Phenotyping heart failure by cardiac magnetic resonance imaging of cardiac macro- and microscopic structure: state of the art review

Abstract

Heart failure demographics have evolved in past decades with the development of improved diagnostics, therapies, and prevention. Cardiac magnetic resonance (CMR) has developed in a similar timeframe to become the gold-standard non-invasive imaging modality for characterizing diseases causing heart failure.

CMR techniques to assess cardiac morphology and function have progressed since their first use in the 1980s. Increasingly efficient acquisition protocols generate high spatial and temporal resolution images in less time. This has enabled new methods of characterizing cardiac systolic and diastolic function such as strain analysis, exercise real-time cine imaging and four-dimensional flow.

A key strength of CMR is its ability to non-invasively interrogate the myocardial tissue composition. Gadolinium contrast agents revolutionized non-invasive cardiac imaging with the late gadolinium enhancement technique. Further advances enabled quantitative parametric mapping to increase sensitivity at detecting diffuse pathology. Novel methods such as diffusion tensor imaging and artificial intelligence-enhanced image generation are on the horizon.

Magnetic resonance spectroscopy (MRS) provides a window into the molecular environment of the myocardium. Phosphorus (³¹P) spectroscopy can inform the status of cardiac energetics in health and disease. Proton (¹H) spectroscopy complements this by measuring creatine and intramyocardial lipids. Hyperpolarized carbon (¹³C) spectroscopy is a novel method that could further our understanding of dynamic cardiac metabolism.

CMR of other organs such as the lungs may add further depth into phenotypes of heart failure. The vast capabilities of CMR should be deployed and interpreted in context of current heart failure challenges.

Graphical Abstract

Graphical Abstract.

The role of cardiac magnetic resonance (CMR) in heart failure phenotyping. CMR can non-invasively assess cardiac structure and function at the systemic, macro-, microscopic and molecular levels. Clear phenotypes are necessary for personalized heart failure therapies.¹⁵⁷

Introduction

Descriptors of illnesses resembling heart failure have existed since the ancient times.¹ The application of X-rays, ultrasound, and more recently, nuclear magnetic resonance (NMR), have provided increasingly sophisticated tools to help us see the heart at work in health and disease. In the late 1980s, the first drug therapies to improve heart failure survival caused a cascade of successive positive drug trials and forced a shift in understanding of heart failure as a multisystemic disorder.^2,3 Despite this commendable progress, mortality remains high, and the demographics of heart failure in the developed world are evolving.^4–9 The same agents that revolutionized treatment of heart failure with reduced ejection fraction (HFrEF) have largely proved ineffective for the increasingly prevalent heart failure with preserved ejection fraction (HFpEF).¹⁰ Treatment allocation by left ventricular ejection fraction (LVEF) as a single phenotypic parameter underpins a one size fits all approach that is no longer sufficient to meet the demands of the modern heart failure demographic. In the field of oncology, personalized medicine has already made tremendous impacts on cancer outcomes.^11,12 Analogous to cancer, heart failure is the clinical manifestation of a diverse range of underlying pathophysiological processes. For targeted heart failure therapy to be possible, precise phenotypic distinction must be linked to clear mechanistic understanding. To this end, cardiac magnetic resonance (CMR) is an indispensable tool. NMR has given us insights into cardiac biology since the 1970s.¹³ With the development of clinical magnetic resonance imaging (MRI) in the 1980s, CMR imaging soon became established as the gold-standard non-invasive modality to accurately and reproducibly document cardiac structure and function.^14–17 The ability of CMR to interrogate multiple additional parameters beyond macrostructure truly distinguishes it from other cardiac imaging modalities. In the present article, we review how CMR can delineate phenotypes and elucidate pathophysiological mechanisms underlying heart failure.

Imaging the heart: more than LVEF

The first in vivo CMR measurements of human hearts were published in 1985 and took 2 hours of scanning using a 0.15 Tesla (T) magnet and electrocardiography (ECG) gating.¹⁸ Faster imaging methods soon made it possible to acquire moving pictures (cine) of the beating heart over a cardiac cycle.¹⁹ Modern cine acquisitions typically produce standard views (short and long axis slices) covering the whole heart with high blood-to-myocardium contrast, spatial resolution of ≤1.5 mm², temporal resolution of ≤45 ms, and all within 5–10 min of scanning (Figure 1).^21–23 Quantitative analysis can be performed on these images to estimate wall and chamber volumes (Figure 2A and B), from which cardiac mass, stroke volume, and ejection fraction can be calculated. The interstudy reproducibility of CMR measurements has been validated in clinical studies and has permitted cardiac remodelling to be tracked over the course of treatment changes.^15,24–30 Published reference ranges for key cardiac dimensions, volumes, and derived values can help identify pathological quantitative parameters and facilitate early disease detection and prognostication.³¹ Abnormal LVEF was one of the first imaging biomarkers to be associated with impaired prognosis in heart failure.^32–34 While LVEF continues to guide prognosis-modifying heart failure therapies, there are many limitations with its use.^35,36 For example, there is poor concordance between measurements of LVEF made using different modalities.¹⁴ Moreover, LVEF cannot account for loading conditions and has low sensitivity to detect early left ventricular (LV) systolic dysfunction at higher ends of the LVEF spectrum.^37,38

Figure 1.

