Left atrial evaluation by cardiovascular magnetic resonance: sensitive and unique biomarkers

Abstract

Left atrial (LA) imaging is still not routinely used for diagnosis and risk stratification, although recent studies have emphasized its importance as an imaging biomarker. Cardiovascular magnetic resonance is able to evaluate LA structure and function, metrics that serve as early indicators of disease, and provide prognostic information, e.g. regarding diastolic dysfunction, and atrial fibrillation (AF). MR angiography defines atrial anatomy, useful for planning ablation procedures, and also for characterizing atrial shapes and sizes that might predict cardiovascular events, e.g. stroke. Long-axis cine images can be evaluated to define minimum, maximum, and pre-atrial contraction LA volumes, and ejection fractions (EFs). More modern feature tracking of these cine images provides longitudinal LA strain through the cardiac cycle, and strain rates. Strain may be a more sensitive marker than EF and can predict post-operative AF, AF recurrence after ablation, outcomes in hypertrophic cardiomyopathy, stratification of diastolic dysfunction, and strain correlates with atrial fibrosis. Using high-resolution late gadolinium enhancement (LGE), the extent of fibrosis in the LA can be estimated and post-ablation scar can be evaluated. The LA LGE method is widely available, its reproducibility is good, and validations with voltage-mapping exist, although further scan–rescan studies are needed, and consensus regarding atrial segmentation is lacking. Using LGE, scar patterns after ablation in AF subjects can be reproducibly defined. Evaluation of ‘pre-existent’ atrial fibrosis may have roles in predicting AF recurrence after ablation, predicting new-onset AF and diastolic dysfunction in patients without AF. LA imaging biomarkers are ready to enter into diagnostic clinical practice.

Graphical Abstract

Introduction

The left atrium (LA) is increasingly recognized as the bellwether,¹ whose characterization provides important prognostic information as a biomarker of subclinical disease,² reflecting diastolic dysfunction³ and the presence of atrial fibrillation (AF). The LA remodels as it responds to higher left ventricular (LV) diastolic filling pressures.⁴ Function of the LA is controlled by mitral valve plane motion in systole and the properties of LV relaxation in early diastole. Only at the end of the cardiac cycle, during the ‘A-wave’ does the LA myocardium contract (a.k.a. ‘booster function or atrial kick’), finishing the work of LV filling (∼20% of stroke volume under normal conditions). While LA remodelling (fibrosis, increased volume, reduced function among other forms) is most strongly associated with AF,⁵ a highly common disease of chaotic electrical activity arising from the LA,⁶ LA remodelling is present in a variety of cardiovascular diseases.

LA anatomy defined by MR angiography and cine

The anatomy of the LA as defined by cardiovascular magnetic resonance (CMR) imaging is shown in Figure 1. The pulmonary veins (PVs) (most often 4, but variants with 3–5 veins exist^7,8) fill the LA with oxygenated blood from the lungs. The left atrial appendage (LAA) acts as a decompression chamber, but also as a location where thrombus can develop.⁹ Anatomy can be defined with contrast-enhanced MR angiography (MRA). Contrast MRA typically uses 0.1–0.2 mmol/kg gadolinium contrast agent and employs a breath-held dynamic time-resolved three-dimensional (3D) gradient echo sequence to acquire multiple phases during injection (4–8 s per phase), with typical spatial resolution of 1–2 mm³, with 400 mm FOV, repetition time of 2–3 ms, and flip angle of 20–40°. The frame with greatest contrast opacification is typically analysed.^10,11 For acceleration of MR image acquisition, parallel imaging (factor 2) can be employed, plus partial Fourier and view-sharing methods such as 3D TRICKS.^12,13 The timing of the breath-hold relative to the injection can be determined using a bolus-timing scan.¹⁴ Semi-automated segmentation software is commercially available for analysis of contrast MRA images of the LA.^15,16 Contrast MRA of the LA and PVs is commonly performed, for planning prior to pulmonary vein isolation (PVI) by ablation, where the number and size of the veins are useful, and the DICOM data are often imported into the advanced navigation software of an ablation system.¹⁷ Electrocardiogram (ECG) gating is not used in MRA of the LA, although some recent approaches have demonstrated the possibility of ECG-gated non-contrast 3D balanced steady-state free precession (bSSFP) methods. bSSFP imaging of the LA and PVs is challenging, because of the blood flow from the highly off-resonant lung generates artefacts in bSSFP.¹⁸ This is present but remains barely noticeable in conventional long-axis cine,¹⁹ but prevents successful 3D bSSFP cine acquisitions. However, new approaches show promise for 3D ECG-gated non-contrast bSSFP based MRA.^20,21

Figure 1.

(A) Standard examples of slices from 3D contrast MR angiography, showing the pulmonary veins and shape and size of the left atrium. (B) 3D rendering of the left atrium, viewing the posterior wall.

