Abstract
Identification of left atrial (LA) fibrosis through late gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) remains controversial due to the heterogeneity and lack of reproducibility of proposed methods. Our aim is to describe a normalized, reproducible, standardized method to evaluate LA fibrosis through LGE-CMR.
Electrocardiogram- and respiratory-gated 3-Tesla LGE-CMR was performed in 10 healthy young volunteers and 30 patients with atrial fibrillation (AF): 10 with paroxysmal AF, 10 with persistent AF, and 10 with a previous AF ablation procedure. Local image intensity ratio (IIR) of the LA was calculated as the absolute pixel intensity to mean blood pool intensity ratio. The healthy atrial tissue threshold was defined in young healthy volunteers (upper limit of normality set at IIR tissue mean plus 2 SDs). Dense atrial scarring was characterized in patients with previous radiofrequency-induced scarring (post-AF ablation patients). Validation groups consisted of patients with paroxysmal and persistent AFs. The upper limit of normal IIR was 1.20; IIR values higher than 1.32 (60% of mean maximum pixel intensity in post-ablation patients) were considered dense scar. Image intensity ratio values between 1.2 and 1.32 identified interstitial fibrosis. Patients with paroxysmal and persistent AFs had less atrial fibrotic tissue compared with post-ablation patients. Endocardial bipolar voltage was correlated to IIR values.
An IIR of 1.2 identifies the upper limit of normality in healthy young individuals. An IIR of >1.32 defines dense atrial fibrosis in post-ablation patients. Our results provide a consistent, comparable, and normalized tool to assess atrial arrhythmogenic substrate.
We establish standardized, reproducible late gadolinium-enhanced cardiac magnetic resonance thresholds based on healthy volunteers and post-ablation patients: an image intensity ratio (IIR) of ≤1.20 identifies normal atrial tissue, an IIR between >1.20 and ≤1.32 identifies interstitial fibrosis, and an IIR of <1.32 identifies dense scarring.
Future research should aim at confirming the clinical validity of our results and test its applicability as a tool in primary prevention, establishing recurrence prognosis and used to guide atrial fibrillation ablation procedures, fostering a personalized approach.
Introduction
Left atrial (LA) structural and functional adaptation to haemodynamic overload (so-called atrial remodelling) contributes to the onset, progression, and perpetuation of atrial fibrillation (AF). In addition, LA remodelling develops as a consequence of repeated episodes of AF, leading to the well-known refrain, ‘AF begets AF’.
Collagen deposition in the myocardial interstitial space is a hallmark of LA structural remodelling.1 Total atrial fibrosis might appear as interstitial fibrosis or dense scarring.1 Fibrosis has been shown in patients with AF but no structural heart disease (lone AF), but is more intense in patients with structural cardiac disease.2
Non-invasive assessment of myocardial fibrosis has proved useful as a diagnostic, prognostic, and therapeutic tool. Gadolinium is a paramagnetic metal that accumulates in the extracellular space of the myocardium and modifies magnetic properties of water. Visualization and quantification of gadolinium in late gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) sequences estimate the extracellular matrix volume and have been used as a myocardial fibrosis surrogate.3
In the left ventricle, LGE-CMR identifies those patients at an increased risk of sudden death, accurately characterizes ventricular myocardial scar, and has been successfully used to support ventricular tachycardia ablation procedures.4 Nevertheless, a reliable and reproducible method to locate and quantify myocardial fibrosis in the atrium is still lacking.5 In recent years, several groups tested the ability of LGE-CMR to detect both pre-existing6,7 fibrosis and post-ablation scarring.8 Although these reports suggested that the extent of fibrosis may predict recurrences after ablation procedures, the lack of reference values for normality has prompted the publication of several image acquisition and post-processing protocols and thresholds to identify fibrosis, eventually limiting the external validation and reproducibility of this technique. Thereby, despite promising findings, the assessment of LA fibrosis has not yet been widely adopted in the clinical practice.
The aims of our study were (1) to establish LA LGE-CMR signal intensity normality in a young healthy population and (2) to provide a normalized, systematic, consistent, and reproducible method to identify LA fibrotic tissue.
Methods
A detailed version of the methods is available in the Supplementary material online.
Sample population
Ten young healthy volunteers (aged 18–30 years with no previous cardiac or other conditions) and 30 patients with a previous diagnosis of AF (10 patients with paroxysmal AF, 10 patients with persistent AF, 10 patients who had previously undergone an AF ablation procedure) were included in this study. Exclusion criteria included claustrophobia, major renal impairment (glomerular filtration rate < 30 mL/min), and gadolinium allergy.