CMR images in the horizontal long axis (HLA) views from 1985 (A)²⁰ and 2023 (B) are shown in the top panels; the bottom panels compare short-axis views from a steady-state free precession (SSFP) sequence in (C) against a free-breathing, real-time cine sequence (D).

Figure 2.

Panels (A) to (C) outline volumetric analysis using a short axis image stack, semi-automatic segmentation and Simpson's rule; myocardial strain (H) can be derived by tracking cardiac motion through tagging sequences (D–E) or feature-tracking (F–G).

Strain analysis

The limitations of LVEF have led to interest in other biomarkers, such as LV strain. While LVEF represents the net volume change from myocardial contraction, strain assesses myocardial deformation in specified dimensions, typically longitudinally, circumferentially, or radially. This allows greater sensitivity in detection of regional myocardial contractile impairment, even when LVEF is ‘normal’ (≥50%). LV global longitudinal strain has been associated with reduced functional capacity and higher cardiovascular morbidity and mortality.^39,40 While strain analysis was popularized by speckle-tracking echocardiography (STE), consistent ultrasonic image quality can be challenging to achieve. Strain analysis by CMR offers solutions to this.

Tagging [Spatial Modulation of Magnetisation (SPAMM)] is accepted as the reference standard tissue tracking strategy for CMR strain assessment. Parts of the myocardium are selectively ‘tagged’ with a magnetic signature, by using selective radiofrequency saturation planes to cause local magnetization perturbations. A grid overlay is typically formed by tagging orthogonally intersecting lines (Figure 2D and E). These taglines deform with myocardial contraction, allowing myocardial strain quantification. The greatest limitation of strain measurement by SPAMM or alternative tagging sequences arises from the need for time-consuming dedicated tagging acquisitions, which limits widespread adoption.^41–43

Feature tracking CMR analysis (FT-CMR) for strain quantification has emerged as a more favoured method. FT-CMR is a post-processing technique that can be applied to existing CMR cine acquisitions (Figure 2F–H). Like STE, FT-CMR uses a block-matching approach, tracking points, or regions of interest throughout the cardiac cycle. Consistent image quality from CMR cine allows most software packages to track 90% of the imaged segments, compared to 67–80% of segments in STE, owing to variable ultrasonic image quality.^39,44 Importantly, FT-CMR strain quantification has been shown to correlate well with the gold-standard myocardial tagging and STE.^45,46 Strain assessment can assess systole and diastole and can also be performed on the right ventricle (RV) and the atria, though this is more challenging owing to thinner walls and more complex morphology.

Left ventricular diastolic function

Left ventricular diastolic dysfunction describes impairment of left ventricular filling. It is a key contributor to clinical manifestations in heart failure and underpins HFpEF. Compared to contractile (systolic) function, diastolic function is more challenging to quantify with non-invasive imaging. Invasive haemodynamic assessment with cardiac catheterization remains the gold-standard, but has inherent risks and cost implications, thus limiting access. Multiple echocardiographic (echo) parameters assess LV diastolic function, including left atrial volume (LAV), Doppler flow assessment of the mitral valve, tricuspid valve, pulmonary veins and tissue Doppler assessment of LV deformation. These have variable correlation with invasive haemodynamic measurements. Probabilistic approaches combine these parameters, but result in a large intermediate group in whom diastolic function cannot be determined.^47,48

There is rising interest in CMR assessment of diastolic function.⁴⁹ One study by Rathi et al.⁵⁰ compared CMR phase contrast flow assessment and Doppler echocardiography in estimating mitral inflow and pulmonary vein velocities, but found systematic underestimation by CMR. Another approach used automated segmentation of routine short-axis cine images throughout the cardiac cycle to plot LV filling curves to correlate with echo-based LV filling patterns.⁵¹ Tissue phase mapping (Figure 2I) with CMR phase contrast (in analogy to tissue Doppler echocardiography) has been used in an obese cohort to describe multidimensional myocardial deformation velocities.⁵² Such methods offer advantages over echocardiography, such as more consistent image quality. However, the threshold for their widespread adoption is high, as echocardiography has superior temporal resolution, is more accessible, and is usually sufficient for estimating similar parameters.