Analysis of morphological features (size, shape, sphericity, etc.) from 3D angiographic images of LA and LAA derived using both magnetic resonance imaging (MRI)^22,23 and CT^24–27 have been shown to be predictive of stroke, AF, and AF recurrence after PVI. Although not all studies using these shape-marker findings have consistently replicated the prognostic value.²⁵

While 3D MRA is used for planning ablation procedures, two-dimensional (2D) cine is commonly used to measure LA volumes and evaluate global LA function (ejection fraction, EF). The conventional technique is shown in Figure 2, in which the two-chamber and four-chamber cine are acquired and the area is planimetered, either manually or using semi-automated tools^28,29 available on many software platforms. Some studies have been performed using Simpson’s method with a dedicated LA short-axis stack,^30,31 or use of two-, three-, and four-chamber views.³² Correlations between values derived by bi-plane and Simpson’s methods are strong.³³ Phasic measurement of volumes is useful for evaluating atrial booster function, but even measurement of minimum and maximum volume yields important information. Volume at any point in the cardiac cycle is measured on long-axis cine by planimetered areas $\left({\mathrm{Area}}_{2Ch}\mathrm{ }\mathrm{and}\mathrm{ }{\mathrm{Area}}_{4Ch}\right)$, and the long-axis length (L) measurements (see Figure 2), using the bi-plane formula, informed by echocardiographic recommendations:³⁴  

$$V=0.85\frac{({\mathrm{Area}}_{2Ch}·{\mathrm{Area}}_{4Ch})}{\mathrm{Length}}.$$

Figure 2.

(A) LA volume and ejection fraction assessment, using two-chamber and four-chamber views, with areas and lengths planimetered. (B) LA volume vs. time curve, showing three characteristic volumes, V_min, V_max and V_pre-A for evaluation.

Normal values are 38 ± 11 mL/m², 14 ± 5 mL/m^2, and 50–62 ± 8% for maximum, minimum volume, and total EF.³⁵ An excellent recent summary³⁵ notes that several areas of non-standardization in volume measurement exist. These include how to measure length (use of averaged value or the smaller value or length from two-chamber only acquisitions), and whether and how to include the PVs and LA appendage (most exclude). Minimum and maximum LA volumes and total LA EF do not change greatly with age, and thus are sensitive markers of disease, but active LA EF increases with age (reflecting mild diastolic dysfunction).^30,36 Women have smaller LA volumes, although this difference disappears after indexing by body surface area.³⁶ More data are needed on normal values for active function (but see³⁷) which requires careful phasic measurement, and might have greater predictive power, e.g. in hypertrophic cardiomyopathy (HCM).³⁸ LA EF is more powerful than LV EF in predicting heart failure outcomes.³⁹ A major pitfall in evaluation of LA volumes and function is the use of long-axis images that are suboptimal due to incomplete presentation of the LA (i.e. ‘foreshortening’). If employed, inaccurate values of volume and function will be measured⁴⁰ and the test–retest reproducibility will be poor.

LA strain and strain rate

Myocardial strain analysis is a relatively recent method to evaluate LA function. Expressed as percentage, strain illustrates the relative deformation of the myocardium. Usually computed in the longitudinal direction of the LA, it can be expressed as a Lagrangian strain: Sl(t) = (L(t) − L₀)/L₀, with L(t) the length of a myocardial segment at time t of the cardiac cycle and L₀ the initial length of the segment, initialized at the QRS complex of the ECG. Strain is complemented by its derivative, the strain rate, an expression of the deformation rate of the LA. The LA strain and strain rate curves are available along the full cardiac cycle and allow a complete evaluation of the LA function and the extraction of indices characteristic of the different atrial phases (see Figure 3): (i) LA peak reservoir strain (Sls) (with ‘l’ denoting longitudinal, ‘s’ denoting systole) after filling to maximal volume prior to the mitral valve opening in end-systole, (ii) LA booster pump strain (Sla), ‘a’ denoting atrial contraction phase, when LV filling is completed in late diastole, and (iii) the LA strain in the conduit phase (Sle = Sls − Sla) (‘e’ denoting emptying phase) corresponding passive emptying in early diastole. Strain rate peaks are evaluated as the maximal strain rate magnitudes during: (i) the LA filling in mid-systole (SRls), (ii) the LA emptying in early diastole after the mitral valve opens (SRle), and (iii) the LA contraction completing the LV filling in late diastole (SRla). Typical normal values for peak strain (%) and strain rate (%/s) are Sls of 30.6 ± 10.6, SRls of 1.29 ± 0.51;⁴¹ or Sls of 25.0 ± 5.4, SRls of 1.2 ± 0.4.⁴²

Figure 3.

(A) LA strain is evaluated by contouring the 2ch and 4ch throughout the cardiac cycle to calculating regional deformation. (B) The strain plotted over the cardiac cycle exhibits a characteristic pattern, and peak global longitudinal strain in systole (Sls), emptying strain (Sle = Sls − Sla) and active strain (Sla) can be measured. (C) The time derivative of the strain curve provides strain rates, with important peaks in LV systole (SRls) and early (SRle) and late diastole (SRla).

Tissue tracking was technically developed in the 90s,^43,44 but the success of LV myocardial tagging methods prevented popularization of this approach with MRI. Meanwhile, LA strain by tagging was not possible, due to the thin LA myocardium, and at that time it was not a metric of interest. Early LA strain evaluations were done using echocardiographic tissue Doppler imaging⁴⁵ or by speckle tracking.⁴⁶ Similarly, the first strain analyses from MRI standard cine images were inspired by speckle tracking and used block matching tissue tracking for the LV^47,48 and were later extended to the LA.^37,49,50 To date, tissue tracking methods applied to long-axis cine images are the only relevant methods for LA strain quantification, as specific magnetization tagging sequences are unable to evaluate the thin LA myocardium.