Late gadolinium-enhanced cardiac magnetic resonance exams were obtained at baseline in healthy volunteers, shortly (<2 weeks) before an AF ablation procedure in paroxysmal or persistent AF patients, and 3 months (±1 week) after a first radiofrequency AF ablation procedure in patients with previous AF ablation. Healthy volunteers and post-ablation patients were used to establish LGE-CMR signal intensity normality and fibrosis thresholds; paroxysmal and persistent AF patients served as validation groups.
Written informed consent was obtained from all participants. The study protocol was reviewed and approved by the Hospital Clinic Ethics Committee.
Image acquisition and late gadolinium-enhanced cardiac magnetic resonance post-processing
The acquisition protocol has been previously reported8 and is extensively described in Supplementary material online, Methods. Briefly, images were acquired 20 min after an intravenous bolus injection of 0.2 mmol/kg gadobutrol (Gadovist, BayerShering, Germany) in a 3-Tesla CMR scanner (Magnetom Trio, Siemens Healthcare, Germany).
Left atrial was segmented using ADAS® image post-processing software (Galgo Medical SL, Barcelona, Spain). Epicardial and endocardial LA wall contours were manually drawn in each axial plane. In order to minimize endocardial and epicardial segmentation artefacts, ADAS constructed a mid-myocardial (50% thickness) layer and built a 3D shell. Pulmonary veins at their ostia and the mitral valve were removed for fibrosis analysis. Pixel signal intensity maps were calculated and projected into the shell (Figure 1).
Segmentation process of a healthy volunteer. (1) Original 3D LGE-CMR image. (2) Manually drawn epicardial and endocardial contours. (3) Deformation of 50% layer. (4) Pulmonary veins and mitral withdrawal. (5) Three-dimensional colour-coded LGE-CMR shell. (5a) Posterior view. (5b) Anterior view.
Signal intensity was normalized to blood pool intensity. Image intensity ratio (IIR)9 was calculated as the ratio between the signal intensity of each pixel and the mean blood pool intensity. Image intensity ratio values were colour coded, projected into the atrial mid-myocardial shell, and presented in histograms.
Normality and assessment of atrial myocardial fibrosis threshold
Two threshold points were sought in IIR histograms: the first delimited an upper limit of healthy tissue signal intensity, and the second discriminated between interstitial and dense fibrosis. Both threshold points were thereafter used to quantify total and dense atrial fibrosis in all groups.
Normal LA IIR values were characterized in healthy volunteers. All IIR values were plotted in a histogram, and the upper limit of normality was set at 2 SDs above the mean (mean + 2 SD, encompassing 97.5% of all signal intensity values of a healthy population).
Dense scarring was characterized in individuals who had previously undergone a pulmonary vein isolation procedure. The ablation procedure in these patients had been conducted according to our centre's usual practice. The target of ablation (paroxysmal and persistent AFs) was pulmonary vein isolation; additional lines were performed in persistent AF patients at the discretion of the treating electrophysiologist. Dense fibrosis was defined as those IIR values above the 60% of the maximum pixel intensity (MPI) in patients who had previously undergone an AF ablation procedure, as previously validated to predict conduction gaps in re-do patients.8
A subset of 14 randomly selected LGE-CMR scans were analysed by a second investigator to assess fibrosis assessment inter-observer reproducibility.
Late gadolinium-enhanced cardiac magnetic resonance intensity and electroanatomic map correlation
In 15 patients undergoing a first pulmonary vein isolation procedure, a point-by-point electroanatomic bipolar voltage map (EAM) (CARTO® 3, Biosense-Webster) was obtained with a multipolar catheter (Lasso NAV®, Biosense-Webster; interelectrode spacing 2–6–2 mm) before ablation. The EAM was merged with the 50%-layer LGE-CMR LA shell, and their correlation was assessed. Only EAM and CMR shell points that were <2 mm apart were used for the correlation analysis.9
Statistical analysis
Continuous variables are shown as mean ± standard deviation (SD) or median (25th–75th quartiles) and compared with a t-test/Mann–Whitney test or one-way ANOVA/Kruskal–Wallis test. Categorical variables are summarized as total number (percentage) and compared with a Fisher's exact test. The correlation between atrial EAM and IIR was fitted in a generalized linear mixed model. Inter-observer concordance was analysed using the Lin correlation coefficient. A two-sided type I error of 5% was used for all tests. All analyses were performed using R v3.2.0 (R project for Statistical Computing; Vienna, Austria).
Results
Baseline characteristics of the 10 healthy volunteers and 30 AF patients are shown in Table 1. Mean age was 22 years for healthy volunteers and 58 ± 10 years for patients with AF. Hypertension was the only factor significantly differing in the four groups.