Recent studies have compared CMR-based diastolic function assessment directly with invasive haemodynamic measurements and clinical outcomes. For example, LV strain derived from feature-tracking CMR has been shown to predict abnormal invasive measurements of LV relaxation.⁵³ Multiple studies have also used FT-CMR to correlate left atrial (LA) strain with diastolic dysfunction, invasive haemodynamics, and outcomes.^54–56 Other studies have suggested a link between CMR RV strain and elevated pulmonary arterial pressure.^57–59 Unique CMR methods for assessment of LV diastolic function are also emerging. For example, CMR tissue characterization can detect LV and LA fibrosis, which have been linked to diastolic dysfunction and HFpEF.^49,60 Four-dimensional (4D) flow may offer novel ways to visualize cardiac haemodynamics non-invasively, with potential applicability to diastolic function.⁶¹

4D flow

Traditional phase contrast flow measures the velocity of blood passing through or within a set imaging plane. In 4D flow, velocity is measured in all three dimensions of space and over the time of the cardiac cycle, comprising the fourth dimension. The acquired data can be post-processed to derive a variety of haemodynamic measures such as blood kinetic energy, vorticity, and flow component volumes, which can be analysed and visualized globally or regionally. Small mechanistic studies have reported reduced kinetic energy peaks in diastole and systole in heart failure patients, and a progressive decline in early diastolic peak kinetic energy. Other studies have described apically displaced vortices in dilated cardiomyopathy (DCM)⁶² or methods to estimate flow inefficiency and energy loss.⁶³ Perhaps the most exciting is the potential to non-invasively estimate intracardiac pressure gradients, either through relative pressure calculations or associations with vortex patterns (Figure 3A).^64,67,68 In 145 patients undergoing right heart catheterization for suspected pulmonary hypertension, invasive measurements of mean pulmonary artery pressure (mPAP) correlated strongly with pulmonary artery vortex duration as measured on 4D flow analysis.⁶⁹ In the left atrium, 4D flow parameters such as peak velocity and vorticity can help distinguish patients with a history of atrial fibrillation from those in sinus rhythm, highlighting their potential to reveal new mechanisms or develop as clinical biomarkers.^70,71 At present, 4D flow remains confined to selected centres with expertise, and is limited by long image acquisition times and intensive post-processing requirements.

Figure 3.

Novel CMR techniques such as 4D flow (A)⁶⁴, diffusion tensor imaging (B)⁶⁵ and pulmonary proton density mapping (C)⁶⁶ may offer new perspectives on heart failure pathophysiology.

Real-time cine and exercise CMR

The commonest clinical cine sequences rely on stitching together frames acquired over multiple heart beats using retrospective ECG-gating, to maximize signal-to-noise ratio (SNR).^22,23,31,72 This method is vulnerable to artefacts and image quality degradation from arrhythmias and respiratory motion. A number of approaches can ameliorate some of these issues, such as prospective ECG-gated imaging and other arrhythmia sorting sequences. Implanted cardiac devices are increasingly CMR-conditional, which combined with wideband imaging, has reduced artefacts on CMR.³¹

More recently, real-time free-breathing cine imaging has become possible owing to continual improvements in image acquisition times. In simplified terms, image acquisition can be shortened several-fold if non-essential parts of raw data acquisition are strategically skipped, or ‘undersampled’. Such techniques have made it possible to image the whole heart with 3D cine within a single breath-hold of 24 seconds, with image quality comparable to conventional 2D cine.^73,74 These faster imaging acquisition methods can obviate the need for stitching, thus permitting real-time cine CMR (RT-CMR), which is less sensitive to heart rate variability and respiratory motion.⁷⁵ RT-CMR opens up opportunities to better assess cardiac function during exercise (Figure 1D), adding a temporal dimension to imaging phenotypes, which traditionally reflect a static snapshot in time. Using RT-CMR, it has been shown that LVEF and right ventricular ejection fraction (RVEF) augmentation to exercise, but not resting LVEF or maximal oxygen consumption (VO₂max), can distinguish patients with DCM from endurance athletes.⁷⁶ Exercise CMR may be of particular value in HFpEF, which typically manifests initially as exertion intolerance. In another recent study using RT-CMR, LV and RV stroke volume augmentation in response to low intensity (20 Watt) exercise was significantly impaired in patients with HFpEF.⁶⁶

Beyond the left ventricle

There is increasing evidence that CMR volumetric and functional assessment of the right ventricle and left atrium adds further understanding and characterization of heart failure phenotypes.

