· Appreciation of the use of EPI in fMRI, dynamic susceptibility contrast imaging, arterial spin labelling, diffusion imaging and chemical exchange saturation transfer imaging.

· Understanding of the different approaches that can be used for each method, and the best ways to minimize confounding artifacts.

MR physicists, plus technologists and physicians with an interest in understanding the underlying MR physicsEcho planar imaging has become an important component of many MRI methods that require rapid and efficient image readout, either because a dynamic time course is required (e.g. functional MRI or dynamic susceptibility contrast imaging), or because a moving organ is being imaged (e.g. cardiac imaging or foetal imaging), or because a large amount of data with different image contrasts are needed to obtain a final image (e.g. diffusion imaging or CEST). This lecture will introduce some of these topics and will highlight how EPI plays a role in these sequences.One of the areas in which EPI made a critical contribution is in the area of functional MRI [1, 2, 3]. Although fMRI is possible using more conventional gradient-echo sequences, EPI offers a robust and reliable acquisition option that avoids many of the phase artifacts that can result from slower line-by-line k-space acquisition sequences. Both gradient echo and spin-echo EPI have been used to perform fMRI, with echo times (TEs) set to the local T2* or T2 for maximum blood oxygenation-level dependent (BOLD) sensitivity. It is also possible to use BOLD weighted sequences to probe other physiological parameters such as oxygen extraction fraction and oxygen metabolism. This can be achieved either by quantitative fitting to a model of BOLD contrast [4, 5], or by using additional inspired gas challenges to ‘calibrate’ the BOLD signal [6-9]. A final widely used dynamic application of EPI in the brain is its use in dynamic susceptibility contrast (DSC) imaging [10]. For DSC a bolus of extravascular contrast agent (usually Gd-DTPA) is injected, which is then imaged as it passes through the tissue bed. Using such data the regional blood volume can be calculated and, theoretically, the blood flow. DSC has found wide use in the diagnosis and acute treatment planning of stroke, where the time-to-peak of the DSC R2* map is used as a surrogate for blood flow.

A variety of preparation modules exist with the intention of labelling blood. In some cases these preparation modules have the intention of crushing the blood signal (e.g., vascular space occupancy – VASO – imaging). In other cases the preparation modules intend to selectively excite the blood signal, for use in perfusion mapping or angiography. This latter class of labelling approaches is that of ‘arterial spin labelling’.

Arterial Spin Labeling

The technique of arterial spin labelling (ASL) was first described by Detre and colleagues in 1992 [11], and relies on the subtraction of two images, one that is prepared with the blood in a fully relaxed state (the ‘control’ image) and one that is prepared with the blood in an inverted (or saturated) state (the ‘tagged’ or ‘labelled’ image). The difference between these two images is proportional to the amount of blood that has arrived in the voxel between the time of the labelling and the time of the readout.

There are a number of ways in which the ASL preparation can be effected in practice. The earliest form described by Detre involved continuous labelling of the arterial blood (CASL) using a long (continuous or as near continuous as the hardware will allow) RF pulse in the presence of a field gradient. The motion of arterial blood through the labelling ‘plane’ leads to a flow-driven adiabatic inversion of the moving spins, which then travel to the tissue region of interest (relaxing as they go, thus limiting the ultimate lifetime of the ‘tag’). The control pulse for a CASL preparation, as well as not perturbing the blood spins, needs to mirror the effects of magnetization transfer, so that those effects cancel along with the static tissue signal and do not lead to a confound. In early publications the control pulse consisted of a ‘label’ placed distally above the slice of interest by the same distance as the proximal ‘tagging’ pulse. However, this approach limits CASL acquisition to a single-slice readout, since MT effects are only cancelled when the proximal and distal distances match. However, an alternative control pulse approach was proposed by Alsop [12] that enables multi-slice CASL data to be collected.

Another class of ASL preparations are the pulsed-ASL (PASL) approaches. In these preparations a volume of tissue is inverted using a short 180º pulse that labels the proximal arterial blood, and this is contrasted with a control pulse that does not label the blood. The first such description was the EPI Signal Targeting by Alternating RF (EPISTAR) preparation [13], in which a proximal slab was inverted and contrasted with an image in which no inversion pulse was used (in both cases the preparation was preceded by a saturation pulse in the location of the imaging slice to minimize effects from an imperfect inversion profile). The EPISTAR sequence was sufficiently insensitive to MT effects that no RF was used for the control pulse, but accurate quantification was problematic for this reason. The later method of Flow Alternating Inversion Recovery (FAIR) allowed a better degree of perfusion quantification [14]. The FAIR method contrasts a global (or large slab) inversion pulse with a slice selective inversion (both centred on the slice(s) of interest). In the case of the global inversion pulse, blood outside the slice of interest is inverted and begins to move into it. In the case of the slice-selective inversion pulse, non-inverted blood moves into the slice to provide contrast. In both cases the static tissue in the slice of interest is inverted and therefore cancels as it has the same magnetization recovery history.

The most widely used PASL methods are those based on the PICORE preparation [15], in which a proximal inversion slab is used for labelling, and an off-resonance MT-matched RF pulse is used for control. The PICORE variants of QUIPSS, QUIPSS II, and Q2TIPS further refine the principle and allow differing degrees of perfusion quantification and slice coverage.

An increasingly prominent ASL method, however, is yet another approach known as pseudo-continuous ASL (pCASL). Rather than using CASL’s lengthy high duty-cycle hard pulses the pCASL approach uses trains of low flip angle selective pulses to induce a flow-driven adiabatic inversion [16]. The theory behind pCASL is derived from SSFP sequences, in which off-resonance-dependent magnetization profiles result from a train of RF pulses. In this case an appropriately shaped Mz transition is arranged to occur about the centre frequency, and the position of the centre frequency of the selective pulses is located at the desired ‘labelling’ plane. Blood flowing through this plane will then experience the imposed Mz transition and will be inverted as it passes through it. The beauty of the pCASL approach is that a perfectly MT-matched control pulse can be applied simply by phase cycling the RF pulses to shift the transition band outside the ‘tuned’ labelling position.

