Advanced flow MRI: emerging techniques and applications

Abstract

Magnetic resonance imaging (MRI) techniques provide non-invasive and non-ionising methods for the highly accurate anatomical depiction of the heart and vessels throughout the cardiac cycle. In addition, the intrinsic sensitivity of MRI to motion offers the unique ability to acquire spatially registered blood flow simultaneously with the morphological data, within a single measurement. In clinical routine, flow MRI is typically accomplished using methods that resolve two spatial dimensions in individual planes and encode the time-resolved velocity in one principal direction, typically oriented perpendicular to the two-dimensional (2D) section. This review describes recently developed advanced MRI flow techniques, which allow for more comprehensive evaluation of blood flow characteristics, such as real-time flow imaging, 2D multiple-venc phase contrast MRI, four-dimensional (4D) flow MRI, quantification of complex haemodynamic properties, and highly accelerated flow imaging. Emerging techniques and novel applications are explored. In addition, applications of these new techniques for the improved evaluation of cardiovascular (aorta, pulmonary arteries, congenital heart disease, atrial fibrillation, coronary arteries) as well as cerebrovascular disease (intra-cranial arteries and veins) are presented.

INTRODUCTION

Cardiovascular function is characterised by the highly integrated and synergistic coupling between the atrial, ventricular, and vascular compartments. The understanding of each individual component has substantially progressed from the molecular to the organ level; however, the direct in-vivo assessment of blood flow, which connects all compartments and plays an important role in the development of cardiovascular disease, remains challenging. Specifically, current imaging techniques are limited with respect to the assessment of certain features of blood flow, which is crucial for the comprehensive characterisation of cardiovascular haemodynamics, such as complex changes in three-dimensional (3D) blood flow patterns, pulsatile nature of arterial flow, flow beat-to-beat variability, blood pressure estimation, and flow quantification in small vessels.

Magnetic resonance imaging (MRI) techniques provide non-invasive and non-ionising methods for the accurate anatomical depiction of the heart and vessels throughout the cardiac cycle. In addition, the intrinsic sensitivity of MRI to motion offers the unique ability to acquire spatially registered blood flow simultaneously with morphological data within a single measurement^1-4. The characterisation of the dynamic components of blood flow with MRI has achieved considerable progress in recent years, in regards to new methodological advances for data acquisition, reconstruction, and analysis. These developments directly impact the ability to assess cardiac and vascular haemodynamics and have opened up new application areas for advanced flow imaging, as summarised in Fig. 1.

Fig. 1.

Summary of recent methodological advances, the impact on MRI flow imaging and quantification of haemodynamic parameters, along with their associated main applications.

RECENT METHODOLOGICAL ADVANCES AND DEVELOPMENTS

Flow imaging with MRI is based on the phase contrast (PC) technique, which can be employed to encode blood flow velocities along all principal dimensions and enables the acquisition of spatially registered information on blood flow simultaneously with morphological data within a single MRI measurement⁵. In current clinical routine practice, PC-MRI is typically accomplished using methods that resolve two spatial dimensions (2D) in individual sections and encode just the time-resolved component of velocity directed perpendicularly to the 2D section. This approach allows measurements of forward, regurgitant, and shunt flows in congenital and acquired heart disease. A number of more advanced and promising flow MRI techniques have been reported, which allow a more comprehensive evaluation of blood flow characteristics, e.g. real-time 2D PC-MRI for the evaluation of flow changes on short timescales and assessment of beat-to-beat variations^6-12; multiple-venc PC-MRI for the enhancement of the velocity dynamic range and/or encoding of flow velocities as a separate dimension along with assessment of sub-voxel velocity distributions^13,14; four-dimensional (4D) flow MRI for the comprehensive analysis of complex time-resolved 3D and three-directional blood flow characteristics^15-20; advanced data analysis and application of fluid dynamics concepts for the quantification of complex haemodynamic properties^21-28,29-41; multi-dimensional data undersampling and advanced respiratory control for highly accelerated 2D and 4D flow MRI ^42-51.

In this article, we will review the status of these MRI techniques and explore emerging techniques and novel applications for the improved assessment of blood flow and characterisation of cardiovascular disease throughout the human body. We will focus on a select number of applications for the improved evaluation of cardiovascular (aorta, pulmonary arteries, congenital heart disease [CHD], atrial fibrillation [AF], coronary arteries) as well as cerebrovascular disease (intra-cranial arteries and veins). A comprehensive review of the many more applications of advanced flow imaging in the setting of different neurovascular and cardiovascular diseases, however, is beyond the scope of this article and the readers are referred to the literature.

FROM 2D TO 4D

Unlike echocardiography (ECG), the assessment of blood flow in the cardiovascular system with PC-MRI is not limited by acoustic window or probe angle. Traditionally, quantitative blood flow assessment based on ECG-gated time-resolved (cine) 2D PCMRI techniques can include such parameters as peak velocity, retrograde flow (or regurgitant fraction), stroke volume, and shunt volumes (QP/QS). Further indices can also be derived from velocity data, such as pressure gradient through the modified Bernoulli equation to assess, for example, valvular function or aortic coarctation severity.