While right ventricular function is persistently difficult to reliably quantify on echocardiography, CMR simplifies RV volumetric analysis. Typically, RV volumes and ejection fraction can be calculated from contour-derived measurements of the same series of short-axis stacks as acquired for the LV (Figure 2). Impairment of RVEF on CMR has been strongly correlated with increased mortality and heart failure hospitalization, regardless of LVEF.⁷⁷ RV strain impairment has also been associated with adverse HF outcomes,^78,79 while RV function on CMR has been found to predict response to cardiac resynchronization therapy (CRT).⁸⁰ These findings are not surprising, given that RV dysfunction may signal later stages of pulmonary arterial hypertension due to left heart failure. CMR measurement of the pulmonary artery to aorta diameter ratio has been correlated with invasive measurements of mPAP and adverse outcomes.⁸¹ In HFpEF, CMR features of RV dysfunction are predictive of mortality, other adverse outcomes, and invasive haemodynamic evidence of pulmonary hypertension.^82,83 In addition to this, CMR is ideally placed to characterize primary RV structural abnormalities, such as in arrhythmogenic right ventricular cardiomyopathy and congenital heart diseases.^17,84

LAV has long been regarded as a surrogate marker of left ventricular diastolic function and is an independent prognostic marker in heart failure.^85,86 Dedicated imaging of the left atrium is rarely performed routinely in most clinical settings, and there is limited consensus as to the preferred method of LAV measurement on CMR. It is worth noting that echocardiography significantly underestimates LA size, while discrepancies are small between LA biplane and short axis measurements on CMR.^87,88 LA volume is a well-validated but likely indirect indicator of LV diastolic function and filling pressure. As such, there is an increasing focus on studying LA function by CMR volumetric and strain analysis.^60,89 In a prospective multi-centre cohort of 121 patients suspected of HFpEF, CMR measurements of left atrial volume index, left atrial reservoir strain, and left atrial area index were found to be the best discriminators between HFpEF and non-HFpEF (as per European Society of Cardiology 2021 guidelines, ± right heart catheterization).^36,54 Impaired LA function on CMR volumetric and FT-CMR has also been associated with exercise intolerance and adverse outcomes in HFpEF.^56,60,90

To the tissue: characterizing the myocardium

The application of NMR to medicine first arose from the discovery that cancer tissues exhibited different NMR properties (higher T1 values) compared to normal tissues.⁹¹ NMR of protons (¹H hydrogen nuclei) in biological tissue primarily arises from water (H₂O) and lipid molecules. The NMR properties of these protons are influenced by their molecular surroundings and their abundance, among other factors, and vary between different tissue types. When probed with pulse sequences of radiofrequency electromagnetic waves, differential NMR signals can be used to distinguish normal myocardium from various disease states.^92–94

Late gadolinium enhancement imaging

While MRI can generate intrinsic tissue contrast, the introduction of gadolinium-based contrast agents to clinical CMR was a game-changer in non-invasive tissue characterization.^92,95 Intravenous gadolinium-based contrast enters and washes out of extracellular tissues before eventual renal clearance. Gadolinium accumulates in tissues with a greater proportion of extra-cellular space, where it exerts a concentration-dependent effect by altering local magnetic properties (significantly shortening T1). Post-contrast images highlight signal intensity differences between tissues depending on their differential concentration of gadolinium. A delay of 5–20 min after contrast injection is typically sufficient for gadolinium to wash out of healthy myocardium. Fibrotic, infiltrated, or oedematous tissue retain gadolinium for longer and will have several-fold increased signal intensity (Figure 4). The distribution of late gadolinium enhancement (LGE) patterns is comparable to histological fibrosis patterns and can indicate the aetiology of cardiac dysfunction.⁹⁶ Typically, ischaemic injury affects the subendocardial layer first, before spreading epicardially, eventually affecting the full thickness of the LV wall in transmural myocardial infarction.^95,97,98 In a seminal paper, Kim et al. used LGE CMR to predict segmental recovery of myocardial contractility after revascularization, by estimating the segmental thickness extents of myocardial infarcts in 60 patients. Based on these findings, myocardial viability by CMR continues to guide revascularization. Non-ischaemic cardiomyopathy patterns of LGE can be more variable, reflecting the heterogeneous group of diseases.^99,100 Typical patterns of cardiomyopathies based on LGE patterns have been described, for example in DCM,^100,101 infiltrative cardiomyopathies such as amyloidosis^102,103 and Fabry’s disease,¹⁰⁴ and inflammatory conditions such as cardiac sarcoidosis^105–107 and myocarditis.^108,109 LGE as a surrogate for scar has been extensively associated with adverse cardiovascular events such as mortality and ventricular arrhythmias.^110–116 Unpredictable gadolinium pharmacokinetics complicate quantitative interpretation of LGE imaging. As scarred regions are detected by comparison to normal myocardium, this method is less sensitive to diffuse pathologies affecting the myocardium more uniformly.

Figure 4.

A transmural myocardial infarction is seen as bright (high-signal) on the late gadolinium-enhanced image because the scar tissue retains gadolinium for longer compared to surrounding normal myocardium; this retained gadolinium shortens T1 and increases signal intensity on an inversion recovery (IR) image adapted to assign low signal (black) to healthy myocardium (myocardial nulling).