Patient demographics and echocardiographic parameters
| Healthy volunteers | Paroxysmal AF | Persistent AF | Post-ablation | P | |
|---|---|---|---|---|---|
| Age (years) | 22 ± 0 | 60 ± 12 | 57 ± 8 | 56 ± 12 | <0.001 |
| Men | 5 (50%) | 9 (90%) | 9 (90%) | 8 (80%) | 0.12 |
| Hypertension | 0 (0%) | 3 (30%) | 5 (50%) | 7 (70%) | 0.01 |
| Diabetes mellitus | 0 (0%) | 0 (0%) | 2 (20%) | 1 (10%) | 0.27 |
| OSA | 0 (0%) | 1 (10%) | 1 (10%) | 0 (0%) | 0.55 |
| Structural heart disease | 0 (0%) | 1 (10%) | 1 (10%) | 0 (0%) | 0.41 |
| Echocardiography data | |||||
| LA AP diameter (mm) | _ | 38 ± 4 | 44 ± 5 | 39 ± 4 | 0.03 |
| LV ejection fraction (%) | _ | 59.4 ± 65.5 | 59.3 ± 32.9 | 60.5 ± 43.6 | 0.82 |
| Healthy volunteers | Paroxysmal AF | Persistent AF | Post-ablation | P | |
|---|---|---|---|---|---|
| Age (years) | 22 ± 0 | 60 ± 12 | 57 ± 8 | 56 ± 12 | <0.001 |
| Men | 5 (50%) | 9 (90%) | 9 (90%) | 8 (80%) | 0.12 |
| Hypertension | 0 (0%) | 3 (30%) | 5 (50%) | 7 (70%) | 0.01 |
| Diabetes mellitus | 0 (0%) | 0 (0%) | 2 (20%) | 1 (10%) | 0.27 |
| OSA | 0 (0%) | 1 (10%) | 1 (10%) | 0 (0%) | 0.55 |
| Structural heart disease | 0 (0%) | 1 (10%) | 1 (10%) | 0 (0%) | 0.41 |
| Echocardiography data | |||||
| LA AP diameter (mm) | _ | 38 ± 4 | 44 ± 5 | 39 ± 4 | 0.03 |
| LV ejection fraction (%) | _ | 59.4 ± 65.5 | 59.3 ± 32.9 | 60.5 ± 43.6 | 0.82 |
AF, atrial fibrillation; OSA, obstructive sleep apnoea; LA, left atrial; AP, anteroposterior; LV, left ventricle.
Patient demographics and echocardiographic parameters
| Healthy volunteers | Paroxysmal AF | Persistent AF | Post-ablation | P | |
|---|---|---|---|---|---|
| Age (years) | 22 ± 0 | 60 ± 12 | 57 ± 8 | 56 ± 12 | <0.001 |
| Men | 5 (50%) | 9 (90%) | 9 (90%) | 8 (80%) | 0.12 |
| Hypertension | 0 (0%) | 3 (30%) | 5 (50%) | 7 (70%) | 0.01 |
| Diabetes mellitus | 0 (0%) | 0 (0%) | 2 (20%) | 1 (10%) | 0.27 |
| OSA | 0 (0%) | 1 (10%) | 1 (10%) | 0 (0%) | 0.55 |
| Structural heart disease | 0 (0%) | 1 (10%) | 1 (10%) | 0 (0%) | 0.41 |
| Echocardiography data | |||||
| LA AP diameter (mm) | _ | 38 ± 4 | 44 ± 5 | 39 ± 4 | 0.03 |
| LV ejection fraction (%) | _ | 59.4 ± 65.5 | 59.3 ± 32.9 | 60.5 ± 43.6 | 0.82 |
| Healthy volunteers | Paroxysmal AF | Persistent AF | Post-ablation | P | |
|---|---|---|---|---|---|
| Age (years) | 22 ± 0 | 60 ± 12 | 57 ± 8 | 56 ± 12 | <0.001 |
| Men | 5 (50%) | 9 (90%) | 9 (90%) | 8 (80%) | 0.12 |
| Hypertension | 0 (0%) | 3 (30%) | 5 (50%) | 7 (70%) | 0.01 |
| Diabetes mellitus | 0 (0%) | 0 (0%) | 2 (20%) | 1 (10%) | 0.27 |
| OSA | 0 (0%) | 1 (10%) | 1 (10%) | 0 (0%) | 0.55 |
| Structural heart disease | 0 (0%) | 1 (10%) | 1 (10%) | 0 (0%) | 0.41 |
| Echocardiography data | |||||
| LA AP diameter (mm) | _ | 38 ± 4 | 44 ± 5 | 39 ± 4 | 0.03 |
| LV ejection fraction (%) | _ | 59.4 ± 65.5 | 59.3 ± 32.9 | 60.5 ± 43.6 | 0.82 |
AF, atrial fibrillation; OSA, obstructive sleep apnoea; LA, left atrial; AP, anteroposterior; LV, left ventricle.