For routinely used 2D cine PC-MRI, the blood flow velocity is typically encoded in one direction through a 2D section^5,52; however, placement of the acquisition plane remains challenging and can lead to the underestimation of peak velocities if misplaced or not orthogonal to the flow of interest. This is a common occurrence in cases involving complex flow and where changes in flow direction occur throughout the cardiac cycle, such as with valvular stenosis, valvular regurgitation, or complex CHD. These underestimations can be improved by taking into account all flow directions, which is achieved by three-directional encoding of all three principal velocity directions inside a section of interest ⁵³. Alternatively, 4D flow MRI (3D cine PC-MRI with three-directional velocity encoding) enables post-hoc time-resolved three-dimensional visualisation and retrospective quantification of blood flow at any location in a 3D volume ^15-20 (Fig. 2).

Fig. 2.

Chart illustrating the increased complexity and information content in flow data as a function of dimensionality and scan time. The role of methods development is shown along the bottom row, with an impact spanning standard 2D cine PC (left) to 4D flow MRI (right), with volumetric coverage and three-directional velocity encoding. Mag, magnitude image.

Several studies have investigated the comparison of haemodynamic indices estimated using 4D flow MRI compared to standard 2D velocity-encoded PC sequences ^21,54-70, and generally found that aortic and pulmonary flow and velocity indices were significantly underestimated when encoding only one rather than three velocity components (see Fig. 3). On the other hand, good agreement was found between three-directional velocity-encoded sequences, either through a 2D section or in a 3D volume based on 4D flow MRI. These results suggest the importance of three-directional encoding of velocity for estimation of haemodynamic indices, especially when considering quantification of peak velocity or the integrated flow volume; however, integrating PC-MRI with three-directional velocity encoding in routine clinical protocols remains challenging due to increased scan time (acquisition of additional data for velocity encoding along all three dimensions), and thus inability to perform data acquisition in short time and/or during breath-holding. In this context, recent and ongoing developments in sparse sampling techniques, e.g., compressed sensing⁴² or radial acquisition⁴³, as well as multidimensional parallel imaging, e.g., k-t GRAPPA^44-46 (generalised autocalibrating partially parallel acquisitions, Siemens), have shown great potential to accelerate data acquisition and shorten overall scan time.

Fig. 3.

Comparisons of regional flow quantification in the ascending aorta (AAo) of a healthy subject by 4D flow MRI, 2D cine PC with three-directional velocity encoding (2D 3-dir), and standard 2D cine PC with one-directional through-plane velocity encoding (2D 1-dir). The imaging plane used for 2D CINE PC was co-registered with the 4D flow data to ensure flow unification at identical locations in the aorta. 3D streamlines (left) depict the systolic 3D velocity distribution and flow patterns in the thoracic aorta. The graphs (right) clearly show underestimation of systolic peak velocities by 2D 1-dir and close agreement between 4D flow and 2D 3-dir.

Alternatively, respiratory gating can be applied to acquire PC-MRI data during free breathing and minimise respiration artefacts. Most widely available approaches include bellows reading or navigator gating^71-74. A drawback of navigator gating relates to the time it takes to acquire navigator data, which only permits a single or two navigators per cardiac cycle in most situations, often leading to poor prediction of respiratory motion⁷⁴. In addition, detection of motion of the organ of interest is typically based on measurements of remote anatomy (e.g., diaphragm). To address these shortcomings, respiratory self-gating methods have been developed ^75,76,77, which repetitively acquire a subset of data of the organ itself to estimate respiratory motion^66,78. In this context, radial imaging methods offer an elegant approach to inherently acquire motion-related information without time penalty ^78,79. As respiratory gating leads to partial rejection of data acquired during rapid motion, overall scan efficiency is typically reduced by 20–60%, leading to long overall measurement times. As an alternative, advanced retrospective motion correction techniques have been developed that permit correction of the acquired data given an estimate of the underlying motion ^80,81.

Methodological improvements based on k-t parallel imaging or compressed sensing have enabled the acquisition of 2D PC-MRI with three-directional velocity encoding during a single breath hold⁸². Their combination with advanced respiration control allowed the acquisition of cardiovascular 4D flow MRI data within clinically acceptable scan times of the order of 5–15 minutes. Continued developments based on alterative data sampling strategies such as radial or spiral data readout have high potential to further reduce scan times. For example, a recent study showed that the combination of highly efficient spiral sampling with dynamic compressed sensing can achieve major acceleration for 4D flow MRI, which allowed the comprehensive assessment of abdominal 4D haemodynamics in a single breath-hold⁸³.

