This paper presents a review of the different approaches existing in the literature to detect and quantify fibrosis in contrast-enhanced magnetic resonance images of the left atrial wall. The paper provides a critical analysis of the different methods, stating their advantages and limitations, and providing detailed analysis on the possible sources of variability in the final amount of detected fibrosis coming from the use of different techniques.
Introduction
Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia, occurring in up to 2% of the general population, and is expected to become even more common due to population ageing, as its prevalence increases with advancing age.1 Atrial fibrillation is associated with a five-fold risk of stroke and a three-fold incidence of congestive heart failure, and doubles the risk of dementia and mortality.2 Catheter ablation using the pulmonary vein (PV) antrum isolation technique has emerged as a first-line therapy for patients with symptomatic AF not responding to pharmaceutical treatment.3
Although the pathophysiology of this arrhythmia is not completely understood, animal and human experimental models have shown multiple disease pathways that can promote an abnormal impulse formation and propagation. Structural, contractile, or electrical alterations are associated with the onset and the perpetuation of AF.4 Fibrosis is the hallmark of atrial structural remodelling; it results from an excessive deposition of extracellular matrix to replace degenerating atrial myocytes.5 Fibrotic tissue can be detected in left atrium (LA) of AF patients with identifiable cardiac disease and in lone AF4,6–8 and is one of the major risk factors for AF progression.9
Magnetic resonance imaging (MRI) with late gadolinium enhancement (LGE) allows the detection and quantification of fibrotic tissue by slow washout kinetics of gadolinium in diseased tissue. While ventricular fibrosis assessment has achieved excellent results with cardiac magnetic resonance (CMR) imaging,10,11 the challenge of atrial fibrosis assessment remains open because of limited image resolution and the thinness (1–2 mm) and unpredictable shape of the LA wall. However, several recent studies managed to assess the feasibility of LGE-MRI in detecting both pre-existing and post-ablation-induced atrial fibrosis and their results have had a strong scientific impact.12–16
The extent of the baseline LGE has emerged as the strongest independent predictor of AF recurrence after ablation14 and a good tool for selecting the appropriate patients and procedural strategy.17 Moreover, the extent of LA structural remodelling correlates with CHADS2 score and stroke risk.18 With regard to ablation-induced fibrosis, LGE-MRI is a powerful tool to identify LA wall injury caused by radiofrequency (RF) energy delivery in the myocardium,12 to recognize ablation line gaps12,19,20 and even to guide the ablation procedure, both in animal models21 and in humans, by integrating the CMR model into the navigation system.16,22
Despite these promising findings from a few specialized centres, the assessment of LA fibrosis has not yet been widely adopted in clinical practice because of the poor reproducibility of the results by other research groups.23–26 A detailed scrutiny of the methodologies used in the major publications on the subject can provide insight on these discrepancies.
Image acquisition
Atrial wall imaging poses a greater challenge since the wall of the LA is much thinner than the left ventricular myocardium, and, therefore, greater spatial resolution is required. Most of the studies to evaluate atrial fibrosis are performed using 1.5 or 3 T scans. In this scenario, several research groups follow similar MRI protocols.12,14,15 First, in order to assess the complex three-dimensional (3D) anatomy of PVs and LA, a single breath-hold MR angiography is acquired after an intravenous bolus of gadolinium contrast (usually 0.1–0.2 mmol/kg). After 15–30 min (this delay assures a T1 differences between blood and scar to improve image contrast), a ‘TI scout’ or ‘Look Locker’ sequence is used to determine the optimal inversion time (TI) to nullify the left ventricular myocardial signal and optimize the contrast between the bright hyper-enhancing scar tissue and the healthy myocardium. This sequence offers a series of images with variable TI obtained over two to three heart beats and allows the examiner to select visually the appropriate TI: this value matches with the image with the best contrast and differs for each patient (due to variation in contrast dose and physiological parameters). After, DE-MRI is performed, commonly, using a free-breathing 3D navigator and electrocardiographically gated inversion-recovery gradient-eco sequence applied in the axial orientation. Images can differ in slice thickness (1.5–2.5–4.0 mm) and in plane resolution ranges (between 1.2–1.6 × 1.2–1.6),12,14,23 usually with low flip angles and parallel imaging. Average scan times for these sequences are 10–15 min, depending on heart rate and breathing patterns.
Left atrial wall image intensity on DE-CMR is affected by parameters such as surface coil proximity, contrast dose, delay time of image acquisition after contrast injection, and individual characteristics such as body mass index, renal function, haematocrit, and adherence to the exam. All these technical variables in MRI acquisition can affect comparison across centres, and together with the absence of standardized image processing protocols, can lead to different (and sometimes contradictory) results.
Image processing: left atrial segmentation and fibrosis detection
The two main steps required to analyse fibrotic tissue from MRI images are segmentation of the anatomical structure of the atrial myocardium and detection of fibrotic areas inside the atrial wall.
Segmentation of the atrial wall
Clinicians or engineers typically segment the atrial wall manually, and this is the most time-consuming part of the analysis. It involves an accurate slice-by-slice 2D tracing of the endocardial and epicardial walls to confine the region of interest (ROI) to only the LA wall, avoiding the blood pool and other anatomical structures such as the aortic ring. Accuracy plays an important role here, as inclusion of outer structures in the segmented myocardium can lead to a consideration of fat or outer tissue as scar or fibrosis, and leaving a part of the myocardium outside the segmentation may preclude detection of fibrotic areas.
Methods to discriminate left atrial fibrosis in magnetic resonance imaging with late gadolinium enhancement
Once the atrial myocardium is properly isolated, we can proceed to detect fibrotic tissue, either visually or by automatic or semi-automatic thresholding techniques. Thresholding consists of applying a cut-off value to the image to distinguish viable from non-viable myocardium, based on the range of signal intensities.
