Abstract

Three dimensional (3D) delayed gadolinium enhancement cardiac magnetic resonance imaging, includes advantages such as: high resolution entire heart coverage without slice–misregistration, shorter or no breath-holds, multi-planes post reconstruction and higher signal to noise ratio (SNR). These are among some of the reasons why it has gained significant popularity within the daily clinical practice; however, parallel imaging, inefficiencies in navigator echoes and prolonged scan times are some of the challenges affecting overall quality of 3D cardiac imaging [1].

1. Introduction to Delayed Gadolinium Techniques

Delayed gadolinium enhancement is currently the gold standard in identifying and assessing myocardial viability, infarction and fibrosis. The development of (IR)-technique based on the premise that myocardial scar enhances after contrast, has led to the significant growth of MRI for diagnostic of cardiac imaging (Figure 1) Because of this improvement and depending on the acquisition timing post contrast, DGE can be utilized for early and late gadolinium enhancement techniques ⁹. The intrinsic gadolinium wash-in/wash-out dynamics within the extracellular space makes it possible to ^{1, 9}:

• Determine hyperemic myocardium.

• Identify and quantify scar, size, shape and fibrotic myocardium.

• Detect partially infarcted myocardial tissue as predictor for post MI mortality.

• Visualize mural thrombus.

• Differentiate between ischemic vs non-ischemic cardiomyopathy patterns.

After administration, Gadolinium contrast normally enters the interstitial space and is excluded from intracellular space by the cell membrane. Normal myocardium enhances slightly, since the overall tissue contains a small fraction of extracellular space. However, infarcted cardiomyocytes in an acute phase, allow a larger amount of contrast to permeate through their ruptured membrane and enhances to a greater extent. Furthermore, chronic scarring, composed of collagenous tissue has a much larger extra cellular volume (ECV) and in combination with slower extravasation (wash-out) is specifically visualized in a later stage due to significant gadolinium contrast retention (b) ².

Our current DGE sequence consist of a multi-breath-hold 2D segmented T1-w fast spoiled gradient echo (turbo-FLASH) with a 180° inversion pulse, that flips (10-15 minutes post contrast) the net magnetization of all tissues into the -z axis along the longitudinal direction (a).

Factors that can influence the appropriate TI timing are type, volume, time and field-strength; contrast types with higher molarity and/or relaxivity (R1) characteristics and administered at higher volumes (>0.2 mmol/kg), will cause the TI time to shorten, as compared to normal myocardium ⁴. Increasing the time between contrast administration and data acquisition leads to longer TI time which decreases the signal difference between normal and abnormal myocardium ⁴.

From 2D to 3D-BH to 3D-FB

DGE sequences nowadays can either be 2D or 3D, FLASH or True-FISP, single shot or segmented, with single or several breath-holds, or as free-breathing (FB).

The traditional 2D-IR-GRE sequence requires multiple (10-14) BH in different predefined planes resulting in long scan times with non-isotropic resolution ⁵. Images are collected slice by slice with thickness (5-10 mm), generally unable to cover the whole heart due to slice gaps, as a result the images have a low SNR and spatial resolution (limited by duration of a breath-hold). Repetitive and prolonged breath-holds in especially weak, fatigued, or non-compliant patients can cause slice misregistration due to variability in BH position ⁶.

To allow time for transverse magnetization to fully recover, data in current 2D sequence is acquired every other RR thus improving signal intensity and becoming less susceptible to irregular heartbeats (arrhythmias) (Figure 5). Imaging every other RR, however, doubles the scan time and inadvertently makes it more difficult for the patient to maintain proper breath-holding during the scan ⁷.

Three-dimensional data acquisition with single or several (1-4) BH provided some compensation for long scan times. Triggering was done on every RR to keep scan duration short at the expense of CNR due to incomplete recovery of the net magnetization. With breath-holds longer than 20 seconds, this technique however, was not very useful for sick and vulnerable patients ⁷.

As heart rate increases and in patients with varying RR due to arrhythmia, the need to trigger on 2 RR or 3 RR increases, doubling and tripling the scan time and leading to poor image quality ⁸. However, with a slightly better temporal and spatial resolution it was still possible to:

• Scan a whole heart continuously (without gap).

• Increase SNR compared to 2D.

• Reformat (isotropic voxel) images in several planes ⁹.

Originally used for imaging of coronary arteries, 3D-free breathing (FB) whole heart sequences have become robust, mainstream, and nearly standard practice in daily protocols. It is one of the most challenging applications of MRI due to: flow, motion and anatomy requiring specific demands for planning, patient cooperation, spatial resolution, signal-to-noise (SNR) and accurate cardiac and respiratory motion correction. The benefits of 3D-FB imaging including no demanding breath-holds, fewer artefacts, full coverage with no gaps and higher spatial resolution, (Table 1) has made it possible to:

• Detect and define even smaller scar tissue size, shape, and pattern.

• Limit partial volume effect that can cause scar and grey zone overestimation.

• Accurately measure degree and extent of transmurality.

• Identify regions of microvascular obstruction (MVO) in acute MI cases and post (PCI).

• Assess scar in thin wall cardiac anatomy such as atria and/or right ventricle

• Guide radiofrequency ablation therapy in patients with A-Fib.

Create 3D models to assess complex anatomy for surgery in patients with CHD ^{1, 10}.

The main issue with 3D-FB DGE, is the increase in scan duration due to demands for higher spatial resolution. Recent efforts to develop and implement improved read-out schemes, motion compensation methods and faster acquisition are discussed in Chapter 2 and Chapter 3.

DGE can either be applied early (within 1-5 minutes post Gadolinium); EGE imaging or later (10-15 minutes post Gadolinium); LGE imaging. These acquisition time differences provide additional contrast characteristics of myocardial pathology, giving EGE a potentially supporting role for improved:

• Detection of inflammation such as oedema, hyperemia and capillary leaks from acute myocarditis or in acute MI (based on the Lake Louise criteria) ¹¹.

