Cardiovascular magnetic resonance physics for clinicians: Part II

Abstract

This is the second of two reviews that is intended to cover the essential aspects of cardiovascular magnetic resonance (CMR) physics in a way that is understandable and relevant to clinicians using CMR in their daily practice. Starting with the basic pulse sequences and contrast mechanisms described in part I, it briefly discusses further approaches to accelerate image acquisition. It then continues by showing in detail how the contrast behaviour of black blood fast spin echo and bright blood cine gradient echo techniques can be modified by adding rf preparation pulses to derive a number of more specialised pulse sequences. The simplest examples described include T2-weighted oedema imaging, fat suppression and myocardial tagging cine pulse sequences. Two further important derivatives of the gradient echo pulse sequence, obtained by adding preparation pulses, are used in combination with the administration of a gadolinium-based contrast agent for myocardial perfusion imaging and the assessment of myocardial tissue viability using a late gadolinium enhancement (LGE) technique. These two imaging techniques are discussed in more detail, outlining the basic principles of each pulse sequence, the practical steps required to achieve the best results in a clinical setting and, in the case of perfusion, explaining some of the factors that influence current approaches to perfusion image analysis. The key principles of contrast-enhanced magnetic resonance angiography (CE-MRA) are also explained in detail, especially focusing on timing of the acquisition following contrast agent bolus administration, and current approaches to achieving time resolved MRA. Alternative MRA techniques that do not require the use of an endogenous contrast agent are summarised, and the specialised pulse sequence used to image the coronary arteries, using respiratory navigator gating, is described in detail. The article concludes by explaining the principle behind phase contrast imaging techniques which create images that represent the phase of the MR signal rather than the magnitude. It is shown how this principle can be used to generate velocity maps by designing gradient waveforms that give rise to a relative phase change that is proportional to velocity. Choice of velocity encoding range and key pitfalls in the use of this technique are discussed.

Figure 1

Parallel imaging. Parallel imaging uses the spatial distribution of coil array elements and their characteristic sensitivity maps to provide spatial information. This allows under-sampling of k-space (skipping of phase encoding steps) during the acquisition, shortening the image acquisition time. (a) shows a full k-space acquisition without parallel imaging. Image signal intensities from anterior and posterior coil elements are combined in phased-array mode to provide uniform signal intensity across the field of view. (b) Parallel imaging using the SENSE method with a reduction factor of 2 (2x SENSE) leads to skipping of alternate lines of k-space. Without the SENSE reconstruction this would lead to an effective reduction of the field of view and image aliasing or ‘foldover’. As an intermediate reconstruction step, separate aliased images are obtained for each coil element. In (c) the SENSE reconstruction uses the images obtained separately from the posterior and anterior coil elements, together with the coil element sensitivity maps acquired from a reference scan to remove the aliasing and reconstruct an unfolded the image with a full field of view.

Figure 2

The three most commonly used magnetisation preparation pulses, showing their relation to the image data acquisition pulse sequence. Beneath each pulse sequence, curves show the behaviour of the z-magnetisation for three different tissues (fat, muscle and fluid). Spoiler gradients, S are applied to suppress any unwanted transverse magnetisation produced by the preparation pulses. In (a) following the 90° saturation pulse the z-magnetisation for all tissues is reduced to zero and recovers according to their T1 relaxation rate. When the imaging pulse sequence is applied after a delay TS the resultant contrast is T1-weighted with the shortest T1 (fat) yielding the highest signal intensity. In (b) following the 180° inversion pulse the z-magnetisation for all tissues is inverted, and then recovers from a negative value. When the imaging pulse sequence is applied after a delay TI, the tissue for which its z-magnetisation is passing through zero yields no signal, effectively suppressing the signal contribution from that tissue. In this example, the signal from muscle is suppressed. In (c) The frequency-selective fat suppression pulse is applied only at the resonant (Larmor) frequency of fat so that only the z-magnetisation of fat-based tissue is reduced. In this case the imaging pulse sequence is applied with no delay. The z-magnetisation of other tissues (muscle and fluid) are unaffected by the fat suppression pulse and will yield normal signal levels.