Echocardiographic (Table 1) and standard CMR measurements (Table 2) showed no differences between groups for left ventricular ejection fraction or diameters. In CMR, LA volume progressively increased from healthy volunteers to patients with paroxysmal AF to patients with persistent AF. Post-ablation patients showed smaller volumes than AFpx and AFps patients who had not yet undergone ablation.
Standard CMR and post-processed LGE-CMR results
| CMR data | Healthy volunteers | Paroxysmal AF | Persistent AF | Post-ablation | P |
|---|---|---|---|---|---|
| LVEF (%) | 59 ± 3 | 59 ± 4 | 57 ± 5 | 60 ± 5 | 0.51 |
| EDV (mL) | 161.3 ± 38.8 | 168.0 ± 23.4 | 157.3 ± 38.9 | 164.3 ± 37.5 | 0.91 |
| ESV (mL) | 67.4 ± 18.3 | 70.9 ± 12.5 | 64.9 ± 15.0 | 67.6 ± 15.5 | 0.86 |
| LA (area, cm2) | 18.6 ± 3.1 | 26.9 ± 7.0 | 30.6 ± 4.3 | 28.2 ± 4.6 | <0.001 |
| LA volume (cm3) | 32.5 ± 8.2 | 83.9 ± 31.5 | 100.7 ± 23.3 | 74.5 ± 19.8 | <0.001 |
| LA BS-indexed volume (cm3/m2) | 18.9 ± 4.36 | 45.32 ± 18.45 | 52.52 ± 10.7 | 38.2 ± 11.71 | <0.001 |
| Post-processed LGE-CMR histogram data | |||||
| Mean IIR | 0.91 ± 0.15 | 0.98 ± 0.2 | 0.97 ± 0.2 | 1.04 ± 0.3 | <0.001 |
| % Overall fibrosis (IIR >1.20) | 2.46% (1.52–4.21) | 8.53% (4.12–12.47) | 11.73% (4.62–22.5) | 34.62% (14.57–43.18) | <0.001 |
| % Dense fibrosis (IIR >1.32) | 0.02% (0.01–0.04) | 1.27% (0.51–2.21) | 1.64% (0.07–2.8) | 14.5% (4.88–22.13) | <0.001 |
| Asymmetry (skewness) | −0.75 | 0.08 | 0.22 | 0.78 | |
| CMR data | Healthy volunteers | Paroxysmal AF | Persistent AF | Post-ablation | P |
|---|---|---|---|---|---|
| LVEF (%) | 59 ± 3 | 59 ± 4 | 57 ± 5 | 60 ± 5 | 0.51 |
| EDV (mL) | 161.3 ± 38.8 | 168.0 ± 23.4 | 157.3 ± 38.9 | 164.3 ± 37.5 | 0.91 |
| ESV (mL) | 67.4 ± 18.3 | 70.9 ± 12.5 | 64.9 ± 15.0 | 67.6 ± 15.5 | 0.86 |
| LA (area, cm2) | 18.6 ± 3.1 | 26.9 ± 7.0 | 30.6 ± 4.3 | 28.2 ± 4.6 | <0.001 |
| LA volume (cm3) | 32.5 ± 8.2 | 83.9 ± 31.5 | 100.7 ± 23.3 | 74.5 ± 19.8 | <0.001 |
| LA BS-indexed volume (cm3/m2) | 18.9 ± 4.36 | 45.32 ± 18.45 | 52.52 ± 10.7 | 38.2 ± 11.71 | <0.001 |
| Post-processed LGE-CMR histogram data | |||||
| Mean IIR | 0.91 ± 0.15 | 0.98 ± 0.2 | 0.97 ± 0.2 | 1.04 ± 0.3 | <0.001 |
| % Overall fibrosis (IIR >1.20) | 2.46% (1.52–4.21) | 8.53% (4.12–12.47) | 11.73% (4.62–22.5) | 34.62% (14.57–43.18) | <0.001 |
| % Dense fibrosis (IIR >1.32) | 0.02% (0.01–0.04) | 1.27% (0.51–2.21) | 1.64% (0.07–2.8) | 14.5% (4.88–22.13) | <0.001 |
| Asymmetry (skewness) | −0.75 | 0.08 | 0.22 | 0.78 | |
AF, atrial fibrillation; LVEF, left ventricle ejection fraction; EDV/ESV, end diastolic/systolic volume; LA, left atrial; IIR, image intensity ratio.