• Highlighting areas of thrombus presented by lower signal intensity (dark) compared to higher signal intensity (bright) of normal myocardium at (TI 440-600msec) ^{6, 12}.

• Differentiation of acute and chronic infarction scar in MVO.

Patients with an acute MI undergo a percutaneous coronary intervention (PCI) to restore the blood flow to the ischemic area which, contains the necrosis, allowing for recovery of the contractile function of the myocardium. This procedure can fail to re-perfuse the small vasculatures of the infarcted area. The presence and/or size of the microvascular obstruction (MVO) is usually associated with poor recovery of cardiac function and increased mortality.

In areas where this myocardial capillary network is poor, disrupted, or nonexistent, contrast enhancement, limited by T1 tissue relaxation, will have an insignificant impact on signal intensity. Choosing a long TI minimizes the signal from MVO and at the same time causes normal surrounding tissues to experience T1 shortening and become hyper-enhanced making this affected area visible as dark (hypo-enhancement, no-reflow) reflecting the core of the necrosis.

2. Motion Correction: Cardiac and Respiratory Motion

The challenging and demanding nature of CMR compared to other organs, relates to complex motion of the heart caused by its pulsating chambers, twisting of the myocardium and blood flow in different directions. The heartbeat consists of 1 cycle divided by 2 phases namely: contraction (systole) and relaxation (diastole), followed by a heart rate dependent “resting” period ¹³.

During a typical CMR, data is being collected over multiple cardiac cycles with the assumption that activity is consistent for every heartbeat, meaning at the same time points or phases during each RR-interval. The reality is that motion, heart rate and breathing, varies naturally on a beat per beat basis due to normal sinus rhythm. During systole, the heart not only shortens in the long axis and contracts radially, but the apex and base also rotates opposite of each other (Figure 10). The diastolic phase on the other hand shortens with increased heart rate whilst the systolic phase remains mostly constant ¹⁴.

Acquiring motion-free images at end-systole or mid-diastole also known as “anatomical MR imaging of heart” is obtained by synchronizing the pulse sequence to the patient’s heart through an ECG vector cardiogram or pulse triggering, known as ECG-gating ¹⁵.

The scan triggers on QRS-complex through leads attached on the patient’s chest or through peripheral blood flow of the fingertip (Figure 11). A pulse trigger can be used as an alternative whenever a poor ECG signal is detected, for instance in post-acute MI or in severe arrhythmias. Pulse-triggering has its own limitations related to inconsistencies due to motion of the fingers, cold hands, weak signal from peripheral artery diseases and delays in triggering from the pulse-wave travelling from the heart to the fingertip ¹⁵.

Accurate synchronization of MR data to reduce cardiac motion artefacts and subsequently improve image quality, starts by tracking and triggering on the heart's electrical activity. This electrical pulse depicted by the ECG as a wave form, can further be divided into different peaks, each one describing various phases of the cardiac cycle (Figure 12).

The key parts of the ECG waveform are:

• P wave: changes from – to + charge in membrane potential= (depolarization) cause atrial contraction.

• QRS complex: due to restoration of membrane potential (repolarization) atria relaxes followed by onset of ventricular depolarization and its subsequent contraction; systole at 50msec. after R wave lasts around 150-300msec.

• T wave: after ejection of blood from heart, the myocardium relaxes, and ventricular re-polarization starts and remains in diastole until next QRS complex.

• T-P segment: includes the diastasis/quiescent phase: period of minimum cardiac motion representing an ideal window for optimal imaging of the heart.

The systolic and diastolic phases are further divided into atrial systole, isovolumic contraction, ejection, isovolumic relaxation, rapid ventricular filling and slow ventricular filling (diastasis) ¹⁷.

A. Atrial Systole

This stage marks the end of diastolic phase when the atrioventricular valves are open and the semilunar valves are closed; the atria’s contract and eject blood into the ventricles. Atrial depolarization is finalized. The end of P wave marks the beginning of atrial systole and subsequent depolarization from atria to AV-node highlights the PR segment ¹⁷.

B. Isovolumic Contraction

As the ventricles are filling up, pressure is rapidly increasing, initiating the contraction. Pressure in ventricles exceeds the pressure in the atria closing the atrioventricular valves (Mitral and Tricuspid valve). The semilunar valves (aorta and pulmonary valve) at this stage are still closed since ventricular pressure has not surpassed the threshold within the aorta and pulmonary arteries. Ventricular depolarization causes the QRS- complex ¹⁷.

C. Ejection

The pressure in ventricles by now has exceeded the pressure in both aorta and pulmonary artery causing the semilunar valves to open. Ventricular contraction forces blood to be ejected through the aorta and pulmonary artery while the atrioventricular valves remain closed.

Ventricles completely depolarized at ejection causing the ST. Start of repolarization causes T wave ¹⁷.

D. Isovolumic Relaxation

Marking end of systole, ventricles are relaxed and the pressure drops. Higher blood pressures in the aorta and pulmonary artery through the ventricles causes the semilunar valves to close at the end of systole.

Completion of ventricular repolarization at end of T wave: (1^st MR anatomical imaging phase) ¹⁷.

E. Rapid Ventricular Filling

At this early diastolic phase, ventricular pressure keeps dropping below the atrial pressure causing the atrioventricular valves to open. Blood from atria now rapidly flows into the ventricles. No electrical activity produced by cardiac cells ¹⁷.

F. Slow Ventricular filling: Diastasis

Atrioventricular valves remain open while semilunar valves are closed. During this mid diastolic phase (2^nd MR anatomical imaging phase), only a tiny bit of venous blood is filling the ventricles, where pressure in both ventricles is almost zero and both systemic and pulmonary pressure systems are steadily decreasing. At the end of diastasis depolarization from SA- node covers the atria producing the P wave ¹⁷.