The increased complexity of PC-MRI data with three-directional encoding and/or volumetric coverage (4D flow) offers the opportunity to obtain more comprehensive and improved information on complex changes of blood flow in cardiovascular disease, compared to standard 2D techniques or echocardiography. A number of investigators have exploited these advantages of multi-dimensional flow imaging to derive new physiological and pathophysiological haemodynamic parameters, such as wall shear stress (WSS) vectors^21,22,84-89, pulse wave velocity (PWV)^29,30, 3D pressure difference maps^{25,26,28,90-92}, or energy loss^13,31,32,93. These advanced haemodynamic measures can provide quantitative information on the impact of vascular disease on aortic or pulmonary blood flow patterns.

APPLICATIONS

Aortic and pulmonary disease

It is important to note that flow imaging is typically performed as part of a standard-of-care aortic/pulmonary imaging protocol, which includes additional MRI techniques for the assessment of cardiac function and wall motion (cine imaging), aortic and pulmonary dimensions and geometry (MR angiography), as well as aortic and pulmonary valve morphology and dynamics (cine imaging). The combination with advanced flow imaging provides a comprehensive assessment of aortic/pulmonary structure and function, which has the potential to contribute to the clinical evaluation of many diseases, such as aortic and pulmonary stenosis (peak systolic velocity) and insufficiency (regurgitant fraction)⁹⁴; aortic aneurysm and/or congenital abnormalities such as bicuspid aortic valve (BAV) or Marfan’s syndrome (outflow patterns, peak velocity, wall shear forces)⁹⁵; aortic dissection (fenestration regions, flow in true and false lumen)⁹⁶; aortic coarctation (peak flow velocity, longitudinal post-repair changes)⁹⁷; pulmonary hypertension (PH; altered flow patterns, pressure estimation)^98,99; and pulmonary branch stenosis (peak velocities, post-stenotic flow patterns).¹⁰⁰

In addition to improvements in imaging methodology, substantial improvements have been made in the analysis of aortic and pulmonary flow data. Emerging applications include post-processing developments for the quantification of effective orifice area^101,102, aortic stiffness-related PWV ^103,104, pressure difference maps ⁹⁷, and WSS^{21,22,105-109}. Modified strategies have also made efforts to use valve tracking to improve the ability to quantify stroke volume and retrograde fraction across the heart valves ^110,111.

Applications of flow imaging in the thoracic aorta have provided evidence that aortic valve disease such as the frequently occurring congenital BAV (affecting 1–2% of the general population^112-115) can result in significant, valve-mediated changes in aortic flow patterns. 4D flow MRI studies have shown that valve abnormalities can be associated with complex flow disturbances, such as altered systolic jet flow patterns (see Fig. 4) and elevated WSS in the aorta^{86,95,106,116-120}. WSS is a known stimulus for arterial mechanotransduction, which can impact endothelial cell function and result in vascular remodelling (e.g., dilation). A recent study⁹⁵ demonstrated that altered aortic haemodynamics can be considered as a physiological mechanism by which valve morphology can influence the phenotypic expression of aortopathy. A subsequent study in BAV patients that included both in-vivo 4D flow MRI and aortic tissue resection in patients undergoing aortic repair provides strong support for this pathomechanism¹⁰⁵. Correlation of aortic haemodynamics with histopathology of the resected aortic tissue showed that regionally elevated WSS patterns were closely associated with abnormal aortic tissue architecture, biomechanics, and protein expression.

Fig. 4.

3D blood flow visualisation (systolic 3D streamlines) based on thoracic 4D flow MRI in a patient with a bicuspid aortic valve with a right–left fusion pattern (RL-BAV, b) compared to an aorta size-matched control subject with normal tri-leaflet aortic valve (TAV, a). Note that BAV resulted in a marked flow jet impinging on the aortic wall compared to TAV.

Other promising applications of aortic 4D flow MRI are related to atherosclerosis and complex aortic plaques (thickness ≥4 mm), which act as potential sources for embolic stroke. Complex plaques in the descending aorta, where they occur most frequently¹²¹, were traditionally not considered a source of stroke as embolisation would require diastolic reverse (upward) flows from the DAo plaque into the brain supplying supra-aortic vessels to establish a mechanism^122,123. Recent 4D flow MRI studies have shown that marked diastolic flow reversal in the DAo is frequently found in patients with aortic atherosclerosis, even in the absence of aortic valve insufficiency^124-126. In fact, there is strong evidence that descending aorta plaques can be sources of embolic stroke via retrograde embolisation. In a study of 94 acute stroke patients, 4D flow MRI was employed for the in-vivo evaluation of aortic 3D blood flow characteristics¹²⁷. In more than 55% of patients, 3D blood flow visualisation revealed reverse flow from descending aorta plaques into the supraaortic arteries supplying the cerebrovascular territory affected by stroke as a possible mechanism for retrograde cerebral embolisation.