Figure 3

STIR, FLAIR and Black-Blood Preparation schemes. Pulse sequence timing and z-magnetisation curves for the three common special cases of the inversion recovery preparation scheme. In (a) the short TI inversion recovery (STIR) pulse sequence has the TI chosen to correspond to the null point for fat tissue (TI_fat) resulting in the suppression of signal from fat. Note that for a modulus reconstruction, the sign of the magnetisation is ignored and the displayed signal intensity from the other tissues increases with increasing T1 value, with fluid yielding the highest signal intensity. In (b) the fluid attenuated inversion recovery (FLAIR) pulse sequence has the TI chosen to be at the null point for fluid to suppress the signal from fluid (TI_fluid). In (c) the first inversion pulse of the black-blood preparation scheme has its TI chosen to correspond to the null point for blood (TI_blood). This pulse is non-slice-selective and so causes the inversion of magnetisation for all tissues. The second slice-selective inversion pulse is needed to undo the inversion applied to the tissues within the imaged slice (myocardial muscle, fat), otherwise their signals would also be suppressed. The combined effect of the two inversion pulses is therefore to invert the tissues including blood outside the imaged slice. The subsequent wash-in of inverted blood into the image slice during the TI period results in suppression of signal from the blood pool.

Figure 4

Triple IR Preparation scheme for Oedema imaging. Triple inversion recovery TSE (Black-Blood turboSTIR) pulse sequence commonly used for oedema imaging. In (a) the pulse sequence timing is shown together with z-magnetisation curves. The preparation scheme consists of the two 180° pulses for the black blood preparation, followed by a third slice-selective 180° pulse to provide the STIR contrast. The sequence has two inversion times, TI_blood and TI_fat. TI_fat has the same value as that used for fat suppression as in the STIR sequence (approx 160 milliseconds). Calculation of the value for TI_blood depends on the heart rate, and the effective TR (number of heart beats per repetition). The z-magnetisation curve shown for blood assumes it washes into the slice after the third 180° pulse has been applied. The image in (b) is acquired in the short axis plane using the black-blood turbo STIR technique. The intrinsic fluid weighting of the STIR technique has been further enhanced by selecting a long TE value (100 ms) to introduce T2-weighting (the technique is sometimes referred to as T2-weighted STIR). Myocardial oedema can be seen as an area of increased signal (arrow). (Image courtesy of Darach O h-lci and Daniel Messroghli, Deutsches Herzzentrum, Berlin).

Figure 5

Frequency selective fat suppression. MR signal contributions from lipid and water molecules are plotted in (a) as a function of Larmor frequency. Both the lipid and water populations exhibit a small range of Larmor frequencies due to the existence of spin-spin interactions, however the central frequency of the lipid population is shifted to a lower frequency than that of water due to the increased shielding effect of the electron cloud surrounding the larger lipid molecule. To achieve fat suppression, a frequency-selective rf pulse (Chemical Shift Selective or CHESS pulse) is applied at the Larmor frequency of the lipid population, causing saturation of the lipid magnetisation. The image data acquisition pulse sequence is then applied immediately after the fat suppression pulse (Figure 2c). In (b) two axial slices acquired using a black blood TSE pulse sequence are shown without (top row) and with (bottom row) fat suppression.

Figure 6

(a) A schematic diagram of a cine tagging pulse sequence. In this example a composite binomial rf pulse is used consisting of three rf pulses with amplitudes in the ratio of 1:2:1. Two modulating gradients are applied in the spaces between the rf pulses to de-phase (modulate) the transverse magnetisation between each rf pulse. The net effect is to cause a variation (or modulation) of the z-magnetisation, creating a series of parallel lines of tissue with a magnetisation that varies alternately between its equilibrium value (untagged) and zero (tagged). A spoiler gradient, S, is applied to destroy transverse magnetisation generated by the tagging pulses. T1 relaxation causes the magnetisation of the tagged lines to recover towards equilibrium, while at the same time the magnetisation of the untagged tissue becomes partially saturated by the rf pulses applied as part of the cine imaging sequence. This causes the contrast between the tagged and untagged lines to reduce as the cardiac cycle progresses. The two short-axis images in (b) are acquired from separate tagged cine acquisitions with tagging applied in two different directions. The arrows indicate the direction of the modulating gradient in each case. Both image examples correspond to a cardiac phase at around end-systole. For stationary tissue, such as in the chest wall, the tagged pattern has remained fixed and is seen as a series of parallel lines. Within the left ventricle, the line pattern has deformed as it follows the motion of the myocardial muscle.