A regular and consistent heart rate (HR) in beats per minute (bpm) and rhythm (cardiac cycle) identified by the RR-interval in milliseconds (msec.), are desirable when it comes to cardiac MRI (CMR). Their interconnection and relationship are presented by the following formula below (Figure 13). Note that as the heart rate increases, the RR-interval and thus acquisition window decreases resulting in less time to collect data.

Patients undergoing cardiac MRI may either have slow, fast and or variable heart rhythms (arrhythmias). The key is being able to identify, interpret and differentiate these activities, rates, rhythms, and irregular patterns and subsequently tailor the acquisition techniques and parameters to achieve optimal image and high-quality exams.

Below are the most common conditions and patterns typically seen in patients referred for CMR:

A: Sinus Bradycardia: heart rate < than 60 bpm; impulse at S-A node at a slow pace

B: Normal Sinus: heart rate between 60-100 bpm; normal complex.

C: Sinus Tachycardia: heart rate >100 bpm; normal complex related to physical exercise, stress, or congestive heart failure.

D: Sinus Arrhythmia (irregular rhythm) common in all ages, where HR ↑ with inspiration and ↓ at expiration.

E: Atrial Fibrillation (A-Fib) irregular baseline and rapid ventricular response; possibly due to pericarditis and/or atherosclerosis.

F: Premature Ventricular Contraction: time interval R peaks is a multiple of RR-intervals; abnormally early ventricular contraction.

G: Ventricular Tachycardia: due to ischemia and myocardial infarction resulting in a high rate (>120 bpm) of ventricular muscle activity.

H: Bradyarrhythmias and tachyarrhythmias are among the most challenging combinations ¹⁹.

Capturing the heart free of motions at an ideal rate of 60 bpm, requires “shooting” images in a few tens of milliseconds or even shorter at increasing heart rates. The scanner with its longer acquisition, limited by repetition times (TR) and number of phase encoding steps needed to fill a K-space to obtain an image, is unable to satisfactory “freeze” the beating heart. Image acquisition is accomplished by splitting each cardiac cycle into shorter frames, (segments or phases) representing a specific time point and collecting these signals, segment by segment over a series of multiple heart beats. This is called Segmentation (Figure 14) ⁹.

Acquiring images of a specific anatomy, requires synchronization to and segmented data collection of a particular cardiac phase. This is called prospective still imaging (single-phase/morphology) triggering technique (Figure 15).

After detecting the R wave, prospective (tissue characterization) synchronization, relies on the following elements and distinctive time frames.

1). Trigger delay (TD)

2). Acquisition window (AW)

3). Trigger window (TW)

1. A time delay between the R wave and the start of acquisition is called trigger delay (TD). This parameter can be set for data acquisition of a part of the cardiac cycle ranging between 0-50msec. for systolic imaging or 150-300msec. for diastolic imaging, or to optimize image quality in situations of inconsistent and variable RR-interval ²⁰.

2. The acquisition window (AW) represents the duration of data collection of either single or multiple RF excitations and echoes produced by the pulse sequence, and it covers the middle 85-90% of the RR-interval for prospective studies ³.

3. If subsequent heartbeats are required for data collection, then the AW needs to fit in every following RR-interval. Since HR’s are rarely consistent, it is advisable to leave a small interval (10-15%) called trigger window (TW) between end of data sampling and the next R-wave. If the QRS complex is initiated slightly early, it will still be detected ^{3, 18}.

Examples of still imaging applications are:

• Black Blood spin echo: double IR- 2D SE for anatomy and tissue characterization.

• Black Blood STIR: triple IR-2D SE for tissue characterization/oedema.

• LGE: IR-2D/3D: Gradient for tissue characterization/myocardial scar (Figure 17).

• Coronary MRA: 3D Gradient for coronary vessel lumen (Figure 18).

Overview of characteristics of prospective single-phase imaging are:

• covers less than entire cardiac cycle; not suitable for imaging end-diastolic functions, mitral or tricuspid valve functions.

• less sensitive to arrhythmia and variable heart rates and ectopic beats.

• manually adjustable AW.

• image acquisition and reconstruction are simpler and easier.

• cardiac phases are determined by segments.

The segmented K-space acquisition described so far is one of the methods for data collection. However, instead of collecting smaller portions of K-space lines over several heartbeats, Single-Shot (Real-time imaging), useful in cases of severe arrhythmias, makes it is possible to collect all data needed to make an image in 1 single heartbeat ¹³. The schematic difference between segmented and Real-time imaging are shown in Figure 19.

One key factor of most fast-imaging techniques is to increase the number of K-space lines per heartbeat. Another way to decrease the acquisition time is to reduce the number of phase encoding steps needed to reconstruct an image. Opting for fewer K-space lines, however, leads to lower overall signal to noise ratio (SNR).

Setting up a cardiac imaging study even more so than any other anatomy, entails a trade-off between temporal and spatial resolution, SNR, scan time and duration of breath-hold. Traditionally 1 line of K-space (non-segmented) was acquired per heartbeat. As an example: imaging a cardiac anatomy with spatial resolution of 2 mm² and FOV of 256mm² ~ (nPE=128) and a pulse sequence with TR=10ms, would require 128 heart beats at 60 bpm ~ (RR=1000ms) resulting in a scan time of 128 sec or over 2 min! (nPE/VPS) × RR. Images would be strongly affected by motion artefacts. Increasing the numbers of echoes/views/lines per K-space to 8 would result in a more acceptable scan of 16 sec (128/8) × RR, but at the expense of other parameters ^{3, 9}.

Protocol optimization and image quality improvement tailored to the patient conditions, requires the technologist to identify, define and understand the relationships of the following key terminologies and formulas:

• Repetition time or “True” TR (echo spacing): time (typically 3-8ms) between 1 or successive RF excitation pulses or gradients needed to refocus the magnetization and thus produce an echo.