Visual assessment
This method relies on visual inspection of images and manually drawn contours around non-viable myocardium. It is considered the gold standard to assess ventricular scar in the absence of histological validation,27,28 and is currently used as the reference technique to test different semi-automatic and automatic algorithms. Current MRI resolution does not allow the same accuracy in detecting scar in the thin atrial wall by direct planimetry of the scar. However, several published studies have used this approach.12,29 A major drawback of visual assessment as a technique for large studies is the subjectivity and non-repeatability of the results. Nevertheless, it was used by Peters et al.12 to elucidate the potential role of LGE in assessing and localizing LA scar in patients after AF ablation. By visually assessing the presence of post-ablation LGE on images, they found MRI enhancement in 100% of patients submitted to AF ablation procedure. When tested in the ventricular fibrosis assessment, inter-observer variability of visual approach varied widely depending on the setting of application, achieving good results in the evaluation of dense scar tissue of ischaemic cardiomyopathy, but showing very low reproducibility in the quantification of diffuse fibrosis, such as in hypertrophic cardiomyopathy.30 Anatomical and histological characteristics of the atria make this method tricky and time-consuming and led to the study and development of automatic thresholding techniques for atrial tissue classification.
Thresholding techniques
Standard deviations above reference
Histogram-based reference
(A) Bimodal distribution of pixel intensities of the LA wall. ‘Normal tissue’ is defined as the first mode of lower pixel intensities and ‘injured tissue’ at N standard deviations above the mean pixel intensity of normal tissue (modified from Jadidi et al.31). (B) Histogram of pixel intensities modified from Oakes et al.14: ‘normal’ tissue is defined as the lower region of the pixel intensity histogram, between 2 and 40% of the maximum intensity within the LA wall. The fibrotic threshold was then calculated as 2–4 SD above the mean of ‘normal’ tissue.
To assess pre-existing fibrosis, Oakes et al.14 used the same SD-above-reference-technique, but applied a ‘dynamic thresholding’ approach. ‘Normal’ tissue is defined automatically as the mean of the lower region of the pixel intensity histogram (between 2 and 40% of the maximum intensity within the LA wall). The threshold cut-off to detect the enhanced tissue is chosen manually at 2–4 SD above the mean of ‘normal’ and assessed slice-by-slice by an expert (Figure 1B). The same method was adopted by a prospective study in 15 centres in the USA, Europe, and Australia, which confirmed that delayed-enhancement MRI estimation of atrial tissue fibrosis was independently associated with likelihood of recurrent arrhythmia in patients with AF undergoing catheter ablation.32
Nulled myocardium-based reference
(A) Blood-pool ROI (in red). The mean and standard deviation were computed in the selected blood-pool region in order to use these values for the threshold computation. (B) Nulled myocardium ROI (in red). In this case, the ROI is manually segmented from the healthy ventricular myocardium. Figures generated with ADAS-AF software (Galgo Medical S.L., Barcelona, Catalonia, Spain).
Blood-pool-based reference
The blood pool was chosen as reference in Harrison et al.,34 the first group to provide a histopathological validation in animals for LGE-CMR and electroanatomic mapping (EAM) capability to perceive fibrosis; signal intensity (SI) thresholds from 0 to 15 SD above the mean blood pool were used to segment the right atrium in CMR images, and segmentations were compared with real lesion volumes. In their study, the SI thresholds that best approximated histological volumes were 2.3 SD for LGE acutely post-ablation, when the tissue shows transmural injury with coagulative necrosis, and 3.3 SD for LGE chronically post-ablation, when a transmural fibrous scar is detected.
As this method has been tested only in pigs, we cannot know if the cut-off value is reliable in human models. However, the study underlines the histological differences in scar tissue depending on the time of execution of MRI after ablation, and the importance of appropriately defining a strict threshold.
Fixed pixel-based thresholding techniques
Full width at half maximum
(A) Distribution of pixel intensities in a patient submitted to a redo procedure: the FWHM scar technique defines fibrotic tissue as the signal exceeding 50% of the maximal SI. (B) Distribution of pixel intensities in a patient submitted to a redo procedure: Bisbal et al.15 proposed two cut-off values to characterize the hyper-enhanced area as border zone or scar, using 40 ± 5 and 60 ± 5% of the maximum intensity as thresholds.
Maximum scar density percentage
Bisbal et al.15 tested the efficacy of a LGE-MRI guided approach in identifying, localizing, and ablating gaps in redo procedures. To distinguish healthy from scarred myocardium, they used SI thresholds based on previous studies by their group, but applied to ventricular myocardium scar detection.38,39 For identification of prior ablation lesions and gaps, a pixel SI-based algorithm was applied to characterize the hyper-enhanced area as border zone or scar, using 40 ± 5 and 60 ± 5% of the maximum intensity as thresholds for border zone and scar, respectively (Figure 3B). The initial threshold was set to 40 and 60%, but a 5% variation of the threshold was allowed according to an expert visual inspection of the appearing fibrosis. After manual segmentation of endocardium and epicardium, five concentric surface layers were created at 10, 25, 50, 75, and 90% (‘onion skin’ model) and a 3D shell was obtained for each layer. A pixel intensity map was projected to each shell and colour-coded to allow the visualization and localization of CMR gaps, defined as any discontinuity of scar core areas, in order to guide re-ablation procedures. Similarly to FWHM, using the percentage of maximum intensity as threshold value (maximum scar density of the patient) limits application of this method to patients who have already undergone ablation prior to MRI acquisition.
Image intensity ratio (blood-pool mean normalization)
The study validated this new measure with EAM in a cohort of 75 patients submitted to AF ablation, proving the correlation between IIR and local intracardiac bipolar LA voltages and established IIR thresholds corresponding to the commonly accepted voltage thresholds of <0.5 and <0.1 mV41,42 to identify atrial fibrosis and dense scar, respectively.
Dewire et al.43 also used this method to detect pre-existent LA fibrosis and show that the extent of LGE in patients undergoing AF ablation is associated with AF type and time. They used the threshold extrapolated from their previous study to distinguish normal tissue (IIR < 0.97), abnormal myocardium (0.97 < IIR < 1.61), and dense scar (IIR > 1.61). Fukumoto et al.44 recently applied the IIR measure in pre- and post-procedural scans of 20 patients submitted to AF ablation in order to define the characteristics of pre-existing vs. ablation-induced LGE of the LA.