Figure 7

This diagram shows how a SPAMM pulse can produce a magnetisation pattern consisting of a series of parallel lines that are alternately fully magnetised and fully saturated, appearing as bright and dark lines (shown on the right). In this example, the composite rf pulse is a 1-2-1 binomial pulse, consisting of three rf pulses with relative amplitudes in the ratio of 1:2:1. As the effective flip angle of the composite pulse is set to be 90°, the flip angles of the three individual rf pulses is therefore 22.5°, 45° and 22.5° respectively. Starting at equilibrium (a), the first rf pulse causes all spins to flip through 22.5° (b). The first modulating gradient is then applied causing the spins to move out of phase along the gradient until there is a 180° phase shift between points that correspond to the desired spacing between adjacent bright and dark lines. The second 45° rf pulse is then applied, causing the magnetisation that is 180° out of phase to be flipped from −22.5° to +22.5°, while magnetisation that is in phase is flipped from +22.5° to 67.5° (d). The second modulating gradient introduces a further 180° phase shift between adjacent bright and dark tag lines (e). The third 22.5° rf pulse is then applied. This causes the magnetisation that is 180° out of phase to be flipped from −22.5° to 0 (aligned along the z-axis), while magnetisation that is in phase is flipped from 67.5° to 90° to become saturated (f).

Figure 8

The relationship between signal intensity and concentration. Signal intensity values over a range of concentrations for a spoiled gradient echo pulse sequence. The two dashed curves show the separate dependencies of the signal behaviour for T1 or T2 alone. The solid line shows the combined effect of T1 or T2 on signal intensity. At low concentrations the effect of T1 shortening is dominant, while at higher concentrations the T2 shortening effect becomes the dominant factor. A series of samples imaged with increasing percentage concentrations of Gadolinium are shown underneath the plot as a visual demonstration of the effect.

Figure 9

Dynamic contrast enhanced cardiac perfusion imaging. Contrast agent is injected intravenously whilst multiple images of the heart are acquired to create a movie showing the contrast agent passing through the heart. Contrast agent can be seen as signal enhancement in the right ventricular cavity (RV) followed by the left ventricular cavity (LV) and more gradually in the myocardium, before finally washing out.

Figure 10

Pulse sequence diagram for hybrid Echo planar imaging (EPI). In echo planar imaging (EPI) multiple lines of k-space are rapidly acquired following a single rf excitation pulse. The slope of the frequency encoding gradient is rapidly alternated, generating a train of gradient echoes. A phase encoding gradient ‘blip’ is applied between each frequency encoding gradient to ensure each gradient echo fills a different line of k-space. Single-shot EPI acquires all the lines of k-space following a single rf pulse. More commonly in cardiac MR applications, a hybrid or segmented EPI approach is used where multiple rf pulses are applied each followed by a shorter echo train. In this example, an echo train length (ETL) = 5 is used.

Figure 11

Choice of imaging pulse sequence for myocardial perfusion MRI. Diagram illustrating the choices of data acquisition pulse sequence for myocardial perfusion MRI. A saturation recovery preparation pulse is employed to ensure T1-weighting followed after a delay by the image data acquisition pulse sequence. The most common data acquisition pulse sequences used are fast/turbo spoiled gradient echo (FGE), balanced steady state free precession (bSSFP) or hybrid (or segmented) EPI. The number of slices that can be acquired (in this case three) is limited by the R-R interval, the saturation delay and the length of data acquisition period.

Figure 12

Trigger Delay and Saturation Time. The trigger delay (TD) sets the point within the cardiac cycle that the centre of k-space, k₀, is acquired within each RR-interval . The saturation time (TS) determines the time between the saturation pulse and the centre of k-space, thereby controlling the T1-weighted contrast of the image for a particular image slice.