MRI tissue tracking for LA strain evaluation involves post-processing of routinely acquired two- and four-chamber cine bSSFP images during the full cardiac cycle. Several packages allow evaluation of LA strain, based on either optical flow-related methods like block matching (2D CPA MR (TomTec Imaging Systems⁴⁸) Multimodality Tissue Tracking (Toshiba⁴¹) CardioTrack⁵¹ or non-rigid image registration.⁵² Other available packages likely also follow one of those two categories (CVI 4.2 Tissue Tracking software) [e.g. Circle Cardiovascular Imaging, MR-Wall Motion Tracking software (Vitrea, Canon Medical Systems), QStrain (Medis Medical Imaging Systems)]. Tissue tracking process is usually semi-automatic and the user is required to initialize the LA contour in two- and four-chamber views; initialization is required at a single time frame (maximal or minimal volume). Strain is by definition strongly related to LA volume and EF, but might also reflect stiffness or fibrosis. 3D tissue tracking analysis based on 3D cine of the LA is an important area of development.^53,54

LA strain pathophysiology

LA strain has been shown to be capable of detecting subclinical changes related to normal physiological conditions; with a global decrease in LA strain indices with ageing, slightly lower LA function values in men compared to women and differences between races.^42,55,56 While MRI studies tend to confirm echocardiography findings, the sub-field is relatively new and population studies are sparse.^42,56,57

LA strain has power as an excellent non-invasive early prognostic and diagnostic aid beyond simple LA size alone.^58–61 Decreased LA strain is a stronger predictor of cardiovascular events compared to LA EF or LA indexed volume.^62,63 A recent large study of the general population from Multiethnic Study of Atherosclerosis (MESA) reported an association of decreased LA strain indices by MRI in subjects with cardiovascular risk factors.⁵⁶ Also in the general population, LA strain predicts new cardiovascular events.⁶⁴

Indications of LA remodelling based on decreased peak reservoir strain, even when LA size is normal, has been described in cardiometabolic diseases like hypertension^65–67 and diabetes.^65,68 Besides onset of new AF, LA strain has also been able to predict AF recurrence after ablation,^69,70 and post-operative AF after cardiac valve surgery.⁷¹ LA strain also provides, in some cohorts, efficient grading of diastolic dysfunction⁷² and was able to evaluate precisely the LV filling pressure elevation.⁷³ LA strain is also considered as a non-invasive surrogate for tissue remodelling and has been shown to decrease with the extent of tissue replacement evaluated by histology^74,75 and by LGE evaluation.^76,77

Commonly, only peak LA strain in systole (Sls, also called ‘global longitudinal strain’) is reported. While other strain or strain rate magnitudes seem to evolve along the same trajectory as Sls, they may hold other information. Indeed, while LA strain is known to be load dependent,⁷⁸ this is less the case for strain rate indices. Moreover, recent studies promote the evaluation of the peak booster function strain, which may hold complementary information. Increasingly, studies show that LA booster pump peak strain is a more intrinsic marker of the LA function, e.g. changing to compensate for loss of LV filling during the LA conduit phase in early remodelling stages like in obesity⁷⁹ and indicating subclinical AF.⁸⁰

Reproducibility, normal values, and limitations

MRI tissue tracking LA strain analyses from routinely acquired cine SSFP long-axis images have shown excellent inter-exam and inter-operator reproducibility^51,81–83 and modest to excellent inter-vendor and inter-modality (echo vs. MR) agreement.⁸⁴ Since, as in echocardiography,⁸⁵ strong variations of LA strain indices between the different vendors remain, this limits the use of absolute LA strain normal values.⁸⁴ Normal values for MRI-cine LA strain exist for normal to low-risk ageing populations.^42,56,57,83 Optimal spatial and temporal resolutions of cine SSFP images for LA strain MRI tissue tracking are still unknown despite preliminary work in this area for the LV.^50,86 Finally, as in echocardiography, the LA is easily foreshortened, and then, just as for LA EFs, this leads to overestimated strain values.^40,87,88

LA late gadolinium enhancement

LA LGE can be used to evaluate fibrosis associated with remodelling, due to either pressure or volume overload, that is present in many cardiomyopathies and thought to be the ‘arrhythmic substrate’ of AF. It can also evaluate acute or chronic scar caused by ablation after a PVI procedure. The earliest histology studies found that atrial fibrosis was implicated in AF.⁸⁹ Histology data showed that in patients with mitral regurgitation and AF, there was greater amount of fibrosis compared to age-matched patients with regurgitation but no AF.⁹⁰ It is now recognized that many diseases, including mitral regurgitation, HCM,⁹¹ and structural heart disease⁹² result in atrial fibrosis, prior to the onset of clinical AF, based on histological analyses and confirmed by MR imaging.^93,94 The mechanisms for this remodelling are well categorized by others.⁹⁵ The first in vivo mapping of this atrial fibrosis was performed by electrophysiology voltage mapping of the atrium in AF patients.⁹⁶ These maps found regions of low-voltage, which was equated with fibrosis. At the same time, the development of late gadolinium enhancement (LGE) methods emerged,^97,98 which was used to evaluate fibrosis and scar. MRI is currently unequalled in the ability to delineate fibrotic tissue, based on increased extra-cellular volume fraction (ECV), and this is especially useful in the LV. LGE was quickly adapted for use in evaluating atrial fibrosis and scar.⁹⁹