Standard CMR and post-processed LGE-CMR results
| CMR data | Healthy volunteers | Paroxysmal AF | Persistent AF | Post-ablation | P |
|---|---|---|---|---|---|
| LVEF (%) | 59 ± 3 | 59 ± 4 | 57 ± 5 | 60 ± 5 | 0.51 |
| EDV (mL) | 161.3 ± 38.8 | 168.0 ± 23.4 | 157.3 ± 38.9 | 164.3 ± 37.5 | 0.91 |
| ESV (mL) | 67.4 ± 18.3 | 70.9 ± 12.5 | 64.9 ± 15.0 | 67.6 ± 15.5 | 0.86 |
| LA (area, cm2) | 18.6 ± 3.1 | 26.9 ± 7.0 | 30.6 ± 4.3 | 28.2 ± 4.6 | <0.001 |
| LA volume (cm3) | 32.5 ± 8.2 | 83.9 ± 31.5 | 100.7 ± 23.3 | 74.5 ± 19.8 | <0.001 |
| LA BS-indexed volume (cm3/m2) | 18.9 ± 4.36 | 45.32 ± 18.45 | 52.52 ± 10.7 | 38.2 ± 11.71 | <0.001 |
| Post-processed LGE-CMR histogram data | |||||
| Mean IIR | 0.91 ± 0.15 | 0.98 ± 0.2 | 0.97 ± 0.2 | 1.04 ± 0.3 | <0.001 |
| % Overall fibrosis (IIR >1.20) | 2.46% (1.52–4.21) | 8.53% (4.12–12.47) | 11.73% (4.62–22.5) | 34.62% (14.57–43.18) | <0.001 |
| % Dense fibrosis (IIR >1.32) | 0.02% (0.01–0.04) | 1.27% (0.51–2.21) | 1.64% (0.07–2.8) | 14.5% (4.88–22.13) | <0.001 |
| Asymmetry (skewness) | −0.75 | 0.08 | 0.22 | 0.78 | |
| CMR data | Healthy volunteers | Paroxysmal AF | Persistent AF | Post-ablation | P |
|---|---|---|---|---|---|
| LVEF (%) | 59 ± 3 | 59 ± 4 | 57 ± 5 | 60 ± 5 | 0.51 |
| EDV (mL) | 161.3 ± 38.8 | 168.0 ± 23.4 | 157.3 ± 38.9 | 164.3 ± 37.5 | 0.91 |
| ESV (mL) | 67.4 ± 18.3 | 70.9 ± 12.5 | 64.9 ± 15.0 | 67.6 ± 15.5 | 0.86 |
| LA (area, cm2) | 18.6 ± 3.1 | 26.9 ± 7.0 | 30.6 ± 4.3 | 28.2 ± 4.6 | <0.001 |
| LA volume (cm3) | 32.5 ± 8.2 | 83.9 ± 31.5 | 100.7 ± 23.3 | 74.5 ± 19.8 | <0.001 |
| LA BS-indexed volume (cm3/m2) | 18.9 ± 4.36 | 45.32 ± 18.45 | 52.52 ± 10.7 | 38.2 ± 11.71 | <0.001 |
| Post-processed LGE-CMR histogram data | |||||
| Mean IIR | 0.91 ± 0.15 | 0.98 ± 0.2 | 0.97 ± 0.2 | 1.04 ± 0.3 | <0.001 |
| % Overall fibrosis (IIR >1.20) | 2.46% (1.52–4.21) | 8.53% (4.12–12.47) | 11.73% (4.62–22.5) | 34.62% (14.57–43.18) | <0.001 |
| % Dense fibrosis (IIR >1.32) | 0.02% (0.01–0.04) | 1.27% (0.51–2.21) | 1.64% (0.07–2.8) | 14.5% (4.88–22.13) | <0.001 |
| Asymmetry (skewness) | −0.75 | 0.08 | 0.22 | 0.78 | |
AF, atrial fibrillation; LVEF, left ventricle ejection fraction; EDV/ESV, end diastolic/systolic volume; LA, left atrial; IIR, image intensity ratio.
Late gadolinium-enhanced cardiac magnetic resonance image post-processing and fibrosis analysis
Late gadolinium-enhanced cardiac magnetic resonance was analysed in all individuals, except for one healthy volunteer who was excluded because of multiple artefacts and poor image quality. A total of 901 390 IIR values were obtained (averaging 23 113 ± 12 137 points per patient). Four IIR histograms including all individuals in the same group were generated: healthy volunteers (HV) and paroxysmal AF (AFpx), persistent AF (AFpt), and post-ablation (P-Ab) patients.