Figure 13

Choices for myocardial perfusion imaging – prep pulse scheme and number of slices. With a single slice acquisition per RR interval (top) there is flexible choice for the optimal cardiac phase and T1-weighted image contrast, but poor coverage of the LV. For multiple slice acquisitions, the use of a separate preparation pulse for each slice (centre) allows the same image contrast for each slice (fixed TS) but the two slices are acquired at different cardiac phases due to their different trigger delays and the number of slices is limited (two in this case). Using a pre-pulse shared by all the slice acquisitions (bottom) potentially allows more slices to be acquired (three slices in this case), but leads to each slice having both a different T1-contrast behaviour, as each slice has a different (TS), and a different trigger delay.

Figure 14

Myocardial Perfusion defect. Under stress conditions an area of reduced signal intensity is observed in the septal and anterior walls of the left ventricle, consistent with disease in the coronary arteries supplying these regions. Image courtesy of John Greenwood [43]. (See Additional file 1 for perfusion movie).

Figure 15

Quantitative Perfusion Data. For every image in the dynamic sequence region-of interest contours describing the myocardium and a region in the blood pool are drawn. The mean signal intensity from within each region is plotted for each time point to generate plots of signal intensity versus time to show the increase in signal intensity in both the myocardium (green) and the blood pool (red). The blood pool curve is also often referred to as the arterial input function (AIF). These curves can be analysed together to give an estimate of myocardial blood flow (MBF).

Figure 16

Non-linearity effects cause errors in MBF. The left hand graph shows the difference between the assumed linear relationship between signal intensity and Gd concentration (dotted line) and the true relationship (solid line). The right hand graph shows how the non-linear relationship at higher concentrations can propagate into a peak height error in the measured blood pool curve (the arterial input function or AIF) causing an overestimate in MBF.

Figure 17

Schematic illustration of an extracellular Gd- based contrast agent kinetics within the myocardial tissue. Following intravenous administration, the extracellular Gd-based contrast agent enters the microvascular network of the myocardium via arterial inflow (left). It is extravasated into the interstitial fluid (extravascular extracellular space) and gradually washed out back into the venous outflow (right) and removed from the body via glomerular filtration. Areas with an increased volume of interstitial space will present a larger distribution volume for the incoming contrast agent, but importantly pathological tissues such as myocardial scar, may also display slower extravasation rates as well as delayed re-absorption (wash out) of contrast agent into the vascular space.

Figure 18

Time-course of T1 for EGE and LGE. This shows the time-course of the changes in longitudinal relaxation time T1 in LV and myocardium following administration of 0.2 mmol/kg Gd-DTPA. T1 values are adopted from the data measured in a study by Klein et al. [75]. In EGE imaging (t_a ~ 5 min post-injection), all tissues apart from MVO experience a significant T1 shortening (second panel from the left). In areas occupied by MVO, a very modest amount of contrast agent is present at 5 min post-injection, and the T1 value within the MVO is high compared to the other three compartments at this time point. The rate of recovery towards the baseline (pre-contrast) T1 value reflects the washout of contrast from individual compartments. Whilst normal myocardium and LV blood T1 values continue to rise between 5-15 min post-injection, scar tissue still maintains low T1 values, due to delayed extravasation and accumulation of contrast agent within an enlarged interstitial water compartment. The low values of T1 may further be maintained by the slow washout kinetics. In MVO, T1 values may continue to decrease, as the areas occupied by MVO may receive contrast agent via passive diffusion from the neighbouring scar.

Figure 19

Pulse sequence diagram for IR-GRE. Pulse sequence diagram representing two consecutive segments of an IR-FGE sequence used in early and late Gadolinium enhancement imaging. In a typical breath hold acquisition of a single DGE slice, 8–10 segments are acquired to fill ~192 -240 lines of k-space. The changes in M_z/M₀ are illustrated in the middle panel, with green and red lines representing M_z/M₀ in short and long T1 regions. The bottom panel illustrates the changes in signal intensity in the modulus MR images (SI ~ |M_z/M₀|). For each shot, several lines of k-space are filled within an image data acquisition window typically limited to 150-200 ms, with a trigger delay corresponding to mid-diastole to minimise the effects of cardiac motion.