Acquisition and technical considerations

LGE⁹⁷ is a method that uses bolus injection of gadolinium contrast agent (usually 0.2 mmol/kg dose), and then images the contrast enhancement later during the equilibrium phase, 10–30 min after injection (thus the name LGE). At this juncture, the contrast agent is distributed throughout the body, in vascular and in extra-cellular spaces (∼25% of myocardial space), but excluded from intact myocytes. In myocardial regions with more extra-cellular matrix (fibrosis), more contrast agent is able to enter, thus the bulk contrast concentration is higher, and therefore the myocardial T1 shortens.¹⁰⁰ With abject scar from infarction or ablation, or even diffuse fibrosis associated with remodelling, ECV increases (from normal values of ∼25% up to ∼85% for replacement scar),¹⁰¹ which is reflected in higher contrast concentration and T1 shortening of the myocardium post-injection. LV fibrosis can be imaged using a T1-weighted ECG-gated inversion recovery sequence, with segmented gradient echo readout.⁹⁷ LA LGE⁹⁹ uses this same technique, but modified in several ways (Figure 4 and Table 1), mainly driven by a need for greater contrast (between fibrosis and blood and fat), and higher spatial resolution to visualize scar in the thin atrial wall (∼2 mm).¹⁰² 3D LA LGE sequence is available on most scanners with excellent results (Figure 5).

Figure 4.

(A) The LA LGE pulse sequence uses an inversion recovery 3D GRE, with navigator-gating, fat-saturation and ECG gating. A 180° inversion pulse is used, with imaging performed at the inversion time (TI) shown when myocardium is nulled. Just before the 3D GRE acquisition, fat-saturation and the NAV pulses are applied. (B) Schematic of magnetization. Blood myocardium and scar are inverted, and imaging is performed when myocardium is nulled, at which time fibrosis is enhanced, relative to blood and myocardium, just as in 2D LGE. Fat is inverted too, and is not fully recovered after a time TI, but can be suppressed with a saturation pulse. The dome of the liver, used for navigator-gating is inverted too, so must be immediately reinverted with a NAV restore, so that at the time of the NAV pulse, the liver signal is high (alternatives exist, e.g. a trailing NAV--see diagram).

Table 1.

Modifications of LV LGE sequence for use in 3D LA LGE

2D LV LGE	3D LA LGE
2D, Standard cardiac views	3D, axial, covering LA
1–2 × 1–2 × 6–8 mm3	1–1.5 × 1–1.5 × 2–4 mm3
TR/θ = 4–5 ms/15–20°	TR/θ = 4–5 ms/15–20°, zero-filling
Breath-hold, 10–20 s	Navigator-gated, 3–6 min
No fat-sat, sequential order	Fat suppression, centric order
1 or 2 RR between IR, TI from scout or use of PSIR	1RR between IR, TI from scout
Phase-encoding AP	Phase-encoding RL
Parallel imaging ×2	Parallel imaging ×2
10–30 min post-injection	15–30 min post-injection
Diastole, segmented GRE (∼150 ms)	Diastole, segmented GRE (∼150–200 ms) or systole (esp. in AF)
Scar signal is 3–6 SDs above myocardium or use FWHM method	Scar signal is 2–3 SD’s above blood signal, or use IIR >1.4, or subject specific method

2D, two-dimensional; 3D, three-dimensional; GRE, gradient recalled echo, IIR, image intensity ratio; IR, inversion recovery; LGE, late gadolinium enhancement; LV, left ventricular; PSIR, phase-sensitive inversion recovery; RL, right-left; RR, R-R interval; SD, standard deviation; TI, inversion time; TR, repetition time.

Figure 5.

Pre- (A) and post-ablation (B) images (two slices in a single subject) exhibit important characteristics of LA LGE. (A) Valves should appear enhanced (yellow arrows, A), and myocardium should appear dark. Aortic wall also usually enhances. Fat is suppressed. The NAV artefact is apparent in this study (white arrow, A), and methods to remove it should be considered, and this artifactual enhancement must not be included as ‘fibrosis’. (B) A good quality study will exhibit sharp features such as papillary muscles well delineated (arrow, B). After ablation, scar is clearly visible (arrows, B).