In the pulmonary system, flow imaging has been employed to assess WSS ^128,129 and flow vorticity ¹³⁰ in the presence of PH. Reiter et al.¹³¹ used time-resolved 3D velocity information to visualise this flow anomaly and found an unusual flow structure in the main pulmonary artery (MPA) of PH patients¹³¹. A vortex in the primary flow direction of the MPA formed below the right pulmonary artery (RPA). Thus, the time persistence of this vortex was found to be correlated with the degree of hypertension as measured by mean pulmonary artery pressure. Visualisation of complex flow patterns in the pulmonary arteries may thus have the potential to provide a noninvasive vortex-based diagnosis of PH.

Complex CHD

The most common congenital disorder, CHD occurs in an estimated nine per 1000 live births worldwide^132,133. CHD can be associated with complex alterations of cardiovascular physiology and flow (e.g., hypoplastic heart) and surgical correction or palliation of the defect are often necessary (e.g., Fontan procedure). As a result, a comprehensive assessment and monitoring of longitudinal changes in cardiovascular physiology from birth to adulthood are critical in these patients. Current diagnostic tools, however, are limited by invasiveness (catheter angiography) or provide an incomplete assessment of the often complex haemodynamics in patients with CHD^134,135.

Although techniques such as echocardiography or CT can provide anatomical or flow information, only MRI is capable of providing non-invasive assessment of haemodynamics, cardiac function, and cardiovascular anatomy within a single examinatino^136-139; however, flow MRI in this patient population is challenging and requires time-consuming and operator-dependent placement of multiple imaging sections to study multiple vessels, which can result in misalignment with respect to the flow direction and thus underestimation of velocity¹⁴⁰. The situation is exacerbated in paediatric CHD by small vessel size, high heart rates, and limited ability to perform long/multiple breath holds. As a result, MRI-based evaluation of CHD can require long total examination times (60–90 minutes) and is often performed during general anaesthesia in children under the age of 6–8 years ^141,142. These challenges drive the need for flow imaging with high spatial and temporal resolution, short imaging times, and full coverage of complex cardiovascular malformations ^143-145.

Advances in MRI hardware such as high-field MRI systems >1.5 T for improved signal-to-noise ratio (SNR) and spatial resolution; improved gradient performance for faster sequences; and high sensitivity multi-channel receiver coils, enabling parallel imaging with high acceleration factors have led to promising new applications for flow imaging in patients with CHD. Specifically, highly accelerated imaging with spatio-temporal undersampling and compressed sensing/parallel imaging reconstruction in combination with compensation for respiratory motion (e.g., self-navigator gating or image registration) can be employed to assess flow with greater flexibility (spatial, temporal resolution) and with short scan times. An emerging and promising technique is real-time flow imaging in CHD^146-148, which can greatly simplify flow imaging by data acquisition during free breathing and without the need for ECG gating. In addition, the technique is well suited to assess important respirophasic effects on flow (e.g., changes in venous flow in the Fontan circulation during inspiration versus expiration), which cannot easily be assessed with standard 2D PC flow methods. Whole-heart 4D flow MRI techniques allow for a non-invasive comprehensive assessment of cardiovascular haemodynamics in the heart and surrounding great vessels (see Fig. 5). Although scan times are still long due to full volumetric coverage of the entire heart (10–15 minutes depending on heart rate and respiration control efficiency), the main advantages of whole-heart imaging are that it facilitates the systematic assessment of blood flow in multiple vessels and enables the retrospective analysis of any region of interest within the imaging volume. Previous studies have shown that the technique can reliably identify altered 3D flow characteristics related to the post-interventional status in patients with CHD ^149-152. In addition, 4D flow MRI based flow quantification has been shown to be equivalent or even improved when compared to 2D techniques, while needing less imaging time than the positioning of multiple planes^{58,61,64,67,68}. More recent developments combining parallel imaging or compressed sensing with 4D flow MRI have high potential to reduce scan times to <5 minutes.

Fig. 5.

3D venous and arterial flow visualisation from 4D flow MRI for a patient with Fontan circulation. Venous return from the bilateral superior venae cavae (SVCs) and the inferior vena cava (IVC) flows to the right and left pulmonary arteries (RPA, LPA) by way of the Fontan connection, a surgically created vascular connection that directly routes venous return to the right and left lungs. Surrounding vessels of interest are shown in grey and include the aorta and pulmonary veins. Time-resolved path lines are colour-coded by vessel of origin: flow from the right and left SVCs (RSVC, LSVC) is shown in blue and flow from the IVC is shown in yellow. These results provide information regarding complex flow patterns and flow distribution, including the identification of vortex flow and the quantification of hepatic (or IVC) flow distribution to the RPA (88%) and LPA (12%).

AF

AF affects 33.5 million patients worldwide. The most serious complication, stroke, is generally attributed to embolism of thrombus from the left atrium (LA); however, current clinical risk scores (e.g., CHA₂DS₂-VASc) have limited predictive accuracy, as they are based on upstream clinical factors (age, gender, diabetes, etc.) rather than individual physiological factors implicated in LA thrombus formation.