Figure 20

Inversion recovery curves at 5-15mins post-injection. Inversion recovery curves at 5, 10 and 15 minutes post-injection and relative signal levels obtained by choosing TI values to suppress signal form MVO (purple curves) at 5 minutes post injection and normal myocardium signal (red curves) at 10 and 15 minutes post-injection. Curves are shown of both the z-magnetisation, M_z expressed as a fraction of the magnetisation at equilibrium, M_o (M_z/M_o) and the modulus of this value, (|M_z/M_o|), representative of the actual displayed signal intensity. In EGE, the difference between T1 in the MVO compartment and surrounding scar and normal myocardium is emphasized by using a long TI which minimises the signal from the MVO. A TI of 440 ms will null the signal from MVO (T1 = 640 ms at 5 minutes post-injection). With this choice of TI, surrounding scar and normal myocardium appear bright, thus enabling identification of MVO (left panel). At 10 and 15 minutes post-injection, the differences in T1 between normal myocardium and scar begin to emerge due to delayed accumulation of contrast in the enlarged interstitial space of the scar. To maximise contrast between scar and normal myocardium, a TI of 300 ms is chosen to null the signal from normal myocardium at 10 min LGE images (middle panel). As T1 values continue to change between 10 and 15 minutes post injection, the optimal nulling time rises to 320 ms in at 15 minutes post-injection for the example presented above (right panel).

Figure 21

EGE images. a) EGE image acquired in the 4-chamber orientation demonstrating a large area of microvascular obstruction in the lateral wall associated with transmural infarction that extends from base almost to the apex (arrowheads). b) Short axis LGE image depicts a transmural scar (arrowheads) that envelops an area of MVO in the lateral wall. Images courtesy of Dr J. Greenwood, MCRC, University of Leeds.

Figure 22

Look-Locker TI scout train. This figure shows the relative signal intensities for the LV blood pool, normal myocardium, scar tissue and MVO for the images acquired at different TI values using a TI scout (Look-Locker technique). The image where the normal myocardium appears darkest is identified as the optimal TI time. In this example, TI of 320 ms nulls the signal from the normal myocardium.

Figure 23

Phase sensitive inversion recovery sequence. By using a phase sensitive reconstruction, the sign of magnetisation is reflected in the displayed signal intensity. The signal intensity is mapped onto a grey scale showing values of zero at the centre of the grey scale, positive values as higher pixel intensities (towards white) and negative values as lower pixel intensities (towards black). Using this method, the effect of a small error in the choice of optimal TI is reduced. The normal myocardial signal can be suppressed by careful windowing of greyscale.

Figure 24

Relative SI levels in Phase Sensitive IR sequence. By restoring signal polarity (Figure 23), normal myocardial signal remains markedly lower than scar, even if the null point is underestimated by 40 ms. Note that negative signal intensities correspond to tissues who’s longitudinal magnetization has not recovered past M_z = 0 at time TI. For image display purposes, tissues with zero magnetisation values are mapped onto the middle of the image pixel intensity range. Tissues with a positive magnetisation are then displayed with image pixel intensities increasing from this middle value, while tissues with a negative magnetisation are displayed with images pixel intensities decreasing from this middle value. ‘Nulling’ the normal myocardium can then be achieved retrospectively by adjusting the window and level on the viewing console and there is no longer an ambiguity between tissues with positive and negative magnetisation values.

Figure 25

This schematic diagram illustrates the key principles of CE-MRA. The diagram in (a) shows the relative timing of the key events in a CE-MRA acquisition. Following the intravenous injection of contrast agent into a subject, there is a delay as the contrast bolus travels through the right and left sides of the heart before entering the arterial system. The red and black curves show the variation of concentration of contrast agent in the arteries and veins within the region of interest over time. Optimum image quality is achieved if the image data acquisition is timed to commence after the arrival of the contrast bolus within the arterial system of interest and is completed before the arrival of contrast in the corresponding venous system (sometimes referred to as the A-V window). The schematic diagram in (b) shows qualitatively how signal intensity relates to tissue type for a typical T1-weighted MRA pulse sequence. Indicative values for T1 relaxation time are also shown. Without the introduction of contrast agent, fat tissue would have the brightest signal in the image and blood within the vessel lumen would not be visible. The introduction of contrast agent into the blood reduces its T1 relaxation time and increases the signal intensity. The maximum concentration during first pass of the contrast bolus through the arterial system must be sufficient to reduce the T1 of blood to significantly below that of fat, so that the arterial vessel lumen yields the highest pixel intensities on the image.