Image intensity ratio seems a good normalization strategy for fibrosis detection, but the choice of the correct IIR threshold to separate fibrotic tissue from healthy tissue is still an open question. Moreover, a multi-centre study that validates this measure using various scanners and contrast agents has never been performed, so it is not yet clear whether it is possible to have a universal IIR threshold or if every centre should find its optimal threshold.
Controversies and limitations
The main limit to clinical application of the fibrosis assessment methods is the poor reproducibility of the results in different centres. The prevalence of baseline LGE in patients submitted to a first AF ablation differs significantly across the published studies, and even across those conducted by a single group, varying from 8.7%13 of the recruited patients with mild MRI enhancement to 4714 with at least moderate enhancement (>15 of LA area).
The RF-induced fibrosis has been identified with visual methods, reporting 100% sensitivity and specificity12; other authors, using computational methods, found that the reliability of CMR in assessing the presence and distribution of ablation lesions ranged from as low as 28 to 54%.23
Nevertheless, not every study that can detect post-ablation scar can also detect gaps. While Bisbal et al.15 observed 94% electrical-CMR concordance using a 3 T scanner and a fixed pixel-based thresholding, other authors29 who visually assessed the scar by 1.5 T scanner images found no association between scar gaps and PV reconnection sites. On the other hand, Badger et al.,19 using a 1.5 T scanner and processing the image with computational methods, described good visualization of gap lesions in redo procedures, which were correlated with recovery of local electrograms or PV electric conduction.
Contradictory results were obtained by studying the correlation between the AF clinical type, paroxysmal or persistent, and the degree of structural remodelling. Contrary to previous traditional theories, some studies32,45–47 found weak or no link between the temporal pattern of AF and structural atrial changes. In DECAAF32 study, the number of patients with paroxysmal AF was surprisingly highest in the group with the most extensive fibrosis (79.2% paroxysmal AF in Grade 4 fibrosis vs. 65.3% in Grade 1); Kuppahally et al.45 did not find difference in the number of patients with paroxysmal and persistent AF when they were classified as mild and moderate–severe LA structural remodelling. Conversely, for Dewire et al.,43 LA fibrosis extent was positively associated with time in continuous AF and AF type (12.4 ± 1.8% scar in persistent and long-standing persistent AF vs. 8.2 ± 0.9% in paroxysmal AF, P= 0.028), while the extent of atrial fibrosis was independent of LA volume.
Numerous potential sources of diversion can be found in image acquisition and processing, and several challenges remain in the attempt to achieve more consistent analysis among different operators.
Image acquisition
The lack of standardized image acquisition protocols for cardiac MRI can affect the reproducibility of the results by different operators and the comparison between patients, altering the accuracy of the analysis. For example, there is no consensus on the choice and dose of contrast agent, nor on the timing in acquisition of LGE sequences after contrast administration. The choice of the optimal TI for the LGE sequence is performed manually by a technician. This is a crucial step, as it can generate the appearance of more or less scar and so an over- or underestimation of the fibrosis. Although scant data are available about the difference between atrial and ventricular myocardium in the kinetics of gadolinium, we can assume, considering the differences in their histology pattern, that the optimal TI of both tissues cannot coincide, generating possible approximation error in the image analysis. Artefacts can be generated from motion blurring, respiratory compensation, and arrhythmias. Prolonged scan time using free-breathing acquisitions, combined with temporal filtering or parallel reconstruction, can improve the image quality as much as a cardioversion before the MRI exam and is, therefore, recommended. Nevertheless, not all the studies were performed under these circumstances.
In fact, a 3 T scanner, compared with a 1.5 T scanner, increases the signal-to-noise ratio and blood-to-myocardium contrast-to-noise ratio, enables better fat suppression, and shortens the scan time. Nevertheless, 3 T scanners also have disadvantages: the increased main field strength causes an increase in the effect of magnetic susceptibility artefacts and of chemical shift artefacts, in RF field inhomogeneity, and in specific absorption rate. Table 1 summarizes the main image acquisition parameters for DE-MRI sequence used for atrial fibrosis assessment reported in previous studies.