3D LA LGE, like 2D LV LGE, requires user input of the inversion time (TI, see Figure 4). In LV LGE, often two R-Rs between inversions are used, in conjunction with phase-sensitive inversion recovery (PSIR),¹⁰³ to perform a phased reconstruction that is somewhat immune to poor TI choice. For 3D LA LGE, one R-R interval (1RR) between inversions is employed, since doubling scan time is infeasible. For the 1R-R LA LGE, a 1R-R TI-scout should be used. The TI at which the LV myocardium is nulled, plus a ‘fudge factor’ (60 ms at our institution) provides a good TI choice. Automated TI estimation based on heart rate, scan protocol, and current myocardial T1-value is possible.¹⁰⁴ Navigator-gated implementations of PSIR for 3D LGE have been developed, and do show utility, although it is critical that both the first and second heart-beat both be (independently) respiratory-gated.¹⁰⁵ The myocardial T1 and therefore optimal TI does drift upward over 5 min, but only slightly (<10 ms),¹⁰⁶ and is not a concern. 3D LGE can only detect atrial fibrosis with a signal higher than blood (i.e. T1 shorter than blood’s). Therefore, LA LGE must begin >15 min after injection, at which point contrast between blood and fibrosis is significant:¹⁰⁶ even later is better (Figure 6B). A rule of thumb is that, if mitral valvular enhancement (Figure 5A) is not observed, the LA LGE is not analysable. Most contrast agents work similarly, although an agent whose concentration remains high in the blood pool longer is less effective,¹⁰⁷ since contrast between blood and scar is lower (and a lower dose must be used). Thus, LGE methods that generate darker blood and increase contrast have been proposed¹⁰⁸ and some have been tested for imaging the LA.¹⁰⁹

Figure 6.

(A) Phase-encoding direction was chosen A-P, causing artefacts (yellow arrow). A repeat scan used phase-encoding R-L, so artefacts are removed, but TI was too short (white arrow). (B) A good quality post-ablation LA LGE image, repeated 10 min later shows improved contrast and greater conspicuity of scar. (C) LA LGE image of a patient with arrhythmia during scan exhibits poor nulling (yellow arrow) and diffuse artefacts (red arrow). A repeat scan using dynamic saturation pulses improved quality.

The use of navigator (NAV)-gating provides respiratory compensation by monitoring the right hemi-diaphragm in real-time, to acquire data selectively during end-expiration. This increases the available scan time for data acquisition from 20 s breath-hold to ∼2–4 min (4–8 min assuming 50% respiratory-gating efficiency). Prospective tracking of the LA is not used, and the tracking factor for the LA motion relative to the diaphragm has not been well characterized. It is essential that the ‘phase-encoding direction’, normally placed on axial images anterior–posterior, be set in the right-left direction. If not, chest wall ghosting (especially contrast-enhanced skin) will appear prominently, since the chest wall moves out of sync with the superior–inferior respiratory motion of the heart (Figure 6A). Use of a leading NAV (see Figure 4) in an inversion recovery sequence requires an NAV restore pulse (also shown in Figure 4), which reinverts the inverted right hemi-diaphragm, to preserve the diaphragmatic signal for monitoring respiration. Because this ‘restore’ pulse also reinverts PV blood, it generates a most problematic artefact, NAV-induced inflow artefacts, especially in the right PVs (see Figure 5A). Several approaches have been proposed to circumvent this artefact. Most easily implemented, the NAV can be acquired after the data acquisition segment (trailing NAV, see Figure 4) thus obviating the need for an NAV restore pulse (practically speaking this might require the technologist to eliminate the NAV restore pulse manually). Other strategies include an NAV acquired ∼100 ms prior to data acquisition segment¹¹⁰ that does not need an NAV restore, and use of an NAV restore pulse <180°,¹¹¹ which also reduces the inflow artefact. Even bellows-respiratory-gating works satisfactorily.¹¹²

Fat-saturation is essential, as fat encases the LA and has a short T1 (∼210 ms). Indeed epicardial LA fat is a biomarker correlated to AF burden, AF recurrence, and new-onset AF,¹¹³ and harbours the ganglionated plexi.¹¹⁴ Fat suppression uses an SPIR (110° flip angle) and centric acquisition. Fat suppression with SPAIR (inversion of the fat, and linear acquisition of data) cannot succeed, because the fat—inverted by the prior 180°—has a very low signal prior to acquisition (see Figure 4B), so inversion is not an option (unless a fat restore pulse is used¹¹⁵) Dixon fat–water separation methods have been used with linear k-space acquisition order.¹¹⁶

Since many patients imaged with LA LGE have AF, a large minority of these patients are imaged during atrial arrhythmia, which results in an irregular RR interval, and no atrial kick. Interestingly, LA LGE quality is sometimes still good. However, quality is more often poor (Figure 6C), and this manifests as poor nulling of ‘normal’ LV myocardium, and ghosts, both problems due to variable time between RR intervals and thus inversion pulses, resulting in variable magnetization in each data acquisition.¹⁰⁴ Approaches to reducing these artefacts include a dynamic TI¹¹⁷ or a dynamic saturation pulse,^104,118 both of which depend on the history of RR intervals, and use MR physics modelling to change the acquisition in real-time. However, these are not commercially available. Cardiac motion blurring is also a factor; imaging during LV systolic rest-period (LA diastole) might help.¹¹⁹ Finally, on average the signal-to-noise ratio (SNR) of LA LGE in patients with arrhythmia is inherently lower. This is because patients with greater AF burden have a slightly higher myocardial diffuse fibrosis,¹²⁰ thereby requiring a slightly shorter TI to null myocardium; surprisingly, this can impact SNR significantly.¹⁰⁴

A study of factors in optimization of 3D LA LGE showed delayed imaging post-injection (and lower contrast dose) improved quality.¹²¹ Alternative acquisition strategies such as spiral¹²² and radial with variable under-sampling and compressed sensing reconstruction^123,124 have been studied for 3D LA LGE, although their role in improving quality is not established.