Gadolinium contrast is also used to assess the coronary vasculature, through first pass perfusion techniques, first performed in humans in the 1990s.¹¹⁷ Briefly, consecutive images are acquired to track the passage of a venous bolus of gadolinium through the cardiac chambers and the LV myocardium. Myocardial segments with impaired perfusion have delayed and reduced gadolinium passing through and show up as lower signal intensity (darker) compared to normal myocardium. Pharmacological vasodilation increases flow through normal coronary arteries, highlighting areas of significant stenosis, which cannot increase flow (stress-induced perfusion defects).¹¹⁸ Studies using perfusion CMR have identified a high prevalence of both epicardial coronary artery disease and microvascular dysfunction among patients hospitalized for HFpEF.¹¹⁹ Semi-quantitative perfusion mapping methods have additionally associated impaired myocardial perfusion reserve with HFpEF diagnosis, and may contribute to further mechanistic phenotyping.^120,121

Parametric mapping

Parametric mapping for tissue characterization overcomes some of the limitations of LGE by presenting absolute rather than relative quantities in image pixels, thus permitting interstudy comparison. T1, T2, T2*, and other parameters of relaxation time (relaxometry) are physical parameters specific to the imaged tissue and scanning conditions. In a parametric map, each pixel of the final image represents an estimate of one of these parameters (Figure 5).¹²²

Figure 5.

Relaxometry charts the recovery of net magnetisation of protons within the main magnetic field (B0) following a radiofrequency pulse, such as shown in this example T1 map, where the pixel values represent T1 values, distinguishing different tissue types and highlighting an area of abnormal myocardium.

T2*

T2* exemplifies one of the earliest successes of parametric mapping. T2* as a measure of relaxation time is significantly shortened by the presence of magnetic field inhomogeneities, such as those caused by nearby iron nuclei. In seminal work, Anderson et al. developed a reproducible method to quantify T2* in vivo, validated its inverse correlation against iron concentrations on liver biopsy, and then established normal and pathological T2* ranges in the heart and liver. Subsequent work led to greater mechanistic understanding of heart failure caused by iron overload, revolutionized clinical management strategies, and ultimately improved overall survival in beta-thalassaemia and other conditions that predispose to iron overload.^123,124

T1 mapping, ECV, and T2 mapping

T1 mapping was made practically possible with the development of a modified Look-Locker inversion recovery (MOLLI) sequence in 2004.¹²⁵ Subsequent modifications such as Shortened MOLLI (ShMOLLI) shortened breath-hold times, increasing practicality in clinical heart failure scans.^125–127 For clinical purposes, typically, three to seven short axis slices can be produced, over 10–15 min. Given certain controlled conditions (e.g. scanner, sequence, breath-holds, heart rate, pre- or post-contrast status), myocardial T1 values are reproducible within narrow reference values. Pre-contrast (native) T1 values are lengthened by increased tissue free water content, a marker of disease. As such, T1 mapping can sensitively detect pathology arising from, for example, acute or chronic myocardial infarction, inflammatory, infiltrative, and other causes of cardiomyopathy.⁹² Beyond diagnostic value, elevated native T1 has been extensively correlated with adverse clinical outcomes, providing prognostic and mechanistic insights in heart failure.^128–130 Similarly, abnormally low T1 values are associated with increased fat content, which may reflect fatty infiltration in conditions such as Fabry’s disease.¹⁰⁴ T2 mapping follows similar principles of T1 mapping but is more sensitive to water content, and therefore is more sensitive to acute myocardial oedema in myocardial infarction, myocarditis, or other causes of myocardial inflammation.^92,109,131

While T1 mapping and T2 mapping are sensitive markers of disease, specificity is limited. In addition, both can be falsely elevated or diminished by a variety of factors such as partial volume and off-resonance artefacts. Although some methods ameliorate for common problems, standardization of techniques and reference values has proved challenging. As a compromise, it is advised that centres establish local reference values for clinical use.¹²² Pre- and post-contrast T1 maps can be used to derive extracellular volume fraction (ECV). Although subject to similar limitations as T1, some studies have suggested that ECV has better agreement between different scanning conditions and better correlation with histological measurements of extracellular collagen volume, but requires longer scan times and the use of gadolinium contrast agent.^129,132,133

Given the time-consuming process of generating successive parametric maps, there has been recent interest in magnetic fingerprinting, a technique that potentially enables multiple parametric maps to be generated in condensed time.¹³⁴ In another potentially disruptive development, Zhang et al.^135,136 recently deployed artificial intelligence (AI) methods to generate images resembling LGE, but without the need for injectable contrast, called virtual native enhancement (VNE). Accuracy was validated against porcine histopathology and on clinical validation cohorts. Further validation and application of this technique to wider disease cohorts could potentially revolutionize CMR tissue characterization.