A final major class of ASL preparation is the velocity-selective (VS-ASL) approach [17]. This preparation seeks to label the magnetization on the basis of its velocity profile, rather than its spatial position. If this can be done at velocities that are sufficiently low (ideally spins decelerating into the capillary bed) then it should be possible to minimize the need to include an arterial arrival delay between the labelling preparation and the imaging readout. If successful VS-ASL can be achieved then it may allow discrimination between absent flow, and flow that is simply delayed in its arrival to the tissue bed (and hence blood that is losing its T1-time-constant label) either due to slow flow or due to circuitous flow.

Vessel-Encoded ASL

Returning to pseudo-continuous ASL, a further feature of this approach is its easy extension to encode the labelling from different feeding vessels. This is possible by adding transverse field gradients between the RF pulses in the pCASL pulse train such that the phase accumulation between successive RF pulses can be adjusted at different spatial locations within the labelling plane to be in either a control state (net phase accumulation 180º between pulses) or a tag state (net phase accumulation 0º between pulses). The locations of the ‘tag’ and ‘control’ locations can then be moved around at will, and the perfusion territories [18] or vascular branches [19] from the feeding vessels can be decoded in post processing. By adding the concept of a variable delay time between labelling and readout it is possible to achieve full vessel-selective quantification via a fit to some form of kinetic model [20], yielding both absolute perfusion as well as arterial transit delay (also known as bolus arrival time). Although other CASL and PASL approaches have also been used to achieve selective vessel encoding, the pCASL approach seems to be the most robust and SNR efficient approach.

Another related use of ASL-like principles is in the study of blood volume changes using the method of vascular space occupancy (VASO) [21]. This method uses inversion pulses to invert the signal from arterial blood, much like conventional ASL, but times the readout of the image to occur at the T1null of the inversion-recovery curve. As such, blood signal from the arterioles and capillary bed is absent, allowing observed signal changes to be ascribed to the tissue compartment, whose signal actually decreases with blood volume increases during activation. In this way VASO can be used as an alternative fMRI approach, and may offer better localization of the fMRI signal changes to the grey matter tissue.

MRI has been used to measure the self-diffusion of water or other substances for many decades. Borrowing from the early work of Stejskal and Tanner [22], who showed how a pair of gradient pulses could be used to measure the diffusion coefficient of a sample, LeBihan was the first to incorporate diffusion measurement into an imaging experiment in the 1980s [23]. The essential experiment has remained unchanged, in that a large gradient pulse is applied that causes phase accrual to occur in the excited spins, that is then reversed using a second balanced gradient pulse. In the absence of diffusion the full signal (albeit T2-weighted) is recovered. However, in the presence of diffusion there is additional phase misalignment and hence signal cancellation caused by random movement of the spins in the direction of the field gradients. By careful control of the strength of the diffusion encoding gradients a quantitative measure of the diffusion coefficient can be achieved.

Diffusion Tensor Imaging

A further extension to diffusion imaging, also pioneered by LeBihan, is the ability of diffusion images to encode diffusion along different physical directions. By measuring a minimum of 6 different directions (e.g., x, y, z, xy, xz and yz) it is possible to characterise the diffusion tensor associated with each voxel [24]. This allows the principal diffusion direction to be determined and hence allows tractography to be accomplished. In order to help differentiate higher order fibre bundles (such as crossing fibres) it is necessary to extend the acquisition to more diffusion directions and even to multiple diffusion b values [25].

Readout Trains

Single-shot EPI is the most common readout sequence used for diffusion imaging, since it offers great robustness to phase errors associated with small motions during the diffusion encoding process. However, single-shot EPI sequences are sensitive to eddy current effects that can differ between different diffusion encoding directions. This has led to the need to develop correction strategies to deal with the resulting subtle mis-registrations between the different directions [26, 27]. An additional tactic to minimize eddy current problems is to use diffusion gradient patterns that are inherently eddy-current compensated [28].

If higher spatial resolutions are desired than are possible with single-shot EPI then multi-shot methods must be used. These require ‘navigator’ information to report on and facilitate correction of any phase errors arising from subject motion. An example of such a sequence is the readout-segmented EPI sequence [29], in which segments of k-space are built up over several TR periods, and a navigator acquisition is acquired alongside every segment.

Magnetization transfer contrast (MTC) in a biological context was first described by Wolff and Balaban in 1989 [30] and involves the exchange of magnetization from slow moving macromolecular spins to fast-moving free water spins. This is accomplished by partially saturating the macromolecular spins using off-resonance RF pulses. Because the macromolecular spins have a broad NMR spectrum they will be affected by off-resonance saturation, which would not normally directly affect the free water pool. However, because there is close contact between the free water pool and the macromolecular pool the saturated macromolecular magnetization will transfer its influence into the free water pool and will result in a lower free water signal. Hence, MTC weighted images can report on the macromolecular content of a voxel, and have applications in the assessment of tissue integrity (e.g., in the assessment of white matter integrity in multiple sclerosis).

A more recent extension of the MTC concept is the method of chemical exchange saturation transfer (CEST) contrast, also first described by Balaban’s lab [31]. It is closely related to MTC and also relies on off-resonance excitation and magnetization exchange, this time between more mobile solute species and water spins. CEST has various potential applications ranging from stroke, to cancer to musculoskeletal MRI.