Utility of LA LGE

3D LA LGE was first introduced at as a method to identify post-ablation scar.⁹⁹ If able to identify gaps in ablation, this approach could inform ‘redo’ procedures and improve our understanding of why AF recurrence after PVI occurs. The value of this approach in identifying gaps in ablation sets has been investigated, but with conflicting results^125,126 for guiding an ablation procedure. Some found that simple extent of scar predicted freedom from recurrence, while others did not.^127–129 However, significant relationships with recurrence were found, i.e. less ostial scar in reconnected PVs,¹²⁷ or more scar around the right inferior PV.¹²⁸ Gaps in LA LGE have been compared to contact force made during ablation.¹³⁰ One recent study¹³¹ found that electrical reconnection 3 months after ablation was found to be very common, by electrical mapping and LGE. There was a linear relationship between scar burden and extent of PV electrical reconnection. However, there was moderate location by location correspondence. Some lesion gaps may be non-conducting (and therefore not culprits in recurrence) and some apparently circumferential scar might have a gap large enough to conduct. Furthermore, possibly not all lesions are transmural, even in the very thin LA wall.¹³² Furthermore, recurrence often is identified with electrical activity from a site not targeted for ablation.¹³³ Acute atrial ablation lesions can be imaged peri-procedurally using LA LGE, and T1- and T2-weighted imaging.^134–137

Very early, 3D LA LGE was pioneered to identify ‘pre-existent’ atrial fibrosis.¹³⁸ The concept that AF patients with highly remodelled atria are less easily cured with ablation and would more readily develop recurrent AF, led to a series of studies^139,140 by the Utah group to show the predictive power of LA LGE. The DECAAF study involved 272 patients imaged prior to ablation for AF, followed for more than a year, and was able to improve prediction of recurrence using atrial fibrosis ‘Utah stage’, compared to clinical variables, with C statistic increasing from 0.65 to 0.69.¹⁴¹ A follow-up trial will determine whether ablation of pre-existent scar improves PVI efficacy.¹⁴² This finding has been partly replicated in studies from separate groups,^127,143,144 who found that patients with AF recurrence after PVI had significantly greater pre-existent scar. The hypothesis that AF is caused by rotors or re-entrant electrical activity in the LA, often anchored by fibrosis, has led some to compare fibrosis to rotors theoretically;^145,146 however, the experimental studies have not found a consistent relationship between rotor location and LA LGE thus far.^147,148 In a large retrospective studies of AF subjects, greater LA fibrosis was associated with stroke, both existing¹⁴⁹ and new onset.¹⁵⁰

Increasingly, the concept of an ‘atrial cardiomyopathy’ has found support^151,152 from CMR studies. Atrial fibrosis is present to some degree in all people,¹⁵³ and especially present in those with a variety of cardiovascular diseases, from mitral regurgitation (where it might be expected due to pressure/volume overload), to HCM,⁹¹ to subjects with coronary artery disease, to those with amyloid¹⁵⁴ or heart failure,¹⁴³ in addition to patients with AF. The concept that ‘AF begets AF’,¹⁵⁵ i.e. that atrial fibrosis develops mainly during AF, may be true, but atrial fibrosis often precedes atrial arrhythmias. In patients without AF, studies have shown that the extent of atrial fibrosis is associated with diastolic dysfunction.¹⁵⁶ Furthermore, the extent of fibrosis predicts new-onset atrial arrhythmias,⁹⁴ and even subclinical arrhythmias as shown in a large prospective trial.¹⁵⁷ A small but unique study¹⁵⁸ showed that patients without AF showed progression of LA LGE (from 13% to 18% in ∼2 years), with both baseline LA LGE and extent of progression showing potential to indicate new-onset AF.

Controversies: thresholding, validation, and reproducibility

Early controversy suggested that LA LGE is not robust and noted that many cardiac structures enhanced (aortic wall, valve leaflets, and the atrium), so enhancement was non-specific. In fact, although not well recognized by the CMR community, these structures, the aortic wall (the tunica media¹⁵⁹), and the valve¹⁶⁰ have significant extracellular matrix and enhance by LGE.

Table 2 summarizes reproducibility studies for LA LGE and average values, including scan–rescan studies, and reproducibility based on inter- and intra-observer variability in fibrosis segmentation for the same scan. Scan–rescan studies of post-ablation atrial LGE show good reproducibility,^129,165 with Sørensen–Dice (Dice) coefficients of up to 50–70% for ablation scar, in studies repeated at 2 days and 3 months. Such scan–rescan studies have not been performed in pre-ablation patients, which is a major gap in understanding.

Table 2.