The histograms for all groups are shown in Figure 2. The morphology of the four LA IIR histograms was significantly different (P < 0.001 for all pairwise comparisons, multiple-comparison adjusted Kolmogorov–Smirnov test). The mean IIR for each group showed a progressive increase from healthy volunteers to post-ablation patients (Table 2).
Histogram distribution (IIR) superimposed in all groups.
The skewness statistic assesses the symmetry of a distribution, with 0 denoting perfect symmetry, positive values a right-tailed distribution, and negative values a left-tailed distribution. Skewness ranged from −0.75 (left-tailed to ‘healthy’ tissue values) for HV to +0.78 (right-tailed to fibrotic tissue values) in P-Ab. Patients with AFpx and AFpt remained in intermediate, slightly right-tailed skewness values at 0.08 and 0.22, respectively.
Threshold assessment
Normal IIR values were identified in a population of healthy individuals. The upper limit of IIR normality was accordingly set at 1.20 (mean tissue IIR + 2 SD) (Figure 3A) and was similar in male and female individuals (1.20 ± 0.04 vs. 1.21 ± 0.09, P = 0.53). Hence, all IIR values >1.20 were considered as fibrotic tissue.
(A) Left panel: Histogram distribution (IIR) in all groups. Red line: Dense fibrosis threshold IIR. Green line: Global fibrosis threshold IIR. Right panel: Examples of posteroanterior and anteroposterior views in each group. (B) Fibrosis quantification in each patient.
Dense scarring was characterized in P-Ab patients as previously described.8 The MPI in the whole atria was recognized and its 60% calculated. All IIR values above 1.32 (60% of MPI) were therefore considered to localize dense scarring.
From our results, we propose that IIR values between 1.2 and 1.32 identify interstitial fibrosis, while an IIR of >1.32 involves dense scarring. We therefore quantified the percentage of fibrosis (IIR > 1.20) in all participants, which progressively increased from healthy volunteers to post-ablation patients [2.46% (1.52–4.21) HV, 8.53% (4.12–12.47) AFpx, 11.73% (4.62–22.57) AFpt, 34.62% (14.57–43.18) P-Ab, P < 0.001] (Figure 3B). Similarly, atrial dense scar percentage was 0.02% (0.01–0.04) for HV, 1.27% (0.51–2.21) for AFpx, 1.64% (0.07–2.8) for AFpt, and 14.5% (4.88–22.13) for P-Ab patients (P < 0.001) (Figure 3B). Data for male and female individuals are shown in the Supplementary material online, Figure.
The inter-observer Lin concordance correlation coefficient for mean IIR, IIR + 2 SD, and global fibrosis percentage were 0.948, 0.987, and 0.998, respectively.
Electroanatomic voltage map and image intensity ratio correlation
The point-by-point correlation between the atrial EAM and IIR shells was assessed in 15 patients (10 persistent AF, 5 paroxysmal AF) undergoing AF ablation after CMR evaluation. Overall, 1729 valid voltage points were obtained (124 ± 81 points per patient). The correlation plot for each of the individuals is shown in Figure 4. A negative correlation between EAM voltage and IIR was found for all individuals, generally of a moderate intensity; correlation coefficients (r) ranged from −0.19 to −0.58 (r2 from 0.036 to 0.34). When all individuals were modelled with GLMM, r was estimated at 0.2 (P < 0.001). Similarly, a negative correlation was obtained after categorizing IIR data into clinically meaningful groups (<1, 1–1.20, 1.20–1.32, >1.32) (Figure 5). Mean IIR values were 1.45V ± 1.51, 1.07V ± 1.39, 0.94V ± 1.16, and 0.72V ± 0.89, respectively (P < 0.001).
Individual patient point-by-point atrial correlation with IIR bipolar voltage (log-transformed).
(A) Representative example of an electroanatomic map and IIR LGE-CMR shell merge. (B) Voltage distributions in the threshold IIR cut-offs in all patients.
Discussion
In this study, we provide a normality range for atrial LGE-CMR signal intensity in young healthy individuals and propose a reproducible, normalized method to assess total and dense LA fibrosis. Specifically, we define the IIR cut-off point of >1.20 for identification of abnormal signal intensity as the threshold for atrial fibrosis, while the IIR cut-off point of >1.32 is the threshold for dense atrial fibrosis. To our knowledge, this is the first study that uses a healthy volunteer population to characterize a LGE-CMR signal intensity threshold in the LA to identify atrial fibrosis.
