MRI is a medical imaging modality utilising the distribution of net magnetic polarisation in an object in the presence of an externally applied magnetic field. Originally discovered in the form of nuclear magnetic resonance (NMR) [204], refinement by Lauterbur, Bloch, and Purcell [205-207] made the technique viable for clinical purposes. Today, MRI is used as a powerful tool for non-invasive structural and functional analysis, and is used for a number of purposes in cardiovascular diagnostics: accurately assessing intracardiac dimensions [208], differentiating myocardial scar tissue from myocardial injury [209], estimating 3D flow features throughout the cardiovascular system [10], and even assessing intracardiac fibre orientation using diffusion tensor MRI [210].
a) Fundamental imaging physics
The fundamental concept of MRI lies in the magnetic momentum or spin of sub-
atomic particles. The magnetic momentum can be likened by a magnet, where the particle has a positive and a counter-acting negative magnetic pole. For clinical imaging purposes the proton has such a defining spin, meaning that it also possesses certain magnetic properties.
Under normal circumstances, the spins of the protons in the human body is randomly oriented. However, if subjected to an externally applied magnetic field 𝐵𝐵0, the net spins will align and start to precess around the axis of 𝐵𝐵0. Specifically,
the precession or Larmor frequency 𝜔𝜔 will be related to the field strength of 𝐵𝐵0 by
where 𝛾𝛾 is the gyromagnetic ratio.
However, all spins will not orientate themselves identically, but rather there will exist a slight favouring of parallel orientation over anti-parallel orientation. Consequently, in the presence of 𝐵𝐵0, there will be a net magnetisation 𝑀𝑀 in the investigated tissue (being parallel to the applied 𝐵𝐵0).
Now by invoking a pulsed external field 𝐵𝐵1 (often called an RF-pulse) orthogonal to 𝐵𝐵0, 𝑀𝑀 will be tipped away from 𝐵𝐵0 (see Figure 3.14). Specifically 𝑀𝑀 will flip down towards the transverse plane orthogonal to 𝐵𝐵0, with the flip angle (deviation of 𝑀𝑀 from 𝐵𝐵0) governed by the strength and duration of the RF-pulse.
Once the RF-pulse is removed, an opposite motion will occur and the spins will start to relax back into their original 𝐵𝐵0-governed equilibrium state. The relaxation
is what forms the actual MRI-image, and is in principle governed by two processes:
i) T1-relaxation or spin-lattice relaxation, describing the time required for
the recovery of magnetisation in the direction of 𝐵𝐵0
ii) T2-relaxation or spin-spin relaxation, describing the time required for
the decay of magnetisation in the direction of 𝐵𝐵1
As the spin orientation and net magnetisation returns into the orientation of 𝐵𝐵0, these relaxation times or net movement of magnetic momentum will be proportional to the Larmor frequency of the tissue, enabling mapping of tissue morphology using dedicated receiver coils.
Figure 3.14: Fundamental principles of MRI signal generation. With the imposed magnetic field 𝐵𝐵0 a net magnetisation 𝑀𝑀 will be present. With an induced orthogonal RF-pulse, the net magnetisation will flip an angle 𝜔𝜔1 (maximum 90°). Once terminated, the net magnetisation will relax back from the flipped to the non-flipped configuration, with the change in magnetisation measured either as a function of magnetisation recovery in z-direction (T1) or as magnetisation decrease in the transverse xy-plane (T2).
b) Image formation
As described above, following an exciting RF-pulse different tissue structures can be identified based on their proton density and relaxation time. However, simple observation of the induced signal is not enough to spatially position the acquired data. For this, a specific pulse sequence needs to be applied to spatially encode the induced signal.
The gradient echo sequence is a typical MRI sequence used for such 3D positioning. In principle, a gradient echo sequence consists of three different sequential steps:
i) Slice selection, or selective RF excitation is represented by a spatial
magnetisation gradient 𝜈𝜈𝑧𝑧 applied in the direction of the main field 𝐵𝐵0, and simultaneous with the RF-pulse. Since the spins will now be excited to a gradient in the z-direction, the RF pulse can be tuned to only excite spins within a specific frequency range; only spins in resonance with the RF-pulse will be excited. Thus, positioning of the retrieved signal in z-direction can be achieved.
ii) Frequency encoding is represented by yet another spatial magnetisation
gradient, but this time by 𝜈𝜈𝑥𝑥. Using 𝜈𝜈𝑥𝑥, the Larmor frequency of the
excited spins will vary as a function of x-position, and can be analysed separately through Fourier decomposition of the retrieved output signal. Thus, positioning of the retrieved signal in x-direction can be achieved.
iii) Phase encoding represents the final spatial encoding, achieved by yet
another gradient 𝜈𝜈𝑦𝑦. Here the phase of the signal (rather than frequency) is analysed, where progressively increasing or decreasing 𝜈𝜈𝑦𝑦 is applied to accumulate different phase changes along the y-axis.
By analysing the final phase shift as a function of location, positioning of the retrieved signal in y-direction can be achieved.