Diffusion tensor imaging

Diffusion tensor imaging (DTI) offers an alternative approach to tissue characterization distinct from relaxometry methods. Briefly, MRI is able to detect the direction and magnitude of restriction to water diffusion. In myocardial tissue, such restrictions arise from cell walls and tissue borders. Thus, the measured directions of diffusion give an indication of average cardiomyocyte orientation within each voxel. The helix angle (HA) is a measure of cardiomyocyte orientation, and normally changes dynamically with cardiac contraction (Figure 3B). Recent human studies have confirmed ex vivo studies of abnormal HA changes in DCM, which may reflect cardiac remodelling.¹³⁷ Fractional anisotropy (FA) measures restrictiveness to water diffusion and is related to myocardial fibre disarray. In humans, myocardial infarction disrupts myocardial tissue architecture and thus reduces restrictions to water diffusion, represented by a lower FA value.¹³⁸ Studies such as these show potential for novel ways to characterize myocardial tissue, but DTI remains limited to a few centres with expertise owing to ongoing technical challenges.

Spectroscopy: tracing cardiomyocyte metabolism

Uniquely among non-invasive cardiac imaging modalities, NMR spectroscopy (MRS) provides a window into cardiac function at the molecular level. One of the first insights into cardiac function with NMR was obtained in 1977 with ³¹P (phosphorus) spectroscopy on a perfused, beating rat heart preparation.¹³ Clinical NMR scanners in 1985 allowed the same to be done in humans in vivo.¹³⁹ In ³¹P-MRS, phosphorus nuclei within different molecules give rise to distinct peaks on a spectrum (Figure 6C), enabling information on relative ratios of key phosphorus metabolites, such as phosphocreatine (PCr) and adenosine triphosphate (ATP). In cardiomyocytes, PCr is an ‘energetic buffer’ and transport molecule, which readily donates its phosphate group to adenosine diphosphate (ADP) in order to maintain ATP concentrations. The PCr/ATP ratio, as a surrogate marker for available cellular energy store, can be altered at various stages of heart failure, giving rise to an energy-deficiency hypothesis of heart failure.¹⁴⁰

Figure 6.

Cardiac metabolism at the molecular level can be assessed using nuclear magnetic resonance spectroscopy (MRS), such as by tracking metabolic destinations of ¹³C pyruvate using hyperpolarized ¹³C-MRS (A)¹⁵⁷, measuring myocardial lipid content using ¹H-MRS (B), or interrogating cellular energy conversion with ³¹P-MRS (C). 2,3-DPG, 2,3-diphosphoglyceric acid; Acetyl-CoA, acetyl coenzyme A; ADP, adenosine diphosphate; ATP, adenosine triphosphate; CK, creatine kinase; Cr, creatine; PCr, phosphocreatine; TCA, tricarboxyclic acid.

A reduced PCr/ATP ratio may signify lower cardiomyocyte energy availability and has been associated with ischaemic or non-ischaemic cardiomyopathy,¹⁴¹ diastolic dysfunction in hypertensive cardiomyopathy,¹⁴² and adverse prognosis in DCM.¹⁴³ As such, ³¹P-MRS has highlighted metabolic mechanisms in heart failure and also showed potential as a prognostic tool. Emphasizing this point, a recent study by Samuel et al.¹⁴⁴ found ATP depletion on MRS before primary prevention defibrillator (ICD) implantation to be predictive of arrhythmic events and cardiac death in HFrEF patients over 10 years of follow-up.

Recent studies applying ³¹P-MRS to HF and risk factor groups, such as diabetes and obesity, suggest a promising cardiometabolic phenotype. Importantly, weight loss improved energy efficiency in obese patients with DCM and also improved ATP handling in obese healthy volunteers.^145,146 Using ³¹P-MRS, Levelt et al.¹⁴⁷ found in 31 patients with diabetes that the cardiac PCr/ATP ratio was impaired, and this was further exacerbated after leg exercise. In patients with HFpEF, compared to controls, resting PCr/ATP ratio as well as exercise-induced cardiac output augmentation were reduced.¹⁴⁸ In another study, patients with cardiac amyloidosis (CA) had the lowest PCr/ATP ratio, followed by HFpEF patients, while control subjects had the highest PCr/ATP ratio, suggesting a correlation with diastolic dysfunction.⁶⁶ In a recent randomized controlled trial of 72 heart failure patients, ³¹P-MRS at rest and on stress found no evidence for a cardiac energetic mechanism of action for empagliflozin.¹⁴⁹

Proton (¹H) spectroscopy (¹H-MRS) plays a complementary role in assessing myocardial metabolism (Figure 6B). ¹H-MRS has emerged as a reliable tool in non-invasive assessment of myocardial triglyceride accumulation, showing good correlation with endomyocardial biopsy results.^150,151 This technique can provide early insight into adverse cardiometabolic changes in at-risk conditions such as diabetes and obesity. Prior studies utilizing cardiac ¹H-MRS have demonstrated the relationship between increased myocardial triglyceride (MTG) levels and decreased LV deformation and diastolic dysfunction in patients with type 2 diabetes mellitus (T2DM).^152,153 Importantly, McGavock et al. demonstrated that myocardial steatosis was detected by cardiac MRS in patients with impaired glucose tolerance, preceding the development of T2DM and LV functional impairment, highlighting it as a potential early biomarker. Crucially, such lipid overload is reversible, and measures such as weight loss have been shown to reduce MTG content and improve diastolic function.¹⁵⁴

Assessment of MTG may also distinguish between LV hypertrophy from hypertensive heart disease versus hypertrophic cardiomyopathy. Animal and human studies have demonstrated increased MTG in response to pressure overload. Sai et al.¹⁵⁵ noted significantly higher MTG levels in subjects with hypertensive heart disease as compared to hypertrophic cardiomyopathy, proposing different metabolic responses depending on cellular and molecular mechanisms driving hypertrophy.

