had regular appointments at the Clementino Fraga Filho University Hospital, Rio de Janeiro, Brazil, and had to be away from an area with endemic CD transmission for more than 20 years, present with chronic CHD according to the Brazilian CD consensus 3 , and test negative for ischemic heart disease on treadmill exercise testing. CD diagnosis was confirmed by positivity in two distinct serological tests 3 .
The exclusion criteria were as follows: clinical history of another concomitant heart disease, atrial flutter/fibrillation, thyroid disease, chronic kidney disease, two or more risk factors for coronary artery disease, pregnancy, and any contraindication to undergo cardiac MRI.
CHD classification criteria followed the Brazilian CD consensus 3 : stage A (electrocardiography [ECG] changes compatible with CHD but no HF symptoms or LV wall motion changes in echocardiography), stage B (asymptomatic segmental or global LV systolic dysfunction, subclassified into B1 or B2 according to LV ejection fraction higher or lower than 45%), stage C (history of HF), and stage D (end-stage HF).
Patients were clinically evaluated, had their epidemiological data collected, and underwent two-dimensional echocardiography, ECG, and treadmill exercise testing. Holter monitoring exams were completed for 14 patients at baseline. All exams for each patient were obtained within one month. After this evaluation was complete, patients were scheduled to undergo ce-CMR.

Electrocardiogram
The 12-lead ECG tracing was interpreted following the Minnesota Code Manual of Electrocardiographic Findings (modified for CD) 19 . The ECG abnormalities that were compatible with CHD followed previously published criteria 3 .

High-Resolution Ce-CMR
All ce-CMRs were acquired using a 3.0-T Siemens Verio (Siemens Healthcare, Erlangen, Germany) equipped with a cardiac 12-element phased-array coil and advanced cardiac-dedicated software. The images obtained were stored de-identified into DICOM format for offline analysis.
The LV images were acquired during a 15 s inspiratory breathhold to decrease breathing movement artifacts. Longitudinal and short axis LV images were acquired by two ECG-triggered pulse sequences at the same spots. LV diameters, volumes, and ejection fraction, right ventricular (RV) volume and ejection fraction, and left atrial (LA) volume were evaluated by cine-CMR using the steady-state free precession protocol. LV mass was the result of the multiplication of LV myocardial volume by its myocardial tissue density (1.05 g/cm 3 ). Images were acquired with 1.4 mm slice thickness with no space between slices up to the LV apex.
Images in the longitudinal and short axis projections were obtained using an inversion-recovery prepared gradient-echo for myocardial delayed enhancement (MDE) protocol 10 to 20 min after intravenous injection of 0.1 mmol/kg of gadolinium-based contrast (Dotarem ® , Guerbet, France) to identify cardiac fibrosis. Fibrosis location and pattern were qualitatively determined. Fibrosis mass was calculated using a visual sub-segmental analysis as previously described 20 and presented in absolute values and as a percentage of the LV mass. The LV 17-segment model was used for segmental MDE analysis 21 . Scar distribution patterns were classified into three categories as previously described 8 : transmural (any segment with fibrosis in more than 50% of its thickness, in up to eight segments); focal (no transmural scar in less than nine segments); and diffuse (scar areas identified in nine segments or more). Segmental fibrosis was categorized as transmural, subendocardial, midwall, or subepicardial.

Ce-CMR Image Postprocessing
The postprocessing ce-CMR analyses were performed in Spain, as previously described 10 . Briefly, Automatic Detection of Arrhythmic Substrate -Ventricular Tachycardia (ADAS-VT) software was used in the Gimias environment (http://www.gimias.org). The LV epicardial and endocardial borders were automatically traced using three anatomic landmarks (mitral and aortic rings and the LV apex) and manually corrected as needed.
Twenty concentric surface layers (from 5% to 95% of the LV wall thickness) were created through the myocardial wall from the endocardium to the epicardium (Figure 1). A 3D shell was obtained for each layer. The signal intensity (SI) distribution in each shell FIGURE 1: 3-dimensional signal intensity maps in 10 out of 20 surface layers (from 5% to 95% of the LV wall thickness) obtained from a representative case depicting the normal myocardium (coded in purple), scar core (coded in red), and border zone (coded in blue-green-yellow).