• Temporal resolution or Tres. (“effective” or “reported”): capturing the fastest phases of systole and diastole means high temporal resolution with a small temporal window (<50msec.) at a nominal heart rate of 60 bpm, which is equivalent to 20 cardiac phases for a “smooth jerk free” cine of the heart. Too low of a temporal resolution leads to inability to capture the high peaks of systole, therefore resulting in underestimation of ejection fraction, blurring of the myocardium and poor visualization of wall motion abnormalities. Below depicts 9 cardiac phases compared to 18.

• Frames/segments/cardiac phases: this somewhat confusing terminology represents individual images of the cardiac cycle formed by groups of data point “snapshots” and is allocated at specific time points per heartbeat. The larger the number of cardiac phases the higher the temporal resolution of the cine series.

• Spatial resolution: the ability to delineate the border between blood pool, myocardium and scarring within the wall itself defined by size and number of pixels (2D) or voxel (3D), determined by the extent of coverage of k space area. The larger the coverage, the higher the resolution and thus the smaller the structures that can be visualized; however, at the expense of prolonged scan time. Moreover, reducing the scan time at the expense of limited voxels across the myocardium ultimately causes blurring ³.

• Views per Segments or VPS (K-space lines/lines per segments/ETL: echo train length): GRE sequences with the shortest repetition time possible, can produce multiple echoes or groups of K-space lines after each RF excitation pulse. More VPS reduces scan time and shortens the breath hold but increases the temporal window which in turn decreases the temporal resolution and limits the number of phases that can be acquired. On the other hand, as HR increases, lowering the number of VPS guarantees that the temporal resolution can capture the fastest motions of the heart ^{9, 26}.

• Number of Phase Encoding; nPE: number of steps (resolution) in the Y- direction of K space needed to complete an image and directly influencing acquisition time.

• Acquisition time (breath-holds) in heartbeats or seconds: duration of the scan time resulting from # of phase encoding steps divided by # of lines per K-space.

○ Temporal resolution Tres = TR x VPS

○ Cardiac phases or segments = AW/TRes

○ Acquisition time = nPE/VPS (heartbeats)

○ Acquisition time = nPE/VPS x RR-interval (seconds)

○ Acquisition window = RR-interval – (TW+TD)

Example: VPS = 7, TR= 6msec., AW=750msec. and phase encoding steps (nPE) = 128.

Temporal resolution is 42msec. with an acquisition time of 18 heartbeats. To increase patient’s comfort, the imaging time is shortened by increasing the segments (VPS) to 11. The new acquisition time is reduced from 18 to 12 heartbeats (128/11) however at the expense of increased temporal resolution (42 to 66). Too high of a temporal window will ultimately lead to risk for increased blurring due to motion artefacts (Figure 24) and less images (cardiac phases) obtained per cardiac cycle (750/42=18 vs 750/66=11) ^{9, 23}. This reduction in numbers of cardiac phases not only negatively affects the quality of the views (“jerky motions”) but also limits volume calculations especially at end systole where motions of the heart are faster lasting between 40-50msec.

As “noise” peaks induced by arrhythmia events can mistakenly be interpreted for normal voltage of the QRS-complex, triggering on these peaks instead of the R-wave leads to random data collection at incorrect phases of the cardiac cycle, resulting into blurring, poor SNR and ultimately degradation of the quality of cardiac imaging.

As previously mentioned, the key to obtaining accurate and diagnostic quality images is by being able to recognize these abnormalities from simple irregularities including slow (bradycardia), fast (tachycardia) and irregular (dysrhythmias) rhythms to more complex combinations such as brady-arrhythmias and tachy-arrhythmias and adjusting parameters accordingly.

Bradycardia: a HR below 60 bpm (RR-interval > 1000msec.) implies a longer RR-interval and thus more data that can be acquired. The acquisition time or breath holds however, are bound to be prolonged potentially, leading to respiratory motion artefacts. Solutions involve:

• Minimizing number of phase encoding steps (nPE) by using rectangular FOV: reduces number of heartbeats needed to make an image.

• Increasing the number of VPS: reduces the temporal resolution which is acceptable since the heart rate is slower

• Reducing the matrix (resolution) in phase encoding direction decreases the (nPE).

• Triggering on every heartbeat for DGE (gating factor 1) reduces breath holds by 50%.

Tachycardia: as the acquisition time is decreasing due to HR’s above 100 bpm (RR-interval < 600msec.), a higher temporal resolution (smaller temporal window) is necessary to capture the faster motions of the heart during end-systole and end-diastole. Improving image quality is done by:

• Optimizing the temporal resolution as the mid-diastolic phase gets shorter can be done by decreasing the VPS at the expense of a higher number of heartbeats needed to obtain an image. Increasing the temporal resolution in patients with tachycardia results in breath hold durations like that of patients with normal HR and thus minimizes blurring.

• As the HR increases, the diastolic phase significantly decreases leading to MR data being acquired over several phases and time points. This difference can lead to image blurring and slice-misregistration. Since systole tends to remain unchanged, synchronizing the AW with the quiescent time at end-systole could be an alternative.

• Performing DGE every third heartbeat (gating factor of 3) rather than every other heartbeat (gating factor 2) allows sufficient recovery of magnetization between successive IR pulses. Increasing the gating factor, however, requires an increment of TI time by 30-50msec.

In challenging rhythm abnormalities such as A-Fib, PVC and/or ventricular tachycardia, fast rates combined with irregular and very short RR-interval cause duration of the cardiac cycle to constantly change during each heartbeat. Possible alternatives include:

• Retrospective gating in these cases leads to a mismatch between the reconstructed phases and their corresponding time points within the cardiac cycle resulting in blurring and ghosting artefacts. One possible solution is to switch to prospective gating and choosing an AW which is shorter than the shortest identified RR-interval (Figure 25) ⁹.

• Arrhythmia rejection (AR) (increasing the TW up to 25%) allows data collection to be switched from retrospective to prospective gating, thus reducing or rejecting the range of heart rates over which data from irregular or false triggers of cardiac phases are being collected. The downside of AR is longer scan times/longer breath-holds and worsening efficiency since more time is needed to acquire data from a smaller acquisition window.