Image acquisition parameters for DE-MRI sequence used for atrial fibrosis assessment reported in previous studies
| Study | Scanner (Tesla) | In-plane resolution (mm) | TR/TE (ms) | Contrast agent | Dose of contrast agent | Acquisition timing (min) | TI (ms) |
|---|---|---|---|---|---|---|---|
| Peters et al.12 | 1.5 | 1.3 × 1.3 | 4.3/2.1 | Gadopentetate dimeglumine | 0.2 mmol/kg | 15–25 | 280 (average) |
| McGann et al.13 | 1.5 | 1.25 × 1.25 | 6.3/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 230–270 |
| Oakes et al.14 | 1.5 | 1.25 × 1.25 | 6.3/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 230–320 |
| Bisbal et al.15 | 3 | 1.25 × 1.25 | 2.3/1.4 | Gadobutrol | 0.2 mmol/kg | 25–30 | 280–380 |
| Badger et al.19 | 1.5 | 1.25 × 1.25 | 5.5/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 270–310 |
| Hunter et al.23 | 1.5 | 1.3 × 1.3 | nr | Gadolinium-diethylenetriaminepentaacetate | (0.4 mL/kg) - 20 mL | 20 | nr |
| Sramko et al.25 | 1.5 | 1.6 × 1.6 | 4.8/1.5 | Gadobutrol | 0.2 mmol/kg | 11 ± 4 | 270 |
| DECAAF32 | 1.5 and 3 | 1.25 × 1.25 | 5.2/2.4 (1,5T)3.1/1.4 (3T) | Gadoterate meglumine; gadobenate dimeglumine; gadobutrol; gadodiamide | 0.1–0.2 mmol/kg | 15 | nr |
| Harrison et al.34 | 1.5 | 1.3 × 1.3 | 3.0/6.2 | Gadobutrol | 0.2 mL/kg | 20 | nr |
| Hwang et al.37 | 3 | 1.5 × 1.5 | 4.7/1.4 | Gadoterate meglumine | 0.2 mmol/kg | 15–25 | 230–270 |
| Khurram et al.40 | 1.5 | 1.3 × 1.3 | 3.8/1.52 | Gadopentetate dimeglumine | 0.2 mmol/kg | 15–25 | 240–290 |
| Dewire et al.43 | 1.5 | 1.3 × 1.3 | 2.5–5.5/1.52 | Gadopentetate dimeglumine | 0.2 mmol/kg | 18 | 240–290 |
| Malcolme-Lawes et al.35 | 1.5 | 1.5 × 1.5 | nr | Gadobenate dimeglumine | 20 mL | 12–20 | nr |
| Study | Scanner (Tesla) | In-plane resolution (mm) | TR/TE (ms) | Contrast agent | Dose of contrast agent | Acquisition timing (min) | TI (ms) |
|---|---|---|---|---|---|---|---|
| Peters et al.12 | 1.5 | 1.3 × 1.3 | 4.3/2.1 | Gadopentetate dimeglumine | 0.2 mmol/kg | 15–25 | 280 (average) |
| McGann et al.13 | 1.5 | 1.25 × 1.25 | 6.3/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 230–270 |
| Oakes et al.14 | 1.5 | 1.25 × 1.25 | 6.3/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 230–320 |
| Bisbal et al.15 | 3 | 1.25 × 1.25 | 2.3/1.4 | Gadobutrol | 0.2 mmol/kg | 25–30 | 280–380 |
| Badger et al.19 | 1.5 | 1.25 × 1.25 | 5.5/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 270–310 |
| Hunter et al.23 | 1.5 | 1.3 × 1.3 | nr | Gadolinium-diethylenetriaminepentaacetate | (0.4 mL/kg) - 20 mL | 20 | nr |
| Sramko et al.25 | 1.5 | 1.6 × 1.6 | 4.8/1.5 | Gadobutrol | 0.2 mmol/kg | 11 ± 4 | 270 |
| DECAAF32 | 1.5 and 3 | 1.25 × 1.25 | 5.2/2.4 (1,5T)3.1/1.4 (3T) | Gadoterate meglumine; gadobenate dimeglumine; gadobutrol; gadodiamide | 0.1–0.2 mmol/kg | 15 | nr |
| Harrison et al.34 | 1.5 | 1.3 × 1.3 | 3.0/6.2 | Gadobutrol | 0.2 mL/kg | 20 | nr |
| Hwang et al.37 | 3 | 1.5 × 1.5 | 4.7/1.4 | Gadoterate meglumine | 0.2 mmol/kg | 15–25 | 230–270 |
| Khurram et al.40 | 1.5 | 1.3 × 1.3 | 3.8/1.52 | Gadopentetate dimeglumine | 0.2 mmol/kg | 15–25 | 240–290 |
| Dewire et al.43 | 1.5 | 1.3 × 1.3 | 2.5–5.5/1.52 | Gadopentetate dimeglumine | 0.2 mmol/kg | 18 | 240–290 |
| Malcolme-Lawes et al.35 | 1.5 | 1.5 × 1.5 | nr | Gadobenate dimeglumine | 20 mL | 12–20 | nr |
CMR, cardiac magnetic resonance; TR, repetition time; TE, echo time; TI, inversion time; nr, not reported.
Image acquisition parameters for DE-MRI sequence used for atrial fibrosis assessment reported in previous studies
| Study | Scanner (Tesla) | In-plane resolution (mm) | TR/TE (ms) | Contrast agent | Dose of contrast agent | Acquisition timing (min) | TI (ms) |
|---|---|---|---|---|---|---|---|
| Peters et al.12 | 1.5 | 1.3 × 1.3 | 4.3/2.1 | Gadopentetate dimeglumine | 0.2 mmol/kg | 15–25 | 280 (average) |
| McGann et al.13 | 1.5 | 1.25 × 1.25 | 6.3/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 230–270 |
| Oakes et al.14 | 1.5 | 1.25 × 1.25 | 6.3/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 230–320 |
| Bisbal et al.15 | 3 | 1.25 × 1.25 | 2.3/1.4 | Gadobutrol | 0.2 mmol/kg | 25–30 | 280–380 |
| Badger et al.19 | 1.5 | 1.25 × 1.25 | 5.5/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 270–310 |
| Hunter et al.23 | 1.5 | 1.3 × 1.3 | nr | Gadolinium-diethylenetriaminepentaacetate | (0.4 mL/kg) - 20 mL | 20 | nr |
| Sramko et al.25 | 1.5 | 1.6 × 1.6 | 4.8/1.5 | Gadobutrol | 0.2 mmol/kg | 11 ± 4 | 270 |
| DECAAF32 | 1.5 and 3 | 1.25 × 1.25 | 5.2/2.4 (1,5T)3.1/1.4 (3T) | Gadoterate meglumine; gadobenate dimeglumine; gadobutrol; gadodiamide | 0.1–0.2 mmol/kg | 15 | nr |
| Harrison et al.34 | 1.5 | 1.3 × 1.3 | 3.0/6.2 | Gadobutrol | 0.2 mL/kg | 20 | nr |
| Hwang et al.37 | 3 | 1.5 × 1.5 | 4.7/1.4 | Gadoterate meglumine | 0.2 mmol/kg | 15–25 | 230–270 |