Fibrosis and post-ablation scar: reproducibility and values in disease groups and controls

Study first author and year	LGE extent/reproducibility	Disease	Age, % male, number of subjects	Comments
Quail et al. (2019)94	Intra-observer (bias ± 2 SDs): 0.3% ± 2.9%, ICC = 0.94 Inter-observer (bias ± 2 SDs): 0.6% ± 7.3%, ICC = 0.71 Mean pre-existent LGE 4.6%± 7.3%	AF and other CVD, pre-ablation	53 years old, 59%, N = 136	Variable threshold based on mitral valve signal, ∼CNR with blood >3
Habibi et al. (2014)49	5.8 ± 4.4% (Parox) vs. 9.2 ± 7.3% (persistent)	54 Parox AF 36 pers. AF	Age: 61 ± 10 years, 76%, N = 90	IIR > 1.6, no reproducibility data
Cochet et al. (2015)161	Intra-observer (bias ± 2 SDs): 0.1% ± 4.1% ICC = 0.96 Inter-observer (bias ± 2 SDs): 0.4% ± 5.4% ICC = 0.93 Mean pre-existent LGE 18.4 ± 8.9%	AF + other CVD	54 ± 4 years old, 65%, N = 190	Variable threshold Corview, CARMA
Bertelsen et al. (2020)162	Intra-observer (bias ± 2 SDs): −1.0 ± 2.8% ICC = 0.99 Inter-observer (bias ± 2 SDs): −1.7% ± 8.5% ICC = 0.96 Mean pre-existent LGE 2.8% (1.3–8.3) healthy, 9.0% (3.9–12.0) lone AF 20.1% (10.2–35.8) (older, no AF)	Healthy, lone AF, and older non-AF subjects	Healthy: 37 ± 6, 82%, N = 11 Lone AF: 39 ± 5, 91%, N = 11. Older, non-AF: 76 ± 5, 55%, N = 11	Galgo software, IIR = 1.2
Oakes et al. (2009)163	Pre-existent Mean LGE 17.1 ± 14.2% Inter-observer (bias ± 2 SDs) −0.9% ± 7% Intra-observer: (bias ± 2 SDs) 0.5 ± 5.5%	AF 51% parox	63.6 ± 12.0. 64%, N = 81	Variable threshold ∼3 SDs, Corview, CARMA
Malcolme-Lawes et al. (2013)127	∼3.5% pre-ablation ∼13% post-ablation	Parox AF	59.6 ± 13, 62%, N = 50	No reproducibility data
Khurram et al. (2014)164	Local IIR thresholds of >0.97 and >1.61 corresponded to bipolar voltage <0.5 and <0.1 mV, respectively	AF, 56% parox, mixed pre-ablation (43%) and post-ablation (57%)	62 ± 8.3 old, 75%, N = 75	IIR
Chubb et al. (2018)129	Mean post-ablation LGE ∼25% Scan–rescan (bias ± 2 SDs) −1% ± 10%, ICC = 0.7–0.8	Post-ablation AF Compared studies 2 days apart	61± 9, 78%, N = 40	Post-ablation AF Compared studies 2 days apart
Kamali et al. (2020)165	Scan-recan study post-ablation: Dice ∼74% (Otsu) Dice ∼64% (CNR > 3.3) Dice ∼56 % (IIR > 1.6)	3 months post-ablation, scan with rescan 3 months later	69.2 ± 12.0, 69%, N = 45	Used Otsu, IIR > 1.6 and CNR > 3.3
Benito et al. (2017)166	Post-ablation 14.5% (4.88–22.13) LGE = 1–2% for parox and perm AF	10 healthy subjects and 30 AF subjects pre- and post-ablation	Healthy: 22, 50%, N = 10 AF: 58 ± 10, 90%, N = 30	Used >1.3 IIR R = 0.2 for voltage vs. IIR Galgo
Karim et al. (2013)167	Pre-ablation Dice ∼45%, post-ablation Dice 75%. Bias +2 SDs: 1–3 ± 2–6 mls (pre-ablation) and 3–5 ± 4–8 mls post-ablation	30 pre-ablation and 30 post-ablation (AF)		Segmentation algorithms vs. expert consensus

AF, atrial fibrillation; CNR, contrast to noise ratio; Dice, Sørensen–Dice coefficient; ICC, intraclass correlation coeffciient; IIR, image intensity ratio; LGE, late gadolinium enhancement; SD, standard deviation.

Regarding the reproducibility of the segmentation of LA LGE, even more is known.^{94,129,161,163,167,168} Dice overlaps of scar are often 40–50% pre-ablation and 70–80% post-ablation. These may seem to be modest value for Dice, but the sparsity of scar itself drives the Dice metrics lower, compared to that reported, e.g. for the LV cavity. Inter and intra-observer agreement is good, even using study-specific thresholds, with low bias (near zero) and narrow limits of agreement (4–7%) on scar extent.^94,161

Histological validation of LA LGE is mainly lacking, compared to LV LGE, where there has been comparison of LGE to histopathology of chronic infarct models.⁹⁸ In one study of 10 surgical ablation patients, agreement between Masson’s Trichrome stain and LGE presence was observed, although statistical comparisons were not reported.¹⁴⁰ A study of ablation in the right atrium of pigs (left atrium is harder to access, requiring an atrial septal puncture) found that LA LGE thresholds of 2.3 standard deviations (SDs) above blood pool signal (acutely) and 3.3 chronically correlated best with histology.¹⁶⁹ Lacking animal models of AF, LA LGE has mainly been validated by comparison to surrogate markers. LA LGE extent has found consistent correlations with LA volume and EF,⁹⁴ LA strain,⁷⁶ CHADS(2) score,¹⁴⁹ age,¹⁶¹ AF burden,¹⁷⁰ and presence of types of heart disease, including HCM, AF, and MR.^94,161 These relationships are typically not strong. The most relevant surrogate may be voltage mapping.