Figure 26

Maximum Intensity projection (MIP). This diagram shows an overview of how projection images can be produced from multiple slice MRA image data. For 2D acquisitions, multiple thin overlapping slices are acquired. For 3D acquisitions, the selected volume is encoded and partitioned into multiple thin, contiguous slices. The MRA technique and its acquisition parameters are chosen to produce a signal intensity from the vessels of interest that is significantly higher than that of the background tissues. The maximum intensity projection (MIP) is produced using computer processing software that traces parallel paths through the data in the direction of the intended projection image, records the maximum pixel intensity encountered and stores this onto the projected image. The lateral and A-P views shown are typical of those produced automatically by the software immediately following image reconstruction. It is usually possible to also prescribe automated reconstruction of customized MIPs at any arbitrary angle. Further MIPs can also be produced retrospectively on a workstation. A disadvantage of the automated MIPs shown is that they often contain overlapping vessels and a relatively high signal from background tissue. This can be removed with interactive post-processing software so the MIP algorithm is only applied to a selected volume of data, sometimes referred to as targeted MIP.

Figure 27

Comparison of 2D and 3D image acquisition. This diagram shows the key differences between 2D and 3D image acquisitions. For 2D acquisitions (a) a slice is selectively excited, and the MR signal is encoded in two dimensions by phase encoding and frequency encoding. For conventional (not segmented k-space) acquisitions, the acquisition time, T_2D is determined by the TR, the number of phase encoding steps, N_P and the number of signals averaged, NSA. For 3D imaging (b), a volume (or thick slice) is excited and then encoded in three dimensions by phase encoding and frequency encoding as for 2D imaging and additionally by phase encoding in the volume selection direction. Compared to 2D imaging, the acquisition time for 3D imaging, T_3D, is increased by a factor equal to the number of phase encoding steps in the volume selection direction, N_S. In practice, N_S is greater than expected as more slices are acquired than specified by the user and discarded after reconstruction. This because the tissue excitation at the edge of the volume is less well defined and these images may also suffer from image wrap which is analogous to the image wrap experienced in the 2D phase encoding direction.

Figure 28

k-space order for 3D image acquisitions. The concept of k-space order for 2D image acquisitions was explained previously (see Part I, Figure 14) [1]. This diagram shows the two approaches to k-space ordering in 3D imaging that are equivalent to the linear and centric k-space orders in 2D imaging. In (a), the linear k-space order for 3D acquisitions begins and ends at the edge of k-space, passing through the centre of k-space at the mid-point of the acquisition. For each phase encoding step in one direction, (vertical in this example) the acquisition steps through all the phase encoding steps in the other direction (horizontal) so that the lines of k-space are filled layer by layer. In (b), the centric k-space order for 3D acquisitions begins at the centre of k-space and works outwards, adding layers in a pattern of concentric elliptical cylinders of increasing size. (The number of k-space lines in each phase encoding direction is usually different, hence the elliptical shape).

Figure 29

The two principle methods used for timing of the start of the CE-MRA data acquisition. For the test bolus method (top) a small bolus (1-2 ml) of contrast agent is first injected and a single-slice acquisition is used to acquire a dynamic series of images at approximately 1 second intervals in the region of interest (ROI). The image time series is analysed to determine the time at which the contrast bolus arrived in the ROI, T_ROI. The time delay T_D for the CE-MRA acquisition is then manually calculated taking into account the time duration of the bolus, T_B, the acquisition time of the MRA pulse sequence, T_A and the k-space order (in this case linear). For the fluoroscopic triggering method (bottom) the first acquisition acquires a dynamic series of images at approximately 1 sec intervals which are reconstructed in real time to allow detection of contrast arrival. On detection of bolus arrival, a trigger signal is generated either manually or automatically that causes the acquisition to switch to the CE-MRA pulse sequence. Manual triggering requires the operator to view the dynamic series of images in real time and to initiate the trigger once they judge that the contrast bolus has adequately progressed into the region of interest. Automated triggering monitors the signal intensity within a region of interest predefined by the operator on the dynamic image data set. The trigger signal is generated once the signal intensity rises above a threshold value.