| Khurram et al.40 | 1.5 | 1.3 × 1.3 | 3.8/1.52 | Gadopentetate dimeglumine | 0.2 mmol/kg | 15–25 | 240–290 |
| Dewire et al.43 | 1.5 | 1.3 × 1.3 | 2.5–5.5/1.52 | Gadopentetate dimeglumine | 0.2 mmol/kg | 18 | 240–290 |
| Malcolme-Lawes et al.35 | 1.5 | 1.5 × 1.5 | nr | Gadobenate dimeglumine | 20 mL | 12–20 | nr |
| Study | Scanner (Tesla) | In-plane resolution (mm) | TR/TE (ms) | Contrast agent | Dose of contrast agent | Acquisition timing (min) | TI (ms) |
|---|---|---|---|---|---|---|---|
| Peters et al.12 | 1.5 | 1.3 × 1.3 | 4.3/2.1 | Gadopentetate dimeglumine | 0.2 mmol/kg | 15–25 | 280 (average) |
| McGann et al.13 | 1.5 | 1.25 × 1.25 | 6.3/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 230–270 |
| Oakes et al.14 | 1.5 | 1.25 × 1.25 | 6.3/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 230–320 |
| Bisbal et al.15 | 3 | 1.25 × 1.25 | 2.3/1.4 | Gadobutrol | 0.2 mmol/kg | 25–30 | 280–380 |
| Badger et al.19 | 1.5 | 1.25 × 1.25 | 5.5/2.3 | Gadobenate dimeglumine | 0.1 mmol/kg | 15 | 270–310 |
| Hunter et al.23 | 1.5 | 1.3 × 1.3 | nr | Gadolinium-diethylenetriaminepentaacetate | (0.4 mL/kg) - 20 mL | 20 | nr |
| Sramko et al.25 | 1.5 | 1.6 × 1.6 | 4.8/1.5 | Gadobutrol | 0.2 mmol/kg | 11 ± 4 | 270 |
| DECAAF32 | 1.5 and 3 | 1.25 × 1.25 | 5.2/2.4 (1,5T)3.1/1.4 (3T) | Gadoterate meglumine; gadobenate dimeglumine; gadobutrol; gadodiamide | 0.1–0.2 mmol/kg | 15 | nr |
| Harrison et al.34 | 1.5 | 1.3 × 1.3 | 3.0/6.2 | Gadobutrol | 0.2 mL/kg | 20 | nr |
| Hwang et al.37 | 3 | 1.5 × 1.5 | 4.7/1.4 | Gadoterate meglumine | 0.2 mmol/kg | 15–25 | 230–270 |
| Khurram et al.40 | 1.5 | 1.3 × 1.3 | 3.8/1.52 | Gadopentetate dimeglumine | 0.2 mmol/kg | 15–25 | 240–290 |
| Dewire et al.43 | 1.5 | 1.3 × 1.3 | 2.5–5.5/1.52 | Gadopentetate dimeglumine | 0.2 mmol/kg | 18 | 240–290 |
| Malcolme-Lawes et al.35 | 1.5 | 1.5 × 1.5 | nr | Gadobenate dimeglumine | 20 mL | 12–20 | nr |
CMR, cardiac magnetic resonance; TR, repetition time; TE, echo time; TI, inversion time; nr, not reported.
Image processing
Image processing techniques are even less standardized than image acquisition protocols. Visual assessment is very operator dependent and the inter-observer variability can be extensive, especially in identifying patchy fibrosis or inhomogeneous grey LGE areas, or in detecting small reconnection gaps.
Some computational techniques are based on a variable (from 2 to 4 SD above reference pixel intensity) and dynamic (slice-by-slice) threshold, and rely on the opinion of an expert to choose the appropriate value.14,32 This makes the methods subjective and prone to user variability.
Comparative analysis of pre-existing fibrosis assessment methods in a healthy subject. The MRI image was processed with the addition of 2, 3, or 4 SD above the mean of the normal myocardium14 and also with the IIR techniques using the validated cut-offs (0.97–1.61)40: the burden of fibrosis varied from the 0.4% detected with the 4 SD method (A) to 3.8% with the 3 SD method (B), reaching 17.95% at just 2 SD above the mean of the defined ‘normal’ tissue (C). With the IIR method, the total burden of fibrosis is very high (39.3%) but contained within the border-zone area (0.97 < IIR < 1.61), while the fibrosis labelled as ‘dense scar’ is completely absent (D). Figures generated with ADAS-AF software (Galgo Medical S.L., Barcelona, Catalonia, Spain).
Image processing techniques and thresholds applied in studies assessing atrial pre- and post-ablation fibrosis with DE-MRI
| Study | n | AF paroxysmal (%) | Pre-/post-ablation fibrosis | Thresholding technique | Fibrosis threshold | Correlation with EAM voltage |
|---|---|---|---|---|---|---|
| Peters et al.12 | 23 | nr | Post-ablation | Visual assessment | Expert's judgement | – |
| McGann et al.13 | 46 | 47 | Post-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + 3 SD | – |
| Oakes et al.14 | 81 | 49 | Pre-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + (2–4) SD | Cut-off: healthy tissue >1 mV; low-voltage tissue 0.1–0.5 mV; scar < 0.1 mV. Positive correlation between low-voltage regions and DE-MRI (R2 = 0.61, P < 0.05) |
| Bisbal et al.15 | 15 | 60 | Post-ablation | Maximum scar density percentage | Th = MIP. (40 ± 5%) (borderzone); Th = MIP. (60 ± 5%) (scar) | 0.21 (IQR 0.10–0.36) (within scar); 0.31 (IQR 0.17–0.51) (border zone); 0.44 (IQR 0.24–0.99) (healthy areas) |
| Badger et al.19 | 144 | 39 | Post-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + 3 SD | Cut-off: healthy tissue >1 mV; low-voltage tissue 0.1–0.5 mV; scar < 0.1 mV. Positive correlation between low-voltage regions and DE-MRI (R2 = 0.57) |
| Hunter et al.23 | 50 | 100 | Post-ablation | SD above Ref nulled myocardium based | Th = mean SI nulled myocardium + 5 SD | – |