Studies utilising transoesophageal echocardiography (TOE) have shown that decreased flow in the LA and particularly in the LA appendage (LAA), which is the typical site of thrombus formation, is an independent risk factor for stroke in AF;^153-155 however, TOE requires oesophageal intubation, and is thus an impractical approach as a screening tool in the population. Furthermore, due to inherent technical limitations, TOE cannot fully assess the complex 3D patterns of flow and stasis in the LA and LAA, and does not provide detailed information on LA and LAA 3D shape and geometry. 4D flow MRI can overcome limitations of TOE by providing the ability to measure complex 3D blood flow patterns in vivo. A number of studies have shown that 4D flow MRI could be a promising tool for the comprehensive assessment of atrial haemodynamics.^156-158

As illustrated in Fig. 6, atrial 4D flow MRI can be employed for the non-invasive invivo assessment of atrial flow dynamics with full spatial (3D) and temporal (cardiac cycle) coverage of the atrium. After 3D segmentation of the LA, anatomical maps of left atrial flow dynamics (mean velocity, peak velocity stasis) can provide intuitive visualisation of atrial flow dynamics (Fig. 6b). Specifically, LA and LAA maps of the relative amount of flow stasis (i.e. fraction [%] of LA or LAA with velocities below a certain threshold) can visualise regions exposed to high stasis or reduced velocities and thus potentially heightened risk for thrombogenesis.

Fig. 6.

4D flow MRI and derived 3D PC angiogram (PC-MRA, a) including the 3D segmentation of the LA which was used to mask blood flow velocities inside the LA for the calculation of stasis (% of LA velocities <0.1 m/s) and velocity maps (b). Ao, aorta, PA, pulmonary artery, RA, right atrium.

A number of pilot feasibility studies have demonstrated the potential of atrial 4D flow MRI and derived maps of LA flow dynamics to detect changes in global and regional LA flow dynamics associated with AF, age, and LA volume^158-160. 4D flow MRI detected significant changes in LA haemodynamics and reduced LA velocities in patients with a history of AF, compared to young and age appropriate control groups. Moreover, reduction in LA velocities was significantly associated with the standard-of-care clinical CHA₂DS₂-VASc risk score.

It is important to recognise that 4D flow MRI data are acquired over multiple heartbeats and the resulting images represent a composite of blood flow over the entire acquisition time. As a result, beat-to-beat flow variations in AF patients with cardiac arrhythmia cannot be captured by this technique. A systematic analysis of the impact of beat-to-beat variations on LA flow characteristics is thus not possible, particularly in the presence of arrhythmia. Recent advances in MRI data acquisition strategies allow for real-time assessment of beat-to-beat variation in the presence of AF and breathing exercises (such as the Valsalva and Mueller manoeuvres) ^8,161,162. To achieve sufficiently high frame rates needed for real-time flow acquisitions, optimised and accelerated 2D flow imaging pulse sequences have been developed, which combine effective data readout modules (e.g., echo planar imaging, EPI) with data undersampling and parallel imaging data reconstruction (e.g., compressed sensing, k-t-acceleration in spatial and temporal dimension). In addition, shared velocity encoding^10,163 has been proposed to further improve temporal resolution and is based on the concept of sharing sets of full k-space data between adjacent time frames in order to double the effective frame rate. Based on these developments, 2D real-time flow imaging with through-plane velocity encoding can be performed with temporal update rates on the order of 30–50 ms (see Fig. 7).

Fig. 7.

Real-time imaging of flow in the LA based on 2D PC with EPI data readout, shared velocity encoding, and spatiotemporal imaging acceleration (temporal resolution per image = 50 ms). Data were acquired without ECG gating and through-plane velocity encoding (total scan time=10 seconds, venc = 80 cm/s). Data analysis included the calculation of LA flow in region of interests delineating the LA boundary. Resulting flow-time curves at mid-atrial short axis level show flow changes over multiple heart beats in AF patients in sinus rhythm (a) and during arrhythmia (b).

The combination of metrics of persistent multi-beat flow patterns (4D flow: velocity histograms, stasis) with measures of beat-to-beat flow variations (real-time flow imaging) may have the potential to better understand the impact of AF on LA and LAA haemodynamics and thus thromboembolic risk.

Coronary artery flow

A non-invasive technique for measuring coronary flow would be highly desirable in the management of coronary circulatory disorders. For example, exercise testing and invasive catheter angiography are often used to induce coronary microcirculatory hyperaemia, thereby increasing coronary blood by up to fourfold. The ratio of stress versus rest coronary flow (or fraction flow reserve, FFR) can be used to quantify the physiological significance of coronary stenosis.