Figure 30

This diagram shows the time- resolved MRA approaches offered by three vendors. In each case, the effective frame rate is defined as the time interval between central zone acquisitions as this determines how often the image contrast is updated. In (a) the vendor implementation of time resolved imaging with contrast kinetics (TRICKS), divides 3D k-space into four elliptical concentric zones. Data is acquired initially for all four zones and subsequently the acquisition of the central zone is interleaved with each of the three outer zones in sequence. The solutions from two other vendors divide k-space into just two zones. In (b) the TWIST technique initially samples the whole of k-space and subsequently acquires the central zone interleaved with the outer zone. As the outer zone is much larger, although k-space lines are updated over the full extent of the outer zone, many lines are skipped. The missing lines are then updated in subsequent outer-zone acquisitions. In (c) the 4D-TRAK technique acquires the whole of k-space at the beginning, followed by between 4–6 consecutive acquisitions of the central zone, after which the outer zone of k-space is sampled once more. The initial whole k-space acquisition is used as a mask which is subtracted from later reconstructions to remove the background tissue signal. Subsequent outer zone acquisitions are combined with the preceding 6 central zone acquisitions (indicated by the curved arrows) to reconstruct time resolved image data sets.

Figure 31

Respiratory gating and navigator echoes. This figure illustrates the principle behind the use of navigator echoes to gate the image data acquisition according to a particular time period within the respiratory cycle. The diagram in (a) shows a cardiac triggered data acquisition with a trigger delay chosen at mid-diastole to minimise the effect of cardiac motion. A curve representing the diaphragm position during respiratory motion is shown above. The effect of respiratory motion is limited by gating the data acquisition, so that image data is only accepted when the diaphragm position lies within a predefined ‘window’ corresponding to end expiration. A navigator pulse (shown in red) is applied immediately before the image data acquisition to excite a column of tissue cutting through the right hemi-diaphragm at right angles (b). The resultant navigator echo is frequency encoded along the length of the column and the navigator echo signal is analysed using a Fourier transform to produce a line of signal. The line signal from each successive R-R interval is added to a navigator display. A computer algorithm detects where the signal intensity changes from a high value (liver) to a low value (lung), representative of the diaphragm position. Where the diaphragm position falls within a predefined gating window the image data acquisition is accepted (indicated by the green dashes). Where the diaphragm position falls outside the gating window the data acquired is rejected and the acquisition is repeated until the diaphragm position again falls within the gating window.

Figure 32

Respiratory navigator-gated 3D Coronary MRA pulse sequence. The key features of a navigator-gated 3D coronary MRA pulse sequence are shown in (a). The image data acquisition is typically performed using a 3D fast or turbo gradient echo (FGE) pulse sequence with a trigger delay set to acquire during mid-diastole. The image data acquisition is typically preceded by three preparation pulses, a T2-preparation pulse, the navigator pulse (see Figure 31) and a frequency-selective fat suppression pulse (see Figure 2 and Figure 5). The fat suppression pulse is necessary to suppress the signal from fat surrounding the coronary arteries and is applied immediately before the image data acquisition to maximise the effectiveness of fat suppression. The T2 preparation pulse is used to reduce the signal from the myocardial muscle (short T2) relative to that of the blood (long T2). This prep pulse consists of a 90° rf pulse followed by a series of 180° rf pulses, similar to a multi-echo spin echo pulse sequence. This produces magnetisation in the transverse plane that is T2-weighted. The T2-weighted transverse magnetisation is then rotated back to the z-axis by a second 90° rf pulse, resulting in z-magnetisation for myocardium that is reduced relative to that of the blood within the coronary arteries. This improves the contrast of the resultant MRA images. Four slices from a 3D coronary MRA dataset are shown in (b). Note the absence of fat signal from around the coronary arteries and the reduced signal contribution from the myocardium.