In addition to the above, when imaging cardiovascular structures, data might also vary as a function of time (due to e.g. the AV-plane motion of the heart, or the respiratory displacements inside the intrathoracic cavity). For such, image gating is
typically utilised, where data is sampled over a number of consecutive cycles and either prospectively or retrospectively assigned to a specific temporal gate. For cardiovascular imaging, simultaneous ECG recording provides excellent correlation to specific cardiac phases, where acquisitions are typically triggered at the detection of a new R-tag (systolic onset). This however also means that data is acquired as a mean average over a number of cycles, and highly transient events might be difficult to capture.
c) Structural imaging
Using the principles in the previous section, the structure of biological tissue can be reconstructed based on their respective relaxation time. As described both T1 and T2-relaxation occur, wherefore structural MRI images are either described as T1-weighted or T2-weighted. For cardiovascular purposes, T1-weighted images are the most commonly used, with these images enhancing soft tissue contrast [211], as well as providing excellent contrast between blood pool and vascular tissue. T2- weighted images are less common for cardiac assessment, but has been employed to identify reperfusion injury [212], myocardial iron overload [213], and even global myocardial inflammation [214].
d) Functional imaging
MRI does not only enable the study of morphology, but also offers the ability to study functional physiological processes. To exemplify, functional MRI (fMRI, evaluating changes in oxygen-saturated haemoglobin) has been utilised to study brain activity [215], and time-of-flight MR angiography (evaluating the temporal process of magnetic saturation) has successfully been employed to identify aneurysmal developments [216]. For cardiovascular purposes, the use of phase contrast MRI (PC-MRI) to assess cardiovascular blood flow has become a fundamental part of cardiac MRI assessment [217]. In the following section, fundamental basics of PC-MRI are provided.
i) Phase-Contrast MRI
The imaging basics of PC-MRI is in part identical to the spatial position imaging described above; by applying varying gradients and utilising spin relaxation times, an image can be formed. In PC-MRI however, the application of a bipolar gradient is utilised: first a de-phasing is introduced by a positive gradient, which is then reversed by an identical but negative gradient. Any static particle will thus return to its original un-phased state, however any particle that has moved inside the object during the application of the bipolar field will have a net phase shift 𝜙𝜙. The concept is depicted graphically in Figure 3.15, showing that particles moving at identical velocity will after the bipolar gradient have identical phase shift.
Since velocities are assessed by an induced phase shift, a pre-defined velocity window or velocity encoding (VENC) will exist corresponding to 𝜙𝜙 = ±180°. Any
larger velocities inducing larger phase shifts will induce aliasing (wrapping), typically encountered in high-velocity PC-MRI flows. In such cases, an increased VENC (by modifying the magnitude of the bipolar gradient prior to acquisition) or
unwrapping algorithms (during post-processing) has to be applied. Importantly, the VENC will affect the resolution in acquired velocities, and is also directly related to image quality. In fact, it can be shown that VENC is related to signal-to- noise-ratio (SNR) as
Figure 3.15: Fundamental principles of phase-contrast MRI. Using a bipolar gradient (+𝜈𝜈/−𝜈𝜈)
spins with increasing phase shift 𝜔𝜔 will be induced by the initial positive gradient, before being reverted by the subsequent negative gradient. Static tissue (upper and lower row) will thus have a resulting net 𝜔𝜔 = 0. However, for moving tissue (middle row, moving to the right with velocity 𝒗𝒗), the movement through the spatial gradient will cause 𝜔𝜔 ≠ 0. Specifically, after the bipolar gradient, a given 𝜔𝜔 will be related to a given 𝒗𝒗 (with constant 𝒗𝒗 in the figure, the entire mid-row ends up with identical 𝜔𝜔).
𝑆𝑆𝑆𝑆𝑅𝑅 = √
2
𝜋𝜋 𝑉𝑉𝐸𝐸𝑆𝑆𝐶𝐶𝜎𝜎 , (3.27)
where 𝜎𝜎 is the standard deviation in the acquired velocity field.
Using slice selection, phase encoding, and frequency encoding, blood flow can also be mapped in 3D. Most commonly though is the selection of a single-plane PC- MRI, where so called 2D time-resolved (CINE) PC-MRI images are acquired, providing information on one-directional velocity through a given image plane.
ii) 4D flow MRI
If applying bipolar gradients in all spatial dimensions, a comprehensive mapping of 3D flow can be achieved. Adding image gating, so called 4D flow MRI (3D flow over time) can be assessed [10, 76, 79]. With such, complete mapping of cardiovascular flow is in principle enabled, and the technique has been extensively applied to study intracardiac flow, derive refined 3D flow parameters for cardiovascular assessment, as well as investigate flow in vascular disease [19, 22, 76, 91, 218-221]. Importantly however is to recall that with bipolar gradients applied in all three dimensions, 4D flow MRI typically requires relatively long scan times (to date around 5-10 minutes). Improvements have been made using accelerated pulse sequences or readout techniques [222, 223], but the acquired flow field will still represent a mean average of the cardiovascular flow over the entire scan time.