| Sramko et al.25 | 73 | 45 | Pre-ablation | SD above Ref histogram based; FWHM | Th = mean SI (‘normal tissue’) + (3–5) SD; Th = SI max − SI min/2 | Cut-offs: scar < 0.5 mV and <0.2 mV (no correlation found between voltage and LGE) |
| DECAAF32 | 260 | 64 | Pre-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + (2–4) SD | – |
| Harrison et al.34 | 16 (pigs) | 0 | Post-ablation | SD above Ref blood pool based | Th = mean SI (blood pool) + 2.3 SD (acute post-ablation); Th = mean SI (blood pool) + 3.3 SD (chronical post-ablation) | 3.3 mV (95% CI 0.8–9.1) (pre-ablation); 0.6 mV (95% CI 0.1–1.5) (acute post-ablation); 0.3 mV (95% CI 0.1–0.8) (chronic post-ablation) |
| Hwang et al.37 | 38 | 32 | Pre- and post-ablation | SD above Ref nulled myocardium based; FWHM | Th = SI mitral valve – SI min/2; Th = mean SI nulled myocardium + 6 SD | Healthy tissue (voltage >1 mV); low-voltage tissue (voltage >0.1 to <0.5 mV); fibrotic scar (voltage <0.1 mV). Concordance in sector-based comparison between EAM voltages and LGE level (P < 0.01) |
| Khurram et al.40 | 75 | 56 | Pre-ablation | IIR | Abnormal myocardium: IIR > 0.97 and <1.61; dense scar: IIR > 1.61 | Cut-off: abnormal myocardium (<0.5 mV); dense scar (<0.1 mV). Local IIR was associated with bipolar voltage (P<0.001) |
| Dewire et al.43 | 60 | 58 | Pre-ablation | IIR | Abnormal myocardium: IIR > 0.97 and <1.61; dense scar: IIR > 1.61 | – |
| Malcolme-Lawes et al.35 | 50 | 100 | Pre- and post-ablation | SD above Ref blood pool based | Th = mean SI (blood pool) + 3 SD | 0.85 ± 0.33 mV (3 SD); 0.50 ± 0.22 mV (4 SD); 0.38 ± 0.28 mV (5 SD). Progressively lower bipolar voltages with LGE level |
| Study | n | AF paroxysmal (%) | Pre-/post-ablation fibrosis | Thresholding technique | Fibrosis threshold | Correlation with EAM voltage |
|---|---|---|---|---|---|---|
| Peters et al.12 | 23 | nr | Post-ablation | Visual assessment | Expert's judgement | – |
| McGann et al.13 | 46 | 47 | Post-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + 3 SD | – |
| Oakes et al.14 | 81 | 49 | Pre-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + (2–4) SD | Cut-off: healthy tissue >1 mV; low-voltage tissue 0.1–0.5 mV; scar < 0.1 mV. Positive correlation between low-voltage regions and DE-MRI (R2 = 0.61, P < 0.05) |
| Bisbal et al.15 | 15 | 60 | Post-ablation | Maximum scar density percentage | Th = MIP. (40 ± 5%) (borderzone); Th = MIP. (60 ± 5%) (scar) | 0.21 (IQR 0.10–0.36) (within scar); 0.31 (IQR 0.17–0.51) (border zone); 0.44 (IQR 0.24–0.99) (healthy areas) |
| Badger et al.19 | 144 | 39 | Post-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + 3 SD | Cut-off: healthy tissue >1 mV; low-voltage tissue 0.1–0.5 mV; scar < 0.1 mV. Positive correlation between low-voltage regions and DE-MRI (R2 = 0.57) |
| Hunter et al.23 | 50 | 100 | Post-ablation | SD above Ref nulled myocardium based | Th = mean SI nulled myocardium + 5 SD | – |
| Sramko et al.25 | 73 | 45 | Pre-ablation | SD above Ref histogram based; FWHM | Th = mean SI (‘normal tissue’) + (3–5) SD; Th = SI max − SI min/2 | Cut-offs: scar < 0.5 mV and <0.2 mV (no correlation found between voltage and LGE) |
| DECAAF32 | 260 | 64 | Pre-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + (2–4) SD | – |
| Harrison et al.34 | 16 (pigs) | 0 | Post-ablation | SD above Ref blood pool based | Th = mean SI (blood pool) + 2.3 SD (acute post-ablation); Th = mean SI (blood pool) + 3.3 SD (chronical post-ablation) | 3.3 mV (95% CI 0.8–9.1) (pre-ablation); 0.6 mV (95% CI 0.1–1.5) (acute post-ablation); 0.3 mV (95% CI 0.1–0.8) (chronic post-ablation) |
| Hwang et al.37 | 38 | 32 | Pre- and post-ablation | SD above Ref nulled myocardium based; FWHM | Th = SI mitral valve – SI min/2; Th = mean SI nulled myocardium + 6 SD | Healthy tissue (voltage >1 mV); low-voltage tissue (voltage >0.1 to <0.5 mV); fibrotic scar (voltage <0.1 mV). Concordance in sector-based comparison between EAM voltages and LGE level (P < 0.01) |
| Khurram et al.40 | 75 | 56 | Pre-ablation | IIR | Abnormal myocardium: IIR > 0.97 and <1.61; dense scar: IIR > 1.61 | Cut-off: abnormal myocardium (<0.5 mV); dense scar (<0.1 mV). Local IIR was associated with bipolar voltage (P<0.001) |
| Dewire et al.43 | 60 | 58 | Pre-ablation | IIR | Abnormal myocardium: IIR > 0.97 and <1.61; dense scar: IIR > 1.61 | – |
| Malcolme-Lawes et al.35 | 50 | 100 | Pre- and post-ablation | SD above Ref blood pool based | Th = mean SI (blood pool) + 3 SD | 0.85 ± 0.33 mV (3 SD); 0.50 ± 0.22 mV (4 SD); 0.38 ± 0.28 mV (5 SD). Progressively lower bipolar voltages with LGE level |
Comparison with EAM voltages when reported.
AF, atrial fibrillation; EAM, electroanatomical mapping; SD, standard deviation; Ref, reference; SI, signal intensity; MIP, maximum intensity pixel; IQR, interquartile range; FWHM, full width half maximum; IIR, image intensity ratio; LGE, late gadolinium enhancement.