• Increasing the VPS in combination with a smaller number of cardiac phases which in turn decreases the RR-interval over which data is collected, leads to fewer mis-triggered data and shorter scan times.

• Capturing optimum temporal windows during mid-diastole becomes challenging. Contour and size of scarring during DGE is optimized by choosing a smaller acquisition window closer to the end-systolic phase which is less sensitive to motion artefacts. With shorter windows at end-systole and TI times possibly longer than RR-interval, proper nulling can be troublesome. An alternative solution is to do PSIR.

In extreme cases where patients suffering from severe arrhythmias either brady- or tachy-arrhythmias or are unable to perform breath-holding or where ECG-tracing despite all efforts is still poor, the only solution is switching to:

• Real-time imaging either gated or non-gated (single shot) sequence. With temporal windows around 100msec., a whole train of single shots images reaching approximately 10 images per second can be acquired for every heartbeat. Therefore breath-holds are not necessary. The trade-off compared to segmented acquisitions is limited temporal and spatial resolutions (underestimation of infarct size) and SNR. Applying parallel imaging (IPAT) and/or radial sampling of K-space might possibly off-set the limited temporal and spatial resolution ⁹.

Switching from multiple breath-holds to free-breathing has facilitated possibilities for higher spatial resolution imaging at the expense of prolonged 3D whole heart acquisition times. Previous chapters have addressed the impact of motion artefacts because of cardiac pumping activities. This chapter will explore the intricacies of respiratory motion and its impact on free-breathing 3D imaging.

Respiratory motion has been and remains one of the greatest challenges for overall clinical adoption of CMR. One key factor impacting cardiac motion is the effect respiration has on the position and shape of the heart. The lungs, as the diaphragm is being pulled downwards during inspiration, fill up with air. The heart attached to this musculature, at rate of 10-12 cycles per minute, deforms and shifts several millimeters (mm) in head-feet (H-F) direction due to lung hysteresis (volume differences between inspiration and expiration) (Figure 26). Other than cardiac motion, these effects are more unpredictable and not only vary from patient to patient, but also occur at different time intervals ²⁷.

As previously discussed in this document, controlling cardiac motion is done by QRS-complex detection and triggering the acquisition at a particular delay. Image quality optimization is possible by “suspending” respiratory motion artefacts such as slice-misregistration, blurring, ghosting and loss of signal by means of the following respiratory gating and motion correction techniques:

• Breath-holding

Breath-holding:

Limited to healthy or compliant individuals, this is the simplest form to indirectly measure and suppress motion of the heart. Acquisition is performed with BH at end-expiration (ideally between 10-20 sec) as this tends to be more reproducible. Several studies however have demonstrated that even during this time, the left margin of the base together with the diaphragm undergo several mm shifts and the heart rate increases at end of the BH. Therefore, the temporal acquisition and spatial resolution are limited. Patients may struggle to consistently hold the same diaphragm position during successive breath-holds which leads to variations of up to 4-13 mm between the position of the heart and imaging plane, ultimately causing slice mis-registration and poor image quality as a result ²⁸.

Free breathing:

A: Respiratory gating

To facilitate patient cooperation, free-breathing imaging with the use of external devices such as abdominal or chest belts (pear belts), monitor the change in pressure caused by movements of the torso, which in turn generates a waveform depicting the breathing cycle and indirectly estimate the respiratory movement of the heart. Image acquisition is set in such a way that data from each cardiac cycle is either rejected or accepted whether registered above or below a gating threshold (Figure 27) ²⁹.

This method offers slightly better resolution and is as effective as a “regular” breathing pattern of the individual and yet still unable to completely remove respiratory motions and blurring and prone to drastic scan increase.

B: Conventional diaphragm 1D navigation (current standard on our scanners)

Accurate measurement in motion correction is essential to facilitate higher spatial resolution imaging. Done with navigator echoes (NE), either as a 2D selective RF “pencil beam” Figure 28a), or as a spin echo with 90° excitation and 180° refocusing pulse “cross-pair” (Figure 28b); this method consists of a narrow pulse positioned at the dome of right hemi-diaphragm, with readout in head-feet (H-F) direction using a 1D representation of the lung-liver interface (Figure 28c and Figure 29) ³⁰.

The result is less blurring performed either retrospectively or prospectively (real-time).

The respiratory efficiency is determined by the amount of ECG triggers that fall within a predetermined navigator acceptance window (2.5-5.0mm) at end-expiration, to either accept or reject images.

Accurate preparation, planning and execution of the conventional 3D DGE requires the technician to do several interactions throughout, highlighted schematically and presented in Figure 31.

• Electrode placement.

• Setting TD and AW depending on heart rate and rhythm.

• Choose the inversion time.

• Select gating/acceptance window.

• Position navigator.

• Plan 3D volume on heart.

High quality and sharper imaging usually require narrow acceptance windows. However, with more data being rejected the result is lower scan efficiencies and prolonged scan times and is specifically challenging in patients with arrhythmias or irregular breathing. Larger gating windows on the other hand, allows for more data to be accepted and acquired, hence a higher scan efficiency and shorter time but at the expense of increased motion, blurring or misregistration ¹³.

The main issue with the conventional approach is that true heart displacement is not measured accurately and since H-F motion of the diaphragm is greater than the H-F motion of the heart, a correction factor of 0.6 is used to compensate for this. This oversimplified correction firstly omits the fact that motion between the diaphragm and heart varies from patient to patient and secondly, the non-rigidity and complex motions of the heart are not accounted for. Continuous temporal changes in patient inter-dependent breathing patterns, respiratory drifts, or in situations of falling asleep or if too anxious at the early part of the exam (compared to later as they become more relaxed), may result in limited respiratory efficiency (between 30-40%), ghosting and extended scan duration ³⁰.