Atrial fibrosis assessment
In the atrium, total fibrosis might result from interstitial reactive fibrosis or from confluent replacement fibrosis after myocyte apoptosis or necrosis.1 Our work provided a threshold to identify both total and dense fibrosis.
Atrial fibrosis assessment: native fibrosis
Several groups have proposed a variety of algorithms to identify native fibrosis, but most of them remain controversial. While most reports agree that a higher signal intensity correlates to a larger amount of fibrosis and a worse recurrence prognosis after AF ablation, a threshold has not been uniformly established.
The most widely validated algorithm was published by the Marrouche group6 and is currently supported by a growing core of evidence.7 Rather than using a fixed algorithm, their method largely relies on an expert decision to use a certain, variable threshold ranging from 2 to 4 SDs above the mean for healthy myocardium intensity. The need for this expertise hampers external validation, with inconsistent findings.10
Most algorithms rely on a certain number of SDs over mean atrial signal intensity to identify fibrosis, likely under- or over-estimating atrial fibrosis.11 Indeed, we have shown that pixel intensity histogram morphology is deeply influenced by patient characteristics (Figure 2), which might lead to inaccurate changes in fibrosis quantification due to differences in the mean and SD between healthy individuals and post-ablation patients.
Khurram et al. first proposed a normalized signal intensity ratio (IIR) to homogenize CMR from all individuals and establish an absolute fibrosis threshold that could be used in all patients.9 Nevertheless, a normality threshold (IIR < 0.97) was set after correlating to bipolar voltage maps in patients with AF. Most groups have exclusively relied on CMR images from patients with AF; the lack of a healthy volunteer group poses a risk of inaccurate identification of normality limits and the under- or over-estimation of atrial fibrosis.
A novel approach in our study incorporated a healthy population in which no fibrosis is expected, in order to obtain a standardized upper limit of normality that is able to provide an objective value beyond which fibrosis is defined and define systematic thresholds. Our healthy volunteer cohort encompasses exclusively young individuals aged 22 years, excluding the pro-fibrotic effects of aging or other co-morbidities such as hypertension. On this basis, an IIR of >1.20 identifies atrial fibrosis, independent of patient characteristics.
Notably, we used young healthy volunteers as the reference population to ensure obtaining normality values from an atrial fibrosis-free population. Ageing has been suggested to associate with progressive atrial collagen deposit, a process likely underlying an increase in AF prevalence over years.12 An elderly or middle-aged reference population, even if providing a normal-for-age threshold, would miss ageing-related fibrosis. When apparently healthy, middle-aged individuals have been used as a control group, average fibrosis burden ranged from 1.76 to 8.9%.13 Our method identifies all ageing- and risk-factor-related atrial fibrosis, both of which may contribute to AF pathology.
Atrial fibrosis assessment: dense scarring
Post-ablation scarring is characterized by large areas of coagulative necrosis and fibrotic replacement and is a hallmark of dense fibrosis. Dense fibrosis is also present in confluent areas in patients with native atrial fibrosis.1 Specific studies focusing on the identification of atrial post-ablation injury with LGE-CMR have been recently carried out, with uneven conclusions.7,14,15
On the basis of histological findings and validation in an animal model, Harrison et al. proposed that blood pool intensity mean plus a certain number of SDs could be used to identify linear radiofrequency lesions in the right atria of sheep.15 Nevertheless, this algorithm was not able to reliably predict previously ablated areas in patients undergoing a second pulmonary vein isolation ablation procedure.16 Similarly, Hunter et al. confirmed that a 5-SD threshold above the mean atrial tissue intensity likely underestimates radiofrequency-induced atrial fibrosis.11 Only 34% of all electrically isolated pulmonary veins were completely encircled by radiofrequency lesions using the Utah criteria.17
Bisbal et al. found a 79% correspondence between the electrical site of reconnection and anatomical gaps by using a value 60% of the MPI to threshold post-ablation scarring.8 Our study used the definition by Bisbal et al. to distinguish dense fibrosis. By using this criterion in our study, large fibrotic patches predominantly encircling pulmonary veins were detected in a 3-month post-ablation LGE-CMR [14.5% (4.88–22.13), Table 2], while it was negligible in young, healthy individuals [0.02% (0.01–0.04)].
Value of cardiac magnetic resonance-identified fibrosis: clinical implications
The non-invasive assessment of atrial myocardial fibrosis extent provides important data on atrial structural remodelling, which might be translated into clinically useful information for daily clinical practice, from primary prevention to prognosis and the guidance of AF therapy. Such knowledge of the arrhythmogenic state of the LA should pave the way for future AF therapy personalization.