Image processing techniques and thresholds applied in studies assessing atrial pre- and post-ablation fibrosis with DE-MRI
| Study | n | AF paroxysmal (%) | Pre-/post-ablation fibrosis | Thresholding technique | Fibrosis threshold | Correlation with EAM voltage |
|---|---|---|---|---|---|---|
| Peters et al.12 | 23 | nr | Post-ablation | Visual assessment | Expert's judgement | – |
| McGann et al.13 | 46 | 47 | Post-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + 3 SD | – |
| Oakes et al.14 | 81 | 49 | Pre-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + (2–4) SD | Cut-off: healthy tissue >1 mV; low-voltage tissue 0.1–0.5 mV; scar < 0.1 mV. Positive correlation between low-voltage regions and DE-MRI (R2 = 0.61, P < 0.05) |
| Bisbal et al.15 | 15 | 60 | Post-ablation | Maximum scar density percentage | Th = MIP. (40 ± 5%) (borderzone); Th = MIP. (60 ± 5%) (scar) | 0.21 (IQR 0.10–0.36) (within scar); 0.31 (IQR 0.17–0.51) (border zone); 0.44 (IQR 0.24–0.99) (healthy areas) |
| Badger et al.19 | 144 | 39 | Post-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + 3 SD | Cut-off: healthy tissue >1 mV; low-voltage tissue 0.1–0.5 mV; scar < 0.1 mV. Positive correlation between low-voltage regions and DE-MRI (R2 = 0.57) |
| Hunter et al.23 | 50 | 100 | Post-ablation | SD above Ref nulled myocardium based | Th = mean SI nulled myocardium + 5 SD | – |
| Sramko et al.25 | 73 | 45 | Pre-ablation | SD above Ref histogram based; FWHM | Th = mean SI (‘normal tissue’) + (3–5) SD; Th = SI max − SI min/2 | Cut-offs: scar < 0.5 mV and <0.2 mV (no correlation found between voltage and LGE) |
| DECAAF32 | 260 | 64 | Pre-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + (2–4) SD | – |
| Harrison et al.34 | 16 (pigs) | 0 | Post-ablation | SD above Ref blood pool based | Th = mean SI (blood pool) + 2.3 SD (acute post-ablation); Th = mean SI (blood pool) + 3.3 SD (chronical post-ablation) | 3.3 mV (95% CI 0.8–9.1) (pre-ablation); 0.6 mV (95% CI 0.1–1.5) (acute post-ablation); 0.3 mV (95% CI 0.1–0.8) (chronic post-ablation) |
| Hwang et al.37 | 38 | 32 | Pre- and post-ablation | SD above Ref nulled myocardium based; FWHM | Th = SI mitral valve – SI min/2; Th = mean SI nulled myocardium + 6 SD | Healthy tissue (voltage >1 mV); low-voltage tissue (voltage >0.1 to <0.5 mV); fibrotic scar (voltage <0.1 mV). Concordance in sector-based comparison between EAM voltages and LGE level (P < 0.01) |
| Khurram et al.40 | 75 | 56 | Pre-ablation | IIR | Abnormal myocardium: IIR > 0.97 and <1.61; dense scar: IIR > 1.61 | Cut-off: abnormal myocardium (<0.5 mV); dense scar (<0.1 mV). Local IIR was associated with bipolar voltage (P<0.001) |
| Dewire et al.43 | 60 | 58 | Pre-ablation | IIR | Abnormal myocardium: IIR > 0.97 and <1.61; dense scar: IIR > 1.61 | – |
| Malcolme-Lawes et al.35 | 50 | 100 | Pre- and post-ablation | SD above Ref blood pool based | Th = mean SI (blood pool) + 3 SD | 0.85 ± 0.33 mV (3 SD); 0.50 ± 0.22 mV (4 SD); 0.38 ± 0.28 mV (5 SD). Progressively lower bipolar voltages with LGE level |
| Study | n | AF paroxysmal (%) | Pre-/post-ablation fibrosis | Thresholding technique | Fibrosis threshold | Correlation with EAM voltage |
|---|---|---|---|---|---|---|
| Peters et al.12 | 23 | nr | Post-ablation | Visual assessment | Expert's judgement | – |
| McGann et al.13 | 46 | 47 | Post-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + 3 SD | – |
| Oakes et al.14 | 81 | 49 | Pre-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + (2–4) SD | Cut-off: healthy tissue >1 mV; low-voltage tissue 0.1–0.5 mV; scar < 0.1 mV. Positive correlation between low-voltage regions and DE-MRI (R2 = 0.61, P < 0.05) |
| Bisbal et al.15 | 15 | 60 | Post-ablation | Maximum scar density percentage | Th = MIP. (40 ± 5%) (borderzone); Th = MIP. (60 ± 5%) (scar) | 0.21 (IQR 0.10–0.36) (within scar); 0.31 (IQR 0.17–0.51) (border zone); 0.44 (IQR 0.24–0.99) (healthy areas) |
| Badger et al.19 | 144 | 39 | Post-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + 3 SD | Cut-off: healthy tissue >1 mV; low-voltage tissue 0.1–0.5 mV; scar < 0.1 mV. Positive correlation between low-voltage regions and DE-MRI (R2 = 0.57) |
| Hunter et al.23 | 50 | 100 | Post-ablation | SD above Ref nulled myocardium based | Th = mean SI nulled myocardium + 5 SD | – |
| Sramko et al.25 | 73 | 45 | Pre-ablation | SD above Ref histogram based; FWHM | Th = mean SI (‘normal tissue’) + (3–5) SD; Th = SI max − SI min/2 | Cut-offs: scar < 0.5 mV and <0.2 mV (no correlation found between voltage and LGE) |
| DECAAF32 | 260 | 64 | Pre-ablation | SD above Ref histogram based | Th = mean SI (‘normal tissue’) + (2–4) SD | – |
| Harrison et al.34 | 16 (pigs) | 0 | Post-ablation | SD above Ref blood pool based | Th = mean SI (blood pool) + 2.3 SD (acute post-ablation); Th = mean SI (blood pool) + 3.3 SD (chronical post-ablation) | 3.3 mV (95% CI 0.8–9.1) (pre-ablation); 0.6 mV (95% CI 0.1–1.5) (acute post-ablation); 0.3 mV (95% CI 0.1–0.8) (chronic post-ablation) |
| Hwang et al.37 | 38 | 32 | Pre- and post-ablation | SD above Ref nulled myocardium based; FWHM | Th = SI mitral valve – SI min/2; Th = mean SI nulled myocardium + 6 SD | Healthy tissue (voltage >1 mV); low-voltage tissue (voltage >0.1 to <0.5 mV); fibrotic scar (voltage <0.1 mV). Concordance in sector-based comparison between EAM voltages and LGE level (P < 0.01) |
| Khurram et al.40 | 75 | 56 | Pre-ablation | IIR | Abnormal myocardium: IIR > 0.97 and <1.61; dense scar: IIR > 1.61 | Cut-off: abnormal myocardium (<0.5 mV); dense scar (<0.1 mV). Local IIR was associated with bipolar voltage (P<0.001) |
| Dewire et al.43 | 60 | 58 | Pre-ablation | IIR | Abnormal myocardium: IIR > 0.97 and <1.61; dense scar: IIR > 1.61 | – |
| Malcolme-Lawes et al.35 | 50 | 100 | Pre- and post-ablation | SD above Ref blood pool based | Th = mean SI (blood pool) + 3 SD | 0.85 ± 0.33 mV (3 SD); 0.50 ± 0.22 mV (4 SD); 0.38 ± 0.28 mV (5 SD). Progressively lower bipolar voltages with LGE level |
Comparison with EAM voltages when reported.