Developments over the last 10 years have steadily been moving away from the traditional diaphragmatic 1D navigator techniques. Correcting for motion by focusing directly on the whole heart, has led to the following innovations:

C: Self-navigation respiratory

This method with its application for assessing complex congenital cardiac malformations uses the position of heart itself for monitoring. It allows acquisition in more than one dimension besides H-F alone, also in right-left (R-L) or (H-F with anterior-posterior, A-P) (Figure 32) ³².

In response to the conventional 1D diaphragm navigator technique, this 3D radial acquisition can correct respiratory motion by collecting data in the Superior-Inferior K-Space profiles. Ease of use is enhanced by limiting the number of steps to only placing electrodes, adjusting trigger delay and 3D volume positioning ³².

Extra scout scans and pre-localizers are simplified and in combination with the key advantage that navigator placement, no longer is needed, data rejection is therefore avoided. As a result, improved image quality because of higher spatial resolution, increased respiratory efficiency and shorter and predictable scan time are possible, even in cases of irregular breathing.

A major limitation, however, is that by including signal from both static and moving tissues this method is unable to distinguish information between rigid organs such as the chest wall or vertebrae and non-rigid organs (heart) causing interference within the FOV ³⁰.

D: Image-based navigation 3D-DGE (I-NAV)

Previously restricted for coronary imaging, this method has now been adapted for 3D-DGE. Other than self-navigation, I-NAV focuses directly on the entire 3D complex motion of the heart including rotational and translational movements (Figure 10). In combination with navigator algorithms this method improves robustness to motion by applying rigid and non-rigid compensation (ability to spatially isolate the moving heart from static tissues) in H-F and R-L direction ^{13, 34}. The results are short and predictable scan times, 100% respiratory efficiency, higher isotropic spatial resolution, low operator dependency since navigator positioning is not required and accurate delineation of scar tissue ³⁰.

Below (Figure 34) is an example of a study conducted in 2015 at the Guy’s & St. Thomas’ Hospital, King’s College London and published in 2017 in the Journal of Cardiovascular Magnetic Resonance: in which for the first time an INAV-3D DGE method has been implemented in clinical setting. Upper rows (a-b-c-d) in Figure 34 from 2D-DGE compared with matching images scanned with I-NAV-3D lower rows (e-f-g-h) ³⁵.

The main challenge of these and other novel respiratory correction methods are widespread adoption and integration into day-to-day clinical practices. Besides shortening scan times and facilitating images almost free of respiratory artefacts, they need to be easier enough to perform, be operator in-dependent and not require too complex trainings to fulfil or lengthy and expensive computational postprocessing ³⁵.

Unfortunately, these advanced methods have so far been limited to a handful of research and dedicated centers and mainly focused on healthy individuals ¹³.

Currently, large, and extensive assessment of clinical merits are being undertaken with the scoop on integration into daily clinical workflows which soon can be expected.

3. Pulse Sequence

Early alternatives for scar assessment, relied on non-gated T1-weighted spin echo and fast (turbo) spin echo sequences resulting in images with poor spatial resolution and motion artefacts due to long breath-holds and very limited CNR.

The introduction of inversion recovery (IR) T1-weighted fast gradient recalled echo (GRE) (Siemens: (t)-FLASH), led to significant improvements related to CNR, contrast-enhancement-ratio (CER), higher spatial resolution and shorter scan times ¹⁸. The basis of DGE imaging nowadays still relies on this principle with alternative options involving PSIR and SSFP (Siemens: FISP), to more state-of-the-art applications such as 3D-whole heart pulse acquisition for instance in resolving left atrial scarring.

A key element in the assessment of infarction and scarring as previously discussed in Chapter 1, is proper selection (nulling) of the delay time by means of an IR-SSFP TI-scout. Different variables can potentially make this process very challenging and possibly resulting in either too short or too long selected TI times (Figure 4) ⁴. If the TI is too short, normal myocardium with a negative magnetization (-Mo) is depicted with an increased image intensity whilst scarred tissue intensity is decreased (shades of grey) (Figure 3), possibly leading to overestimation of infarcted region ³⁶. Overshooting the TI value leads to both normal myocardium and scarred tissue with high signal intensity thus limiting overall contrast ³.

Even if the optimal TI time is identified, this value might still not match the true due to the discrepancy between TI-scout which is a SSFP readout and DGE which is a spoiled-GRE readout ⁹.

Optimal nulling in 3D IR-GRE segmented DGE sequence presents yet another challenge since contrast continues to wash-out during the long acquisition, therefore changing the optimal TI and resulting in poor CNR and sub-optimal image quality (Figure 35).

Phase-sensitive inversion recovery (PSIR) technique has been proposed to address these problems. By applying a second reference background image (proton density), PSIR can differentiate between positive and negative magnetization producing corrected “real” images whereby -Mz appears darkest, nulled tissue mid-grey and +Mz bright (Figure 36, phase-corrected column) ²⁴.

By correcting the polarity of the signal intensity, PSIR sequence (Figure 37):

• facilitates a consistent contrast difference over a longer range of TI times.

• avoids the need for scan interruption to re-assess optimal TI times.

• reduces inter-operator dependency, greatly improving image quality and thus an ideal option for 3D-FB whole heart sequences combined with faster acceleration techniques.

• is suitable for patient with arrhythmias or with difficult breathing.

Another area where PSIR might play a key role is in the visualization of scars adjacent to blood-pool at the subendocardial border. DGE usually achieves great contrast differentiation between normal and scarred myocardium. The challenge at times is the contrast differences between myocardium and blood-pool. Some articles have therefore suggested a Dark-blood PSIR DGE technique involving the improvement of scar-to-blood contrast by choosing the TI time when blood-pool is dark instead of healthy myocardium (Figure 38) ^{39, 40}.

By selecting this time as default, images would still have myocardial nulling but with reduced signal from the blood-pool resulting in improved identification and delineation of size and location of subendocardial scar tissue.