Voltage mapping has been used to validate LA LGE. Khurram et al.¹⁶⁴ found that a signal >1.6 image intensity ratio (IIR) corresponded to voltage <0.1 mV. Malcolme-Lawes et al.¹²⁷ found that LA LGE defined as >3 SDs above blood pool collocated to regions of lower voltage (<1 mV) on both pre-existent fibrosis and post-ablation scar. The relationship was moderate, which likely reflects reality. Others suggest that a threshold of 1.3 IIR identifies post-ablation scar best.¹⁶⁶

LA LGE is segmented (Figure 7) by including enhanced signal located on the atrial wall. Subject-specific thresholds (e.g. based on mitral valve⁹⁴ or Otsu threshold¹⁶⁵) or fixed thresholds, based on blood intensity and SD of signal in the blood pool (see Table 2) have been utilized. Thresholds that are advocated include signals >3 SDs above blood signal (CNR > 3) and signal 1.2–1.6 higher than blood signal (so-called IIR > 1.2–1.6). One scan–rescan study found that subject-specific Otsu threshold (using an atrial wall signal histogram) performed well.¹⁶⁵  Figure 8 illustrates differences of these inter-related thresholds.¹⁷¹ An IIR of 1 (i.e. a CNR of 0), sometimes advocated, is not practical, as it implies that half the atrium will be categorized as fibrotic even in the case where no fibrosis exists, since normal (partial-volumed) atrial wall is approximately isointense with the blood. The best thresholding approach would estimate ECV based on T1 mapping¹⁷² for each study, and use as a standard threshold an ECV cut-off such as 55%. For example, based on analysis of LGE signal vs. T1, assuming a blood T1 of 300 ms, an IIR threshold of 1.4 (equivalent to a CNR of 3 if blood SNR of 7.5, see Figure 8), would identify atrial fibrosis with ECV >∼55%.¹⁷¹ Another pitfall in segmentation relates to the identification of the atrial wall. PV sleeves are sometimes completely omitted from segmentations and enhanced mitral valve annular tissue (Figure 5A) is sometimes included as ‘atrial fibrosis’. The LA LGE volume is often normalized by the LA myocardial volume (i.e. using a measurement of LA surface area multiplied by the LA wall thickness). The optimality of this normalization has not been explored. The LGE expressed in percent is also often log transformed for improved statistical testing (closer to normal distribution).

Figure 7.

Segmentation of LA LGE. (A) Slice from LA LGE volume, with some enhancement regions of the atrial wall. (B) After thresholding, LA LGE is identified—excluding the enhanced signal of the mitral valve. (C) Segmentation of the LA cavity. (D) 3D reconstruction of LA LGE, and quantification of LA LGE volume in mls, can be performed.

Figure 8.

A single slice from a 3D LA LGE of a post-ablation subject is segmented with multiple thresholds (here shown as both image intensity ratio, IIR and CNR). IIR (vs. blood) and CNR above blood signal are two inter-related metrics for setting thresholds for segmenting fibrosis. They are linked since CNR = (IIR − 1)·SNR_b. Use of a threshold which is too low (e.g. IIR = 1.2) yields non-specific segmentation, while too high a threshold (e.g. IIR = 1.6) misses scar.

Future directions

Several machine-learning networks have been developed and validated to replace tedious manual segmentation of atrial volume and atrial scar on 3D LGE.^173–175 Limitations are that there is no consensus regarding the optimal scar thresholds and the definition of atrial wall, as noted above. This variability of expert segmentations and non-standardization is problematic for automated segmentation approaches for atrial fibrosis. Still this approach is highly valuable.

Echocardiography has shown that stasis of LA and LAA flow are important in thrombus formation.¹⁷⁶ LA and LAA flow can be measured with four-dimensional (4D) phase contrast; the earliest study documented the presence of vortical flow in the atrium, which it hypothesized is a method of conserving kinetic energy until needed during diastolic filling.¹⁷⁷ Later MR studies have been performed to document blood stasis, especially in patients with AF.^178–180 Mean and peak velocities in the LA were lower for AF vs. controls, while LA volumes were much greater. Such findings of stasis have been confirmed by other groups.¹⁸¹ Other studies have compared 4D flow metrics to CHA2DS2-VASc score.¹⁸² The reproducibility of 4D flow is reasonable for application to the LA.¹⁸³

Diffusion tensor imaging, challenging even in the LV, has been performed in the atria, but only in ex-planted hearts.¹⁸⁴ Myocyte hypertrophy, a common pathological finding in the myocardium,¹⁵¹ would be a useful imaging biomarker.¹⁸⁵ LA wall thickness measurement might be possible.²¹ 3D strain maps are feasible and would add value. LA LGE would benefit from standardization and consensus regarding segmentation methods, automated analyses, further scan–rescan studies in pre-existent fibrosis, and strategies for stronger validation. Atrial contraction function may soon be more greatly utilized, as its evaluation is now more automated. In conclusion, LA imaging biomarkers are ready to enter into routine diagnosis and prediction!

Conflict of interest: none declared.

Funding

National Institutes of Health, HL155992.