FIGURE 2:
Signal intensity maps acquired from contrast-enhanced cardiac magnetic resonance representing in color-coded shells the left ventricle of a patient with a scar core area surrounded by a border zone (BZ) area with several long BZ channels (A) and another patient with a scar core area with a lower number of BZ channels with shorter extension (B). Myocardial delayed enhancement-depicted areas of cardiac fibrosis in all segments of the septum, apex, and apical segment of the inferior wall in A and in the basal and mid segments of the inferolateral wall in B.
The signal intensity codification is as follows, normal myocardium: purple, scar core: red, and BZ: blue-green-yellow. The BZ in light blue contains the BZ channels coded in white.
obtained from the ce-CMR was depicted by means of a color code in which the scar core areas were coded in red and BZ areas were coded in blue-green-yellow, using an algorithm based on maximum pixel SI in the LV wall. The automatic algorithm used a SI threshold above 60% of the maximum pixel SI to identify the core and from 40 to 60% of the maximum pixel SI to identify the BZ 22 . The extent and shape of the scar, categorized as core or BZ, was represented in three dimensions (3D) across the LV wall by the color-coded SI maps of the 20 transmural shells using the threshold-based classification.
The software enables the 3D identification of the core, BZ channel number, mass, location, and extension using a specific function to calculate the geodesic distance between two given points in a shell. The BZ was coded light blue, with the BZ channels coded white (Figure 2).

Study End-Point and Follow-up
The study end-point was a composite of cardiovascular death or ICD implantation. Patients were regularly followed up at Clementino Fraga Filho University Hospital during the study followup, and phone contact was made if a patient missed their clinical appointments. Additional Holter monitoring exams were performed whenever patients developed new symptoms of syncope or near syncope. Amiodarone was prescribed for all patients with: 1 -LV ejection fraction under 40% and non-sustained VT on Holter, and history of syncope or sustained VT or ventricular fibrillation (VF) inducible during electrophysiological study; 2 -sustained VT/VF associated with symptoms or not. Electrophysiological studies were performed in case of ICD implantation indication or the presence of non-sustained VT and LV ejection fraction under 40%. ICD implantation was indicated for all patients with either: 1-sudden cardiac arrest due to sustained VT/VF and LV ejection fraction under 35%; or 2-sustained VT and LV ejection fraction under 35%; or 3-nonsustained VT and LV ejection fraction under 35% and sustained VT/VF inducible during electrophysiological study.
Death was classified as sudden if a previously stable patient died within 1 h of symptom onset, during sleep, or unwitnessed. Acute non-cardiac illness, drug overdose, terminal disease, metabolic causes, and vascular non-cardiac disease should be ruled out before the etiology was considered cardiac 23 . Information about death was obtained from institutional medical records and from a direct interview with participants' relatives, who were also requested to provide copies of the death certificates.

Statistical Analysis
Calculations were performed using MedCalc 12.5.0.0 statistical software. Normally distributed continuous variables are expressed as the mean and standard deviation (SD), while skewed distributed variables are expressed as the median (interquartile range). Kolmogorov-Smirnov tests demonstrated that the continuous variables were normally distributed if the obtained P-values were >0. 10. Categorical variables are expressed as absolute and percentage values. Comparisons between groups were performed using an unpaired Student's t-test or Mann-Whitney test, as appropriate. Separate multivariable Cox proportional hazards regression analyses adjusting for LV ejection fraction and LV fibrosis mass were performed to test if the BZ channel parameters were associated with the studied end-point. LV ejection fraction 24 and fibrosis mass 9 were chosen as the adjusting variables based on previous studies that demonstrated their value as mortality predictors in CD. Model Harrell's C-index (Harrell) 25 was calculated for each multivariate model and represents a goodness of fit measure for models. Values of C were interpreted as the area under the receiver operating characteristic curve. A value of C close to 0.5 indicated that the model was not superior to random chance to predict an outcome, while values > 0.7 indicated a good prediction model. The modified τ method was used to determine the confidence interval of Harrell's C-index 26 . Schoenfeld residuals did not discard the proportional assumption of Cox proportional hazards regression analysis. The significance level was set at 0.05.  Helsinki, 1964Helsinki, , as revised in 1975Helsinki, , 1983Helsinki, , 1989Helsinki, , 1996Helsinki, , and 2000. All patients gave informed consent before participating.