The major challenge in using MRI for the assessment of flow in coronary arteries includes the need for high spatial resolution, due to their small size. Additionally, coronary flow MRI should resolve and/or provide the ability to follow the tortuous path and highly dynamic cardiac and respiratory induced motion of the coronary arteries during the cardiac cycle. Several studies have addressed these challenges and successfully acquired coronary flow images by using advanced imaging acceleration techniques (see Fig. 8) or more efficient non-Cartesian data sampling strategies, such as spiral trajectories in combination with motion control (e.g., navigator echoes)^164-167. Reports include the validation of PC-MRI in coronary arteries by assessing flow reserve after stent deployment in comparison with intracoronary Doppler ultrasound ¹⁶⁸. Another study showed good agreement for MRI-derived coronary flow reserve compared to myocardial blood flow obtained by positron-emission tomography (PET) ¹⁶⁹. To date, only few clinical applications have been reported. Among them, studies include the assessment of myocardial flow reserve in patients with coronary artery disease (CAD) after stent implantation in the proximal left anterior descending artery (LAD) by Saito et al.¹⁷⁰; the comparison of coronary flow during rest versus exercise stress in healthy adults and CAD patients by Hays et al.¹⁷¹; and profiling of CAD by combining MRI-based assessment of vessel wall remodelling and related myocardial blood flow by Jahnke et al.¹⁷².

Fig. 8.

Coronary 2D PC MRI with through-plane velocity encoding. Data were acquired using k-t acceleration with a high acceleration factor of R=5 which allowed for data acquisition during a single breath hold and an in-plane spatial resolution of approximately 1 mm². The images show magnitude and flow data for peak coronary flow during early diastole. LV, left ventricle; RV, right ventricle; LAD, left anterior descending coronary artery.

Although methodologically challenging, comprehensive MRI protocols including coronary flow assessment could provide a promising non-invasive approach for the direct assessment of coronary vessel wall remodelling, and the resultant pathophysiological consequences on the level of epicardial coronary and myocardial blood flow in patients. Further improvements in hardware and imaging at high field strength (3 T and above)¹⁷³ are expected to enhance the spatial resolution needed for reliable coronary flow imaging methods. Nevertheless, coronary motion remains a major challenge. As a result, the measurement and quantification of coronary blood flow is easier to perform at the larger and more easily identifiable coronary sinus compared to whole-heart flow quantification. Retrospective motion correction for navigated 2D cine velocity mapping ⁷⁴ or data acquisition using readout modules less sensitive to motion (radial, spiral) are promising but further developments and studies evaluating the reliable assessment of flow in all coronary segments are warranted. If successful, an emerging application would be the quantification of FFR, which is currently performed invasively (catheter) or using a combination of computed tomography (CT) angiography and computational fluid dynamics (CFD) ^174,175. MRI could provide a valuable non-invasive and radiation free alternative by combining coronary MRA with rest and stress coronary flow imaging.

Cerebrovascular flow

In clinical practice, transcranial Doppler ultrasound is routinely used for cerebrovascular flow measurements; however, the technique is operator-dependent and significantly limited by the available acoustic windows of the head. 2D PC-MRI can provide reliable flow measurements in large intracranial arteries and veins, not limited by location. Challenges for using 2D PC-MRI for flow measurement include small and tortuous vessels¹⁷⁶, complex vascular anatomy, and the need for manual placement of 2D imaging planes in multiple vessel segments. Recent improvements in intracranial flow MRI include automatic positioning of 2D planes for flow quantification based on a previously acquired angiogram ¹⁷⁷ or the application of 4D flow MRI, which additionally offers 3D blood flow visualisation and retrospective flow quantification. Emerging applications of cerebrovascular flow imaging include the haemodynamic evaluation of intra-cranial aneurysms, arteriovenous malformations (AVM), and intracranial atherosclerotic disease (ICAD), as well as venous flow.

A large number of studies investigating flow patterns in intracranial aneurysms were based on CFD ^109,178-187 techniques in conjunction with subject-specific geometries extracted from medical images ^{179,184,188,189}. Findings from these studies revealed a wide variety of complex intra-aneurysmal flow patterns that were strongly dependent on patient-specific vascular geometry. As an emerging pattern, a number of studies showed that changes in WSS along the wall of intracranial aneurysms may be associated with risk of aneurysm growths or rupture^109,190; however, CFD has limitations, such as assumptions concerning blood properties, boundary conditions, and vessel properties ^191,179,192. As an alternative, 4D flow MRI is increasingly used to assess intra-aneurysmal 3D haemodynamics in vivo. Several groups have reported the successful measurement and evaluation of intra-aneurysmal flow and WSS in patient feasibility studies ^{109,185,193-198}, indicating the potential of flow MRI to assist in the classification of individual aneurysms pre-intervention.