While ³¹P-MRS and ¹H-MRS have provided useful insights into metabolic mechanisms potentially underlying heart failure, there are a number of limitations, such as long scanning times and limited specificity. Importantly, cardiac cellular metabolism is a complex and dynamic process. Conventional spectroscopy techniques are limited by SNR, such that only a few metabolic molecules can be studied reliably, and usually mandate a steady state, given long acquisition times.¹⁵⁶ Human hyperpolarized carbon (¹³C) spectroscopy (Figure 6A) has recently been used to track the downstream metabolites of pyruvate in a small number of patients following myocardial infarction and in patients with diabetes.^157,158 This novel technology is currently still expensive and requires expertise not widely available, but has tremendous potential to provide unique insights into physiological and pathological cardiac metabolism.

The big picture: at the heart of a system

MRI is ideally placed to assess extracardiac structures, adding multisystemic perspectives to heart failure pathophysiology. The lung and pulmonary vasculature form an integral part of the circulatory system. Recent studies have quantified lung water density (LWD) as a surrogate for pulmonary congestion. Thompson et al.¹⁵⁹ used a widely available water-sensitive MRI sequence to image the lungs in patients with heart failure, patients with risk factors for heart failure, and a group of healthy volunteers. LWD was estimated by referencing the lung signal intensity to that of the liver, which was assumed to have a water density of 70%. Patients with HF or risk factors had increased LWD compared to controls. Increased LWD was also associated with adverse heart failure events and increased left ventricular filling pressure by invasive measurements. Another approach by Burrage et al.⁶⁶ used a novel proton density mapping sequence. LWD at rest and following low intensity (20 W) exercise was compared between patients with cardiac amyloidosis (CA), HFpEF, diabetes, and healthy volunteers. Patients with CA or HFpEF were found to have greater increases in LWD on exercise compared to diabetic patients or controls. These studies suggest a role for lung MRI to detect subclinical pulmonary congestion in heart failure, with potential for further mechanistic studies as well as clinical applications.

The bigger picture: the role of CMR

Since its inception almost four decades ago, CMR has progressively advanced our knowledge and diagnostic toolbox for heart failure (Figure 7). Increasingly detailed descriptors of cardiac structure and pathophysiology are arising from its multiple capabilities. At the same time, biomarkers from other imaging modalities and non-imaging domains are expanding.¹⁸⁷ Cardiomyopathic processes are increasingly aggregated according to common genotypes and phenotypes. For example, inflammatory, inherited, and infiltrative cardiomyopathies are increasingly recognized and streamlined towards specific treatment options and care pathways. Cardiac MRI has already been instrumental in the progress towards personalized medicine in heart failure, and its use is likely to grow further, as technology develops.

Figure 7.

A timeline plotting major heart failure advances on the left and relevant cardiac magnetic resonance advances on the right. Left: advances in cardiac surgery significantly improved heart failure resulting from prevalent valvular disorders such as mitral stenosis in the 1960s.^160,161 The advent of cardiopulmonary bypass led to the first successful coronary artery bypass graft surgery (CABG) in 1961¹⁶² followed soon after by the first heart transplant in 1967.¹⁶³ Motion-mode echocardiography (M-mode) was a key development in phenotyping of heart failure using non-invasive imaging.¹⁶⁴ Until the late 1980s, heart failure therapy largely consisted of digoxin, diuretics, vasodilators, and cardiac surgery if this was an option.¹ In 1977, A. Grüntzig performed the world’s first percutaneous coronary angioplasty, pioneering the field of interventional cardiology and paving the way for primary percutaneous coronary intervention (PCI) to become the gold standard treatment of acute myocardial infarction.^165,166 The first implantable cardioverter defibrillator (ICD) was implanted in 1980, followed in 2004 by the establishment of cardiac resynchronization therapy with biventricular pacing for heart failure.^167,168 Medical treatment for HFrEF progressed with successive clinical trials showing mortality and symptomatic benefits in hydralazine and nitrates,² angiotensin converting enzyme (ACE) inhibitors,³ betablockers,¹⁶⁹ mineralocorticoid receptor antagonists (MRA),¹⁷⁰ angiotensin receptor-neprilysin inhibitor (ARNI),¹⁷¹ and most recently sodium-glucose cotransporter-2 (SGLT2) inhibitors.¹⁷² In 2021, SGLT2 inhibitors were shown to benefit HFpEF prognosis for the first time.¹⁷³Right: several Nobel prizes have been awarded for discoveries in nuclear magnetic resonance and development of medical imaging based on it.^174–178 In 1977, the first human NMR images and the first NMR spectra from a perfused rat heart were obtained.^13,179 In 1985, ECG-gating made it possible to acquire good quality CMR images in human patients and volunteers.^18,20 From the 1990s onwards, cardiac NMR spectroscopy in humans gave rise to new understanding of the energetic deficits underlying heart failure pathophysiology.^140,143,180 Development of gadolinium contrast greatly improved clinical applications of myocardial tissue characterization with CMR.¹⁸¹ Further advances in tissue level CMR arose from parametric mapping methods.^{123,125,127,182} Diffusion imaging, 4D flow, and hyperpolarized ¹³C spectroscopy have potential to develop into novel clinical and research tools in heart failure.^{61,157,183,184} Artificial intelligence is playing an increasingly prominent role in CMR at all stages.^135,185,186