The prerequisite in both tissue characterization (static imaging) and function (cine imaging), is that the repetition times (TR) be kept very short. This is only possible with the use of gradient echo-based pulse sequences. The most common pulse sequences, -as part of these 2 bright-blood imaging techniques, are acquired through spoiled gradient echoes (SPGR) and bSSFP. As the TR is kept short, transverse magnetization signal from a previous RF pulse still exist when next RF is applied. This signal can either contribute or interfere with the subsequent signal during following TR's.

Spoiled or gradient recoiled echo (GRE) sequences use a spoiler to remove this signal resulting in very short TR and TE and therefore produce contrast which is predominately T1-weighted ⁹.

Balanced SSFP gradient sequences on the other hand do not spoil this signal but re-phase it to therefore contribute to the transverse magnetization at the next RF. Steady state condition is reached after a few RF's when signal from previous excitations is combined to give an even greater signal thus producing a contrast which is related to the T2/T1 ratio of tissues.

Blood with a high T2:T1 ratio (large transverse magnetization) appears bright (hyperintense) while myocardium with a lower T2:T1 ratio (small transverse magnetization) appears dark (hypointense).

Compared to spoiled gradient echo, bSSFP produces higher SNR which in turn allows for higher bandwidth resulting in an even shorter TR and TE and thus faster with better temporal and spatial resolution ⁹.

Thus, makes it the preferred sequence for cine imaging. BSSFP however, is more sensitive to field inhomogeneity causing dark-banding (off-resonance) and with a higher flip angle leads to increased SAR values and thus less suitable for higher field strength scanners >3T ⁹.

Contrast administrated during DGE, reduces the T1 relaxation time and in combination with IR- preparation, makes T1-weighted (turbo) spoiled gradient the most common technique of choice in demonstrating signal enhancements, between normal and affected myocardium. This ultimately leads to turbo-FLASH sequences producing higher CNR and CER compared to bSSFP. Recently, the combination of the “faster” bSSFP with single-shot IR or PSIR, have become popular alternatives to acquire DGE imaging in patients with arrhythmias and or with breath-hold limitations ⁹. Table 2 presents an overview of different DGE possibilities.

4. How to Do It: High Resolution 3D Whole Heart DGE [1,10]

The sequence normally used is a 3D HR-DGE inversion recovery spoiled (turbo) gradient (t-FLASH or GRE) planned strictly transversal and positioned on free-breathing localizers (Figure 40). Since this is a long acquisition; extra attention, preparation and instruction to the patient is critical. Below are some key tips to consider:

• Patient positioning: comfortable pillow, pleasant room temperature, relaxing music and/ or warm blanket to lie still during exam.

• Patient information: fully cooperative and well-informed patient is likely to be more motivated in following instructions thus leading to successful image quality.

• ECG: bad leads and or connections result into poor images.

• Instruct patient to maintain a shallow and regular breathing pattern for increased navigator efficiency and acceptable scan time (Figure 40). Heavy and or irregular breathing leads to severe respiratory drifts and longer scan duration, potentially affecting image quality.

• It is advisable to run a second free-breathing localizer prior to the 3D sequence. The patient has had some time to relax in the scanner, breathing is then naturally and more consistent, resulting in an accurate navigator positioning on top of the liver dome.

• Plan volume (yellow box) in straight axial position covering the ventricles on localizers.

• Locate top liver dome on Coronal and Axial localizers and plan middle (red block) of cross-pair navigator on it.

• Saturation bands are positioned over arms and chest to reduce (R-L) fold-over-, ghosting, and residual breathing artefacts, and subcutaneous fat signal.

• Select zero shimming for better homogeneity (green box).

Tracking respiratory motion as previously discussed in Chapter 2.7 is dynamically done by a crossed-pair navigator positioned on the liver dome at the liver-lung interface; making sure it does not cross any other anatomy of interest. Several parameters can be adjusted as seen in Figure 41.

• Either run the 3D sequence with scout mode (19 sec.) (A1) on, or (start-stop-start) method without scout mode on. Tracking factor default (0,6) is a correction between displacement of diaphragm and heart.

• Navigator data displayed in (B2) shows an upper line through the lung and a lower line through the liver. Expiratory phase is automatically detected (Mode 156).

• Green box represents acceptance window of (3-5 mm) in which data is accepted.

• Scan efficiency window next to “Accept” shows 40%. Minimum of 30% is required for acceptable scan times.

• Once the navigator adjusts (C3), enter new value (157) manually into Search Position (red) to center both green and red boxes on this position.

• Scout Mode can be deselected to run the real sequence.

• During the actual scan, the navigator can be monitored in (D4) and remaining time in percentage to the right of Ima. Selecting motion adaptation allows the green box to dynamically adapt over time, thus compensating for respiratory drifts.

In situations of bradycardia or bradyarrhythmia’s whereby the navigator efficiency is compromised, scan duration will negatively extend beyond the 1-5 minutes window prior to the TI scout. It is advisable to run this sequence after the 2D DGE planes, requiring a new TI value (from second TI-scout) to be selected and incremented with 30-50msec. Accurate visualization of enhancement differences within the myocardium is optimized when this “moving target” is captured correctly as seen in Figure 42.

5. Conclusion

For decades now, DGE has been the cornerstone in CMR. Based on the principle of hyperenhancement post contrast, this tool has been indispensable for assessing necrotic and fibrotic myocardial tissue as well as several non-ischemic cardiomyopathies. A critical aspect of this diagnosis is the ability to “capture” these enhancement differences as acquisition times progress as seen in Figure 3 and Figure 42. The primary technique used for DGE is based on a segmented 2D IR fast spoiled gradient echo (GRE) or a balanced steady state free precession (BSSFP) sequence, which is acquired over multiple breath-holds. The limitations of the 2D standard paradigm include slice misregistration due to poor or incomplete breath-holds, suboptimal heart coverage, respiratory motion artifacts and poor spatial resolution ⁹.