Patient Characteristics
A total of 22 patients (55% females; 65.5 years (SD 10.1) were included in this study. Most of the patients were in the earlier stages of CHD (stage A or B1). Nine of these patients (40.9%) presented the associated digestive form. There was a similar sex distribution. Most patients were white, originated from the Northeast Brazilian region, and had arterial hypertension. The most common symptoms were palpitations, near syncope, and syncope ( Table 1). The most common ECG abnormalities were primary T wave changes, right bundle branch block, left anterior hemiblock, and premature ventricular complex ( Table 1). The most common abnormalities recorded on 24 h Holter monitors were premature ventricular complexes, followed by non-sustained VT and sinus pause >2 s. Only one participant had sustained VT ( Table 1).

Ce-CMR
The LV volumes, end-diastolic diameter and ejection fraction, LA volume, and RV end-diastolic volume and ejection fraction are depicted in Table 2. A total of 12 (54.5%) patients presented with LV systolic dysfunction, while only one presented with an RV ejection fraction below 45%. Among the patients with LV systolic dysfunction, five had mild, six had moderate, and one had severe, according to criteria previously published 27 .   A total of 44 BZ channels were identified ( Table 2). One patient presented with seven BZ channels, three with four BZ channels, two with three BZ channels, three with two BZ channels, and 13 with one BZ channel. Regarding CHD stages, one patient with stage C disease had seven channels, while another had one channel; the patient with stage B2 disease had four channels; those with stage B1 disease had 2.1 channels (SD 1.2), and those with stage A disease had 1.1 channels (SD 0.3). Twenty-one (47.7%) BZ channels were confined to just one layer; five (11.4%) to two layers; five (11.4%) to three layers; four (9.1%) to four layers; one BZ channel (2.3%) each was confined to five, six, seven, eight, 10, 13, 15, 16, and 18 contiguous layers. Most BZ channels were identified in the midwall (n=28; 63.6%), followed by the subendocardium (n=6; 13.6%) and subepicardium (n=2; 4.5%). The other four BZ channels (9.1%) had a path through all layers of the myocardium (transmural), two BZ channels (4.5%) showed a path involving the midwall and the subendocardium, and two BZ channels (4.5%) showed a path involving the midwall and the subepicardium.

Study End-Point and Survival Analysis
Six patients presented the studied combined end-point during a mean follow-up of 4.9 years (SD 1.6). Five patients had cardiac sudden death, and two underwent ICD implantation due to inducible sustained VT during an electrophysiological study that was performed due to non-sustained VT and LV ejection fraction under 40%. One of the patients who underwent ICD implantation died 22 months after placement. Those patients who presented the studied end-point during the study follow-up had larger LV fibrosis mass, larger BZ channel mass, and higher BZ channel extension in layers at baseline ( Table 2).
The results of the univariable analyses are depicted in Tables 3, 4, and 5. LV fibrosis mass was associated with the studied end-point in the univariable analysis. Age (HR=0.94; 95% CI 0.85-1.04) and sex (HR=2.8; 95% CI 0.5-15.3) were not included as adjustment variables of the multivariable models.
BZ channel extension in layers was associated with the studied end-point independent from LV ejection fraction or LV fibrosis mass, while other BZ channel characteristics (mass and number) had a borderline univariable association with the studied end-point. However, the BZ channel mass and number were not associated with the studied end-point after multivariable analysis (Tables 3, 4, and 5). LV fibrosis mass remained with a borderline association with the studied end-point after multivariable analyses. The average number of layers of the BZ channels of the patients that presented events were 3.3, 3.9, four, 6.25, and seven (two cases). No patient in whom all BZ channels had a single layer in extension presented a cardiac event.

DISCUSSION
To our knowledge, this is the first study to evaluate BZ channels and cardiac fibrosis by postprocessing ce-CMR analyses using ADAS-VT software in patients with CHD. This software performs a 3D analysis of the scar and BZ, allowing the identification of the conducting channels corresponding to the electrophysiological substrate of the VT isthmus 10 . This information can be handy to the electrophysiologist during catheter ablation procedures by facilitating the identification of the ideal location for the ablation. Furthermore, the BZ channel extension in layers was associated with the occurrence of cardiovascular events independent of LV fibrosis mass and ejection fraction.