There are a number of limitations to the role of cardiac MRI. Globally, CMR remains a relatively specialized tool that requires costly equipment, maintenance, operator expertise, and comparatively long scan times. Efforts to shorten scanning times through novel scanning and post-processing methods can drastically change the role of CMR for heart failure phenotyping. AI is likely to play an expanding role for cardiac MRI at every stage of the process. For example, at the image acquisition stage, machine learning has helped optimize image quality and shorten acquisition times.¹⁸⁵ On the other end of the CMR process, large prospective cohorts such as the UK Biobank and the Multi Ethnic Study of Atherosclerosis (MESA) generate large data sets of standardized CMR images, which can be used to train and apply machine learning models.^188,189 Cohort and population level studies may play an important role in linking certain imaging features (radiomics) directly with clinical outcomes to delineate phenotypes and predict prognosis.¹⁹⁰ These big data approaches are extremely powerful for generating hypotheses, but smaller mechanistic studies will still be required to verify and probe hypotheses to consolidate our understanding and enable development of targeted, personalized treatment strategies.

In the time that cardiac MR has been developed since the 1980s, the landscape of heart failure has also changed dramatically. Across several longitudinal epidemiological studies, such as the Framingham Heart Study, a clear pattern is emerging.^5,7–9 Incidence of HFrEF and heart failure hospitalization rates have fallen, at least partially as a result of better risk factor management and treatment options developed over the past few decades (Figure 7). Increasingly, heart failure presents in the community, initially with exercise intolerance, rather than overt congestion and hypervolaemia, and is associated with preserved LVEF. Diagnosis of HFpEF remains a challenge. In a vicious cycle, the scarcity of effective treatments encourages diagnostic inertia, while the lack of accurate diagnostic tools prevents development of targeted treatments.

Going forward, it is essential that CMR continues to develop in line with the dynamic clinical milieu. Outlining ‘downstream’ manifestations of disease is no longer sufficient. Rather, efforts should focus on defining cardiac diseases by their underlying biological mechanisms. New technologies should seek validation against biologically and clinically meaningful markers, in order to surpass and complement, rather than approximate, existing methods. Established techniques should be adapted to address challenges of the present, for example, through more widespread exercise testing. As heart failure is underpinned by intolerance to exertion, exercise testing could be the key to delineating pathophysiological responses from health, early in the disease progression.

Conclusion

Heart failure is a common endpoint for a multitude of diseases, and targeted approaches are increasingly needed. CMR can non-invasively interrogate cardiac morphology and function from systemic to molecular levels, contributing to detailed phenotyping of cardiomyopathy, on the path towards personalised medicine in heart failure.

Funding

J.P. is funded by a British Heart Foundation Clinical Research Training Fellowship (BHF FS/CRTF/22/24293).

S.M.N. is supported by a Wellcome Trust Clinical Research Training Fellowship (102176/B/13/Z).

S.N. acknowledges support from the Oxford NIHR Biomedical Research Centre and the Oxford British Heart Foundation Centre of Research Excellence.

O.J.R. is funded by a British Heart Foundation Senior Clinical Research Fellowship.

Data availability

No new data were generated or analysed in support of this research.

Images

Images created with BioRender.com.

Figure 1A was adapted from Higgins et al.²⁰ with permission.

Figure 3A was adapted from Ramos et al.⁶⁴ under a Creative Commons License.

Figure 3B was adapted from Ferreira et al.⁶⁵ under a Creative Commons License.

Figure 3C was adapted from Burrage et al.⁶⁶ under a Creative Commons License.

Figure 6A was adapted from Adapted from Rider et al.¹⁵⁷ under a Creative Commons License.

The Graphical Abstract features a panel adapted from Rider et al.¹⁵⁷ under a Creative Commons License.