In patients with cerebral AVMs, the pathological vascularisation (direct shunting of blood from arterial to the venous sides without an intervening capillary bed) leads to abnormal haemodynamics. Flow information is potentially valuable for a better understanding of the impact of a focal AVM on the flow redistribution in the brain and/or in treatment planning by attempting to identify the feeding arteries with highest flow, enabling efficient and targeted embolisation treatment. 2D PC-MRI has been used in a number of studies to quantify flow in pre-defined vessels, such as feeding arteries. Earlier reports demonstrated that flow MRI could successfully be used to visualise and quantify flow and velocity¹⁹⁹, to compare flow velocity in feeding and contralateral vessels and following embolisation treatment²⁰⁰, and to compare total cerebral arterial inflow (basilar and internal carotid artery flow rates) ²⁰¹. More recently, 2D PC-MRI in combination with automated 2D plane placement at multiple anatomical regions was employed to assess haemodynamic changes after AVM embolisation²⁰². To date, there are only a few studies investigating the diagnostic value of 4D flow MRI for the assessment of AVM flow characteristics. Recent reports include the quantification of flow and WSS in patients using a highly optimised radial 4D flow technique ^203,204. Additional studies demonstrated the potential of 4D flow MRI for the evaluation of global and regional AVM flow characteristics as illustrated in Fig. 9 ^205,206. The findings showed that 4D flow MRI can assess treatment-induced changes in cerebrovascular flow distribution and were able to demonstrate significant associations between 4D flow metrics, cerebral perfusion indices, and AVM risk factors such as the Spetzler–Martin grade²⁰⁶.

Fig. 9.

Comparisons of time-integrated 3D path lines and regional flow quantification between a 41-year-old male normal volunteer and a 49-year-old male patient with cerebral arteriovenous malformation (AVM). The path lines of the normal volunteer (a) show symmetric and coherent flow patterns in the cerebral arteries, whereas the path lines of the AVM patient (b) exhibit disturbed cerebral flow patterns and the velocity in the feeding right middle cerebral artery (RMCA) is much higher than the contralateral counterpart. The peak velocity (c) and flow-time (d) curves at the RMCA M1 segment (yellow arrows) quantitatively demonstrate higher blood flow and velocity in the AVM feeding artery compared to the same artery in the normal volunteer. Note that the average length of the cardiac cycle was shorter for the AVM patient compared to the normal volunteer during the scan.

Intra-cranial atherosclerotic plaques can alter local and global haemodynamics (particularly proximal or distal to stenosed vessels). Currently, intracranial haemodynamic disturbance in patients with ICAD is primarily assessed using Doppler ultrasound. 2D PC-MRI has also been used to classify stenosis severity, predict risk of recurrent stroke, and detect in-stent restenosis after stent placement;^207-210 however, very few studies have been performed to characterise the 3D blood flow disturbance and flow redistribution across the major cerebral arteries in patients with ICAD. An early study by Hope et al.¹⁹⁵ reported that TOF MRA overestimated the degree of stenosis and that 4D flow MRI velocity measurements could improve accuracy of diagnosis, when compared to catheter angiography. It should be noted that current flow imaging techniques (2D and 4D) are also limited by insufficient spatial resolution for the characterisation of blood flow at sites of critical or severe stenosis. Instead, post-stenotic flow is typically used to represent the regional flow in the stenotic artery; however, the accuracy of flow quantification in the smaller arteries (e.g., anterior/posterior cerebral arteries) may be compromised by partial volume effects. Higher magnetic field strength (7 T) with increased spatial resolution may be required for improved flow assessment in the smaller vessels.

Cerebral venous blood flow attracts less attention compared to arterial flow, but the venous system is of great importance to the understanding of some cerebrovascular diseases, such as cerebral venous thrombosis or cerebral venous sinus stenosis in patients with intracranial hypertension. The assessment of intra- or extra-cranial venous flow can be particularly challenging as there are several factors that can affect flow measurement ²¹¹, such as head and limb position²¹², respiration, heart rate, or hydration status. Nevertheless, 2D PC-MRI has been applied to quantify physiological cerebral venous flow²¹³, for the concomitant analysis of arterial, venous and cerebrospinal fluid flows²¹¹, the assessment of intracranial venous haemodynamics in normal and patients with cerebral venous thrombosis²¹⁴, and to evaluate physiological variation in dural venous sinus flow²¹⁵. Volumetric measurement of cerebral venous 3D flow dynamics^216,217 and assessment of cerebrospinal venous flow using radial 4D flow MRI²¹⁸ have also been reported.

Limitations of intracranial flow MRI include the spatial resolution and low sensitivity to slow blood flow. For the detailed evaluation of cerebrovascular flow, it would thus be desirable to gain quantitative information on both lower venous and high arterial blood flow velocities, which can differ by one order of magnitude even in normal subjects (10 versus 150 cm/s). Current flow MRI techniques, however, measure blood velocity (v) based on a single user-defined velocity sensitivity (venc), usually set above the expected maximum velocity. This single-venc acquisition can result in either velocity aliasing for unexpected higher blood flow velocities (v>venc) or high noise levels for slow flow (v<<venc). Current MRI protocols thus lack the dynamic range to reliably assess the full velocity spectrum. To improve upon this limitation, several studies have investigated multi-venc approaches based on several flow acquisitions with a set of two or more vencs ^219-221. The resulting high-venc data can then be used for complete anti-aliasing of the low-venc data while maintaining the favourable velocity-to-noise characteristics of the latter. Alternative approaches include five-point balanced flow encoding to reduce noise and aliasing in phase images ²²², or varying velocity encoding during systole or diastole ²²³.