Figure 33

This diagram illustrates how velocity-related phase shifts are caused by blood flowing along the direction of two equal but opposite magnetic field gradients applied in succession, (bipolar gradient pulses). Two gradient pulses are shown with gradient amplitudes G and duration t, separated by a time T. Immediately after the rf pulse is applied, both moving and stationary spins have the same phase (top left). When the first positive gradient pulse is applied (1), spins at position A experience an increase in magnetic field due to their position along the gradient. When the second negative gradient pulse is applied [2], stationary spins that remain at position A experience an equal decrease in magnetic field, causing the spins to move back into phase with one another. Spins in moving blood with velocity v, however, will have moved to position B, and experience an additional decrease in magnetic field that is proportional to the distance moved, vT, from position A. The signal from moving blood therefore acquires a phase shift, ϕ, relative to that of stationary tissue, that is proportional to velocity, v as shown (top right). Bipolar gradient pulses are used to design velocity-sensitive pulse sequences. From the equation shown (top centre) it can be seen that the velocity-related phase shift, ϕ, is also proportional to amplitude G, the duration, t, and the separation, T, of the gradient pulses. Velocity sensitive pulse sequences are designed to be sensitive over different velocity ranges through careful choice of these gradient pulse parameters.

Figure 34

Signal losses caused by velocity gradients. This diagram illustrates how signal losses occur in the presence of stenotic or regurgitant flow jets. The two images (bottom left) show signal loss (white arrows) caused by aortic valve regurgitation at two end systole and early diastole. Flow jets consist of a large range of velocities, from very high velocities at the centre of the jet, to relatively low velocities at the edge of the jet (top left). The change in velocity from the centre to the edge of the jet is often referred to as a velocity gradient. Figure 33 shows that when bipolar gradients are applied within imaging pulse sequences, velocity related phase shifts occur. Where there is a large range of velocities due to a velocity gradient within a pixel, this results in a large range of phase shifts (right), causing significant dephasing and therefore signal loss within that pixel. Outside the jet there is only a small range of velocities leading to negligible dephasing and no signal loss.

Figure 35

A retrospectively-gated, cine spoiled gradient echo pulse sequence (top) is shown in detail for four cardiac phases. In this example the standard imaging gradients are modified in the slice-selection direction to encode flow velocity components perpendicular to the image plane. For each cardiac phase the gradient echo pulse sequence is acquired twice, with two different flow sensitivities, indicated here by the red and green gradient pulses. The heart phase interval or ‘effective TR’ of the velocity encoded sequence is double the actual TR of the gradient echo pulse sequence. The M_x and M_y components of the MR signal are used to calculate both the signal magnitude, M, and the signal phase, ϕ, for each pixel. The phase maps for the two different flow sensitivities are subtracted to remove background phase shifts, creating a velocity map, which contains only velocity-related phase shifts due to the difference in flow velocity sensitivity between the two acquisitions. The flow sensitivities are chosen such that a subtracted phase difference of 180° corresponds to a predefined maximum velocity or VENC. The sign of the subtracted phase indicates the direction of flow along the encoded direction. The value of relative phase shift is mapped onto the pixel intensity scale such that a zero subtracted phase (stationary tissue) maps on to the centre of the pixel intensity scale (mid-grey). Positive and negative subtracted phases (corresponding to opposite directions of flow) are mapped onto higher and lower pixel intensities, respectively.

Figure 36

This figure illustrates some of the common pitfalls encountered with velocity-encoded MR imaging. Images (a) and (b) show a transaxial magnitude image and velocity map with ‘through-plane’ velocity encoding demonstrating forward flow in the ascending aorta (arrows). In (b), the maximum velocity is within the VENC chosen for the velocity-encoded acquisition, so that the phase shifts for all the pixels within the ascending aorta are less than 180° and are shown as high pixel intensities. In image (c), the VENC is set too low and the pixels at the centre of the ascending aorta (arrow) appear as negative velocities. This artefact, known as ‘velocity aliasing’ arises because phase shifts greater than 180° are interpreted as negative phase shifts and are mapped on to the lower pixel intensity scale (d). Images (e) and (f) show a magnitude image and velocity map acquired in an oblique sagittal plane to demonstrate flow in the aortic arch. The velocity encoding direction is chosen as ‘in-plane’ to demonstrate velocity components in the head-feet direction. In (f), the maximum velocity in both the ascending and descending aorta is lower than the VENC so that no aliasing is visible. For image (g) the VENC has been set too low resulting in velocity aliasing. For image (h) the velocity encoding direction has been incorrectly chosen to encode velocity components through the image plane, so that only relatively low transverse components of blood flow velocity are visible in the aorta.