AF, atrial fibrillation; EAM, electroanatomical mapping; SD, standard deviation; Ref, reference; SI, signal intensity; MIP, maximum intensity pixel; IQR, interquartile range; FWHM, full width half maximum; IIR, image intensity ratio; LGE, late gadolinium enhancement.
Comparative analysis of post-procedural fibrosis in a patient submitted to AF ablation. The MRI image was processed at (A) 3 SD above the reference threshold (first mode of lower pixel intensity), (B) the fixed pixel-based threshold of 40–60% of the maximum intensity pixel (MIP),15 and (C) the IIR of 0.97–1.61.40 The first method detects 14.07% of the LA area as scar, very similar in percentage and localization to the ‘dense fibrosis’ identified with the second technique (14.28%), while the IIR (0.97–1.61) detects only a small amount of ‘dense scar’ (2.89%). Meanwhile, the total burden of fibrosis detected by the three different methods differed widely (14.07, 88, and 57%, respectively), as did the ‘border-zone’ area, when assessed. Figures generated with ADAS-AF software (Galgo Medical S.L., Barcelona, Catalonia, Spain).
Magnetic resonance imaging with late gadolinium enhancement validation process
Validating the capability of LGE-CMR to detect fibrosis in the atria is a controversial issue. The most secure and reliable validation should come from histological analysis, but this would be difficult to perform in humans, and can be performed only in localized areas. Thus, studies usually compare LGE-CMR intensities to voltage maps.
This can be seen as a surrogate of the histological study, but it is an approximate approach affected by several technical variables, such as the contact force and the orientation of the mapping catheter, the catheter tip size, and the number of voltage points registered. Another considerable problem is that the voltage cut-offs for defining atrial scar are not well established. Traditionally, voltage threshold was set at <0.05 mV for identifying atrial scar (noise level of the first measurement equipment).48 The most widely used voltage values are now <0.1 mV for dense scar and <0.5 mV for border zone.14,37,40–42 Recently, Harrison et al.34 found a voltage cut-off of 0.3 mV, derived from histological studies in pigs, but this has not been validated for humans. Another study49 has demonstrated that a bipolar voltage of <0.15 mV best predicted sites without pace capture in patients undergoing redo PV isolation.
Even after agreeing on a voltage threshold, matching voltage points to resonance points is not straightforward. First of all, segmentation errors can add noise to the LGE-CMR intensity values (e.g. blood-pool voxels or aortic wall voxels included in the myocardium can be seen as scar). Besides, fusing the shell obtained from the resonance imaging to the shell obtained with EAM can be tricky: a slight miss-alignment can cause healthy pixels to fall into a scar zone of the other shell, and vice versa. Thus, the correlation measure used to validate that a certain LGE-MRI intensity is in a scar zone or in a healthy zone is noisy and prone to errors. These errors can be reduced by comparing areas or sectors of LGE-MRI shell and EAM shell, instead of comparing single points or very small circular areas. Using sectors averages the errors and reduces the measurement noise, but this makes it impossible to see gaps and very small and localized structures, which can provide valuable information for the diagnosis and procedure planning.
Future perspectives
There is perhaps as much potential to be gained from correct CMR assessment of atrial fibrosis as there are limits to the techniques that we have reviewed.
Cardiac magnetic resonance T1 mapping is an emerging alternative tool, based on the longitudinal recovery time of hydrogen atoms of tissues after excitation. It has been used for global diffuse interstitial ventricular fibrosis assessment50–54 and has been recently proposed as an alternative approach to quantify diffuse fibrosis in the atria.55,56
New developments in T1 mapping techniques have permitted non-invasive quantification of myocardium extracellular volume fraction (ECV), an imaging marker of diffuse cardiac fibrosis, oedema, and amyloid, validated against histological studies.57,58 Extracellular volume fraction maps are based on measurement of longitudinal relaxation time (T1) before and after gadolinium contrast administration both in myocardial tissue and blood pool, calibrated using the value of haematocrit. Compared with native and post-contrast T1 mapping, ECV maps are less dependent on physiological parameters, by reducing the confounding factors related to renal clearance of contrast, contrast dose, body composition, timings of acquisition post-contrast, and haematocrit.
Nevertheless, T1 mapping techniques do not provide a localization of the fibrosis, and therefore are not suitable to guide ablation procedures or to recognize gaps. To detect focal and cohesive fibrosis, such as that induced by RF application, LGE remains the most promising tool, if correctly utilized.
With new software, advanced technology, and improvement in MRI resolution, it is likely that MRI images will be processed more accurately with automated methods. Meanwhile, in order to obtain reliable and clinically applicable results, the use of standardized protocols is necessary to ensure uniformity of image acquisition and processing.
Conflict of interest: L.M. and L.P. are consultants for Medronic, St Jude Medical, Boston Scientific, and Sorin Group.