Scar Architecture and BZ Channel
The architectural scar pattern in patients with coronary artery disease is pyramidal, with the subendocardial area being more extensive than the subepicardial area and a more significant number of BZ channels having a subendocardial location. Moreover, the BZ channels with a subepicardial location correspond to 35% of the BZ channels in transmural scar zones 10 . In our study, the fibrosis pattern of most patients was focal and midwall, as described previously in patients with earlier stages of CHD 28 . Although the BZ channel mass and number reported by us in CHD is within the range previously published in other cardiac diseases 14,15 , the BZ channel pattern differed in CHD as almost half of the BZ channels were confined to a single layer, and most were identified in the mesocardium rather than in a subendocardial location. This information is crucial for radiofrequency ablation planning.

Clinical Implications
As sudden cardiac death (SCD) is the primary mode of death in CHD 16 , strategies to treat malignant ventricular arrhythmias may increase survival in these patients. This increase in survival can be even more significant in patients with earlier-stage CHD since they do not present with HF. Moreover, as the proportion of patients with earlier-stage CHD is higher than that of patients with CHD and HF, most episodes of SCD occur among the former 16 . ICD implantation would increase survival in CHD and is recommended for secondary prevention 17 . However, the longterm benefit of ICD use for primary or even secondary prevention remains controversial since no randomized controlled trial to date has tested ICD against pharmacological treatment. One of the reasons for this controversy is the high mortality that patients with CHD present despite ICD implantation 18,[29][30][31] . However, there is great heterogeneity among mortality rates reported after ICD implantation, possibly due to discrepancies in study populations, concomitant use of amiodarone and carvedilol, and ICD programming characteristics. Nevertheless, Gali et al. compared patients treated with ICD plus amiodarone against a historical control of patients treated with only amiodarone for secondary VT prevention and described a decrease in 72% of the risk of all-cause mortality and of 95% in SCD among those treated with ICD implantation 32 . The knowledge gained with BZ and 3D scar characterization may aid in the identification of patients with higher SCD risk among those at earlier stages of CHD and change the balance toward an absolute survival benefit among patients selected for ICD implantation. Even in a small sample of patients, our results indicate that the BZ channel extension is a potential new risk factor for SCD in patients with CHD that deserves further studies in larger populations. Similarly, BZ channel extension, characterized by its transmural extension, was an independent predictor of post-ablation VT recurrence in patients with ischemic and non-ischemic cardiopathy, while BZ channel mass or number were not 13 . The extension of the BZ channel may be an important arrhythmic predictor as BZ conductive channels are a fundamental component of the reentrant circuit of ventricular arrhythmias. BZ channel mass was also associated with arrhythmic events in postmyocardial infarction patients 12 and in patients with ischemic and non-ischemic cardiomyopathy 14 . Therefore, ce-CMR allows the analysis not only of the scar but also of the BZ conductive channels and may improve SCD prediction in CHD.
One fact that could contribute to the mortality of patients with CHD despite ICD implantation is the large number of inappropriate versus appropriate shocks. In fact, among patients with CHD and ICD for secondary prophylaxis, those who received more shocks within 30 days had higher cardiac mortality rates 18 . The mapping of BZ channels may aid in the identification of patients at higher risk of malignant arrhythmic events who could benefit from VT ablation. The mapping of BZ channels may also help plan VT ablation, as the ablation technique depends on whether it has to be done via an endocardial or epicardial approach.

Limitations
The small sample of patients and clinical events included in the current study limit the conclusions. The small number of events precluded the inclusion of other relevant adjustment variables, such as age, sex, and medications used, and may cause overfitting. The small sample size also did not allow the necessary statistical power to imply conclusions over variables not associated with the combined outcome. The BZ mass was not analyzed in this study. This variable was described by others as a predictor of cardiovascular outcomes in other clinical settings 11,14 and could have been included among the adjusting variables. Another limitation is that Holter monitoring exams were not obtained at baseline in all patients. However, additional Holter exams were obtained during follow-up whenever patients presented with symptoms of syncope or near syncope. Regardless, the findings of our survival analysis support the hypothesis that BZ channel architecture may provide additional predictive value to SCD prediction models.

CONCLUSIONS
In this group of CHD patients, ce-CMR with ADAS-VT allowed the identification of BZ channels as well as their characteristics, precise location, and trajectory. Moreover, BZ channel extension was independently associated with the occurrence of clinical events. Our findings underscore the potential importance of BZ channel identification by ce-CMR for therapeutic guidance and identification of high-risk patients. Further studies are warranted in this area.