Studies performed at ultra-high field (7 T) have shown that the assessment of smaller calibre-vessels as well as small aneurysms can be improved (e.g., posterior cerebral artery) and their pulsatility index can be measured^173,224,225. In addition, several advanced acceleration methods for the improvement of spatial resolution and/or reduction of total scan time have been successfully explored including k-t BLAST²²⁶, k-t GRAPPA, k-t SENSE²²⁷, radial²²⁷ and spiral²²⁸ sampling schemes as well as compressed sensing approaches^229,230.

Fig. 10 shows results based on k-t accelerated dual-venc 4D flow MRI at 7 T and illustrates that the combination of the novel developments described above can enable flow imaging with significantly improved dynamic range for the assessment of both low venous and high arterial flow velocities within a single scan.

Fig. 10.

Intra-cranial dual-venc 4D flow MRI at 7 T (spatial resolution = 1.1 mm³) for the assessment of cerebral vascular flow and angiography. Example flow images (top row) illustrate substantial velocity aliasing for the low venc data (left) but excellent velocity–background contrast and depiction of regions with low flow velocity. Conversely, velocity aliasing is absent in the high venc data (mid) but velocity background noise is increased and low flow regions are less well delineated. Dual-venc reconstruction of flow data (right) successfully removed velocity aliasing while maintaining the favourable dynamic range of the low venc data. Resulting 3D PCMRA maximum intensity projections (MIPs, lower row) of the entire 4D flow data corroborate these findings. The depiction of the cerebral vasculature (especially small arteries and veins with slow flow) is considerably improved for the dual-venc reconstruction (right). Data were acquired at UM CMRR (Minneapolis) using an actively shielded 7 T system with high-performance gradients (G_max 72 mT/m, slew rate 200 mT/m/ms) and a 32-channel head coil.

DISCUSSION AND CONCLUSIONS

In the past decade, substantial advances in flow imaging (including imaging techniques and data analysis) have been made that have led to a number of emerging applications for the improved assessment of cardiovascular and cerebrovascular diseases. The application areas presented in this review reflect the view of the authors. A comprehensive review of the many more and growing applications of advanced flow imaging in the setting of different neurovascular and cardiovascular diseases, however, was beyond the scope of this article and the readers are referred to the literature.

A number of hurdles and limitations for the more widespread application of the emerging flow imaging techniques presented in this review exist. Eddy currents and phase offsets remain longstanding technical challenges that introduce measurement error across all phase-contrast approaches, including the advanced methods presented here²³¹. These errors, which manifest as phase shifts on a regional, global, and temporal basis, will compromise the ability to accurately quantify integrated parameters such as stroke volume^231-234. A number of studies have investigated various correction approaches^57,234-236; however, there is no widespread agreement as to a universal correction protocol. The lack of standardised analysis tools for regional flow quantification and analysis of PC data with three-directional velocity encoding and/or 3D volumetric coverage is currently a major hurdle for their more widespread clinical application. In addition, many promising techniques for advanced sparse sampling and timing acceleration have been developed in the past decade; however, current limitations include a wide disparity of techniques used for sparse sampling and imaging acceleration, as well as a lack of standardised quality metrics that define the optimal use of these methods for clinical applications.

To further the availability of these techniques and facilitate their translation into the clinic, workflow efforts towards standardisation across different MRI vendor platforms and data analysis software are warranted. These include fundamental aspects such as the consistent presentation of flow directions in MRI images and the clear definition of data header fields with flow relevant information (e.g., velocity sensitivity) across all platforms. In addition, software solutions should report standardised flow metrics and include quality control to provide the user with feedback regarding phase offset errors and corrections that have been applied to the flow data. To achieve these aims, multicentre studies are necessary to establish the repeatability of various aspects of the technique across centres and identify the most robust and reproducible flow metrics. In addition, further longitudinal studies of emerging novel haemodynamic measures, such as WSS or stasis maps, are needed to evaluate their diagnostic value for patient risk stratification and therapy management.

In summary, flow imaging with MRI has undergone and continues to undergo a substantial transformation, from simple techniques measuring one-directional blood flow velocities at a specific location to a more compressive diagnostic tool that can assess 3D blood flow, quantify real-time beat-to-beat respirophasic changes in flow, or derive advanced metrics of cardio- and cerebrovascular haemodynamics, such as pressure gradients or WSS.