As the number of exams and demand for specific diagnosis such as size, shape and location of the infarcted zones, or early detection of fibrosis have increased, so have the demands for a paradigm shift with higher image quality. With high isotropic voxels and whole heart continuous coverage, 3D sequences have been proposed to address the limitations of the standard paradigm ¹. Key differences are presented in Figure 43.

The “early adoption” of (1-4 BH) 3D-DGE sequence, limited by (too) long breath-holds was not very suitable for the most ill and vulnerable patients. Free-breathing 3D-DGE alternatives introduced by Saranathan et al. in 2004, incorporate a 1D navigator-echo followed by a gradient echo acquisition ⁴¹. Thus, facilitates higher spatial resolution imaging making it possible to identify small scars which are otherwise non-visible with 2D DGE. These and other benefits of the 3D-FB DGE, require specific attention to the physiology and artefacts resulting from the heart activity, technical aspects of the sequence, and patient cooperation.

The current 3D DGE approach with conventional 1D diaphragm navigator echo is characterized by:

• Unpredictable and long scan times due to limited scan efficiencies.

• Indirect measurement of heart motion.

• Hysteresis effects.

• Complex acquisition planning.

• Patient inter-dependency

The heart has complex motions and is influenced by varying breathing patterns; placing a huge demand on cardiac MRI to identify, locate and tailor its acquisition to the period of minimum cardiac activity. This is achieved by synchronizing the MR data to the heart’s electrical activity, as well as interpreting different rhythm patterns such as sinus bradycardia to more complex events like tachyarrhythmias.

A successful implementation of tissue characterization in delayed gadolinium enhancement, requires knowledge of the interconnection of parameters such as trigger delay, acquisition window and trigger window and how they relate to varying heart rates and rhythms.

In situations of bradycardia where acquisition times or breath-holds are bound to be prolonged, reducing the temporal resolution or triggering on every RR cuts down the acquisition by 50%. In conditions of tachycardia, where the diastolic phase decreases potentially leading to blurring or slice misregistration, synchronizing the acquisition window to end-systole could be an alternative ⁴².

One of the greatest challenges in shifting the paradigm from 2D multi BH’s to free-breathing 3D imaging remains the effects of respiratory motion on the shape and position of the heart. Other than cardiac activities, “breathing (hysteresis) effects” are difficult to estimate and locate and are very patient inter-dependent ¹. Limited by the amount of heartbeat acceptance (navigator efficiency < 40%) together with inaccurate measurements of true heart displacement, makes the conventional 1D diaphragm navigator prone to ghosting and extended scan times. Shallow breathing in “healthy” and cooperative patients with normal sinus and RR around 1000 msec., combined with a narrower acceptance window will produce the best image quality and is easily performed in the 1-5 minutes slot post contrast. In patients with more complex conditions potentially causing extended duration of the 3D DGE, and to not jeopardize the continuity of the protocol, it is advisable to perform this sequence with a wider acceptance window after the 2D DGE planes. During stress exams it is perhaps recommended to run this scan at the end when the effects of Regadenoson have worn off and the heart rate and rhythm have returned to normal. Maintaining proper contrast differences, as previously discussed will require “following” the TI time obtained from a new TI scout sequence. Another key and no less important element are planning and patient support in which high-quality scans depend on comfortable, well informed and engaged patients.

There are however extreme situations such as (brady- and tachy-) arrhythmias or worsening irregular breathing patterns, whereby the diaphragmatic 1D navigator with its limited efficiency, is either unable to produce diagnostic images, lasts too long or where the acquisition even fails to perform ⁹. Ongoing and promising developments in respiratory compensation methods, with the focus on direct position of the heart itself, are soon expected to be incorporated into daily clinical routines. These techniques not only have substituted the current navigator echo paradigm by a more advanced self and image-based navigation algorithm approach, but will drastically alter the existing scan protocol (Figure 44) allowing for:

• Shorter and predictable scan times.

• Increased efficiency up to 100%.

• Simpler scout scanning.

• No navigator planning required.

• Lower operator-dependency (ease of use) ³¹.

While the implementation of the 3D whole heart DGE remains challenging depending on the clinical and technical indications and conditions of the patient, it is the intention of this document to provide an in-depth understanding of the principles of this novel sequence, and hopefully facilitate knowledge and serve as an indispensable tool for patient care.

Abbreviations

AAR: Area at risk

AR: Arrhythmia rejection

ARVD-(C): Arrhythmogenic right ventricular dysplasia/ cardiomyopathy

AV-valves: Atrioventricular valves

AW: Acquisition window

BH: Breath-hold

BPM: Beats per minute

BSSFP: Balanced steady state free precession

CHD: Congenital heart disease

DGE: Delayed gadolinium enhancement

ECG: Electrocardiogram

ECV: Extra cellular volume

EGE: Early gadolinium enhancement

FB: Free breathing

FOV: Field of view

FPP: First-pass perfusion

GRE: Gradient recoiled echo

HR: Heartrate

IR: Inversion recovery

LGE: Late gadolinium enhancement

MI: Myocardial infarction

MVO: Microvascular obstruction

NE: Navigator echo

NICM: Non-ischemic cardiomyopathies

nPE: Number of Phase encoding steps

PCI: Percutaneous coronary intervention

PSIR: Phase sensitive inversion recovery

PVC: Premature ventricular contraction

RF: Radio-frequency pulse

RR: RR-interval representing 1 cardiac cycle/1 heartbeat

RVOT: Right ventricle outflow tract

SA: Short axis

SNR: Signal to noise ratio

SPGR: Spoiled gradient recoiled echo

STIR: Short tau inversion recovery

TD: Trigger delay

t-FLASH: Turbo-Fast low angle shot

TR: Repetition time

Tres: Temporal resolution

True-FISP: True-fast imaging with steady state precession

TW: Trigger window

VPS: Views-per-segment

1D: 1 Dimensional

2D: 2 Dimensional

3D: 3 Dimensional

References