The selection of patients at high risk of AF might allow optimization of primary prevention programmes. Hypertension is the most common AF risk factor in the community, but the positive predictive value for AF is relatively low.18 The AF risk of structural heart disease is higher (6- to 14-fold increased risk), but its prevalence in overall AF is much lower than hypertension. A set of 14 clinical and electrocardiogram markers proposed by the CHARGE consortium predicts AF incidence, but accuracy remains moderate (Area under the curve 0.66–0.71 in validation cohorts).19 The non-invasive assessment of the structural arrhythmic substrate by means of LGE-CMR might provide a more direct estimation of AF risk; those patients with extensive fibrosis and at high risk of AF might benefit from a closer rhythm follow-up, more intensive antihypertensive drugs, or even early instauration of antiarrhythmic therapies.
Atrial fibrillation outcomes after an ablation procedure remain unsatisfactory. Plasmatic biomarker profiles suggesting enhanced collagen turnover predict a higher recurrence rate after an AF ablation procedure. Non-invasive estimation of atrial fibrosis by means of LGE-CMR might be a valuable tool to improve patient selection for AF ablation.7,20 Ablation gaps assessed by LGE-CMR predict AF recurrences after an AF ablation procedure.21
The pre-procedural identification of fibrotic areas might not only serve to foresee a group of patients with an ominous recurrence prognosis but also to guide AF ablation procedures. A recent subanalysis of the DECAAF trial suggested that encircling atrial fibrosis is beneficial to prevent AF recurrences after pulmonary vein isolation.22 Identifying dense fibrotic lesions in patients undergoing repeated AF ablation procedures localizes conduction gaps surrounding pulmonary veins and simplifies re-do procedures.8 Confirmation in larger, randomized trials is warranted.
Limitations
Some limitations of our work need to be acknowledged. First, individual factors are an indisputable source of variability. Individual characteristics such as body mass index, renal function, and haematocrit might deeply change gadolinium wash-in and wash-out and limit the validity of our (and most other) algorithms. Blood pool normalization partially accounted for these potential biases in our study and allowed inter-individual comparability.
Second, technical and post-processing parameters might influence IIR estimation. Our work was conducted in a 3 T set-up as it provides a higher signal-to-noise ratio and enhanced temporal and spatial resolution in thin-walled atrium than 1.5 T set-ups. The accuracy of manual and semi-automated identification of the LA endocardial and epicardial boundaries remains a critical aspect of image post-processing protocols, largely depending on investigator experience. Nevertheless, these errors were minimized by using a mid-myocardial layer (50% of atrial wall thickness) that prevents against mild inaccuracies in boundary segmentation. Moreover, reproducibility of our technique was confirmed by a high Lin correlation coefficient. Further studies are needed to guarantee external validity in 1.5 T set-ups, different patient groups and operators experience.
Third, we describe a normality threshold value for LGE-CMR and liken it to fibrosis burden. Although LGE-CMR has been commonly used as a myocardial fibrosis surrogate, it is possible that other causes increasing extracellular volume such as oedema or infiltration could also contribute to LGE.3
Last, collagen deposition in the myocardial interstitial space is likely continuously distributed with a variable degree of overlap between HV and AF patients, thus making any threshold arbitrary. Small changes in any threshold of a continuous variable might result in large changes in the percentage of fibrosis.15 Nevertheless, we believe that finding an upper limit of normality that encompasses the 97.5% of all values in a healthy population is the more appropriate way to discriminate healthy vs. pathological tissue. As in any normality threshold, external replication is required.
Conclusion
In healthy individuals, the LGE-CMR threshold for healthy atrial tissue is IIR ≤ 1.2. Higher values identify variable degrees of fibrosis. An IIR of >1.32 identifies dense atrial scar. Our results provide a consistent, reproducible, and normalized tool to identify atrial fibrosis that might be useful for prognostic and therapeutic purposes.
Supplementary material
Supplementary material is available at Europace online.
Funding
This work was partially supported by Fondo de Investigaciones Sanitarias-Instituto de Salud Carlos III (PI13/01747), European Regional Development Fund (ERDF. European Union. A Way of Making Europe), the European Union's Horizon 2020 research and innovation programme under grant agreement No. 633196 (CATCH ME) and a grant by La MARATÓ-TV3 (Id 20152730).
Conflict of interest: D.A. was an employee of Biosense-Webster at the time that this work was conducted. L.M., JB and A.B. are shareholders of in Galgo Medical Company.
Acknowledgement
The authors thank Mrs Neus Portella for research assistance.
