«INTRODUCTION The topic of magnetic resonance imaging (MRI) started for us at Nottingham in the early summer of 1972. During a discussion with one of ...»
Nobel Lecture, December 8, 2003
Sir Peter Mansﬁeld Magnetic Resonance Centre, Department of Physics
and Astronomy, University of Nottingham, Nottingham, NG7 2RD, U.K.
The topic of magnetic resonance imaging (MRI) started for us at Nottingham
in the early summer of 1972. During a discussion with one of my graduate stu-
dents, Peter Grannell and my post-doc Dr Allan Garroway, concerning mul- tiple-pulse line narrowing experiments in solids, the idea occurred to me to use the line narrowing technique as a means of effectively removing dipolar interactions in a material like CaF2 and at the same time impose an external linear gradient on the sample thus broadening the line shape to reveal the atomic or molecular structure within the sample.
It soon became apparent, however, even with the achievable narrowed line- widths of around 1 Hz for CaF2, corresponding to a line width reduction of 3 104, that the residual line width was still too broad using practical external gradients to resolve the atomic structure in a single crystal of CaF2. Despite this setback, work continued with artiﬁcial one-dimensional lattices made up of several thin plates of camphor.
Peter Grannell and I continued this work during the course of 1972 and it resulted ﬁnally in a paper presented at the First Specialized Colloque Ampère, Krakow in 1973 (1). Formal publication appeared shortly after (2).
These papers emphasized the Fourier transform approach used, even though the images of the camphor stacks were one-dimensional. It was clear that we had made our task much more difﬁcult by choosing to work with solids.
Thoughts rapidly turned to liquid-like spin systems where the line narrowing approach would be unnecessary.
The imaging approach considered so far was essentially one or two-dimen- sional. The next step was to deﬁne a thin slice of material so that this would be imaged without spill over to adjacent planes. This was achieved using a technique called selective irradiation (3).
One of the major practical difﬁculties encountered with MRI to this point was the time it took to acquire the data. Line-scanning, for example, took typ- ically 10–20 min to acquire an image comprising 64 64 pixels (4).
The breakthrough came in 1977 with the introduction of echo-planar imaging (EPI) (5). This snap-shot technique meant that in principle com- plete two-dimensional images could be achieved in extremely short times 266 ranging from 20–50 ms. However, to achieve these acquisition times, yet an- other inventive step was required. This was the introduction of active mag- netic screening (6,7). Ordinary magnetic gradient coils, necessary to deﬁne the slice thickness and image axes, were found to interact strongly with the metal cryostat structure of the super-conductive magnet. Time dependent currents induced in these structures would decay away with their own inde- pendent time constant, adding an undesirable and unpredictable time dependence to the otherwise static magnetic ﬁeld.
By actively shielding the gradient coils, all extraneous time dependence is obviated, together with all undesirable reﬂected static magnetic ﬁelds. Of course, the magnetic screening process itself introduces magnetic ﬁelds which change the character of the gradient ﬁeld but in a well deﬁned and calculable manner.
Magnetically screened gradient coils now form an integral part of virtually all commercial MRI scanners.
EXPERIMENTAL APPARATUSFigure 1 shows a home built 0.5 T whole body MRI scanner together with a patient bed support. Figure 2 shows a doubly screened gradient coil assembly used in the experimental scanner of Figure 1. With a doubly screened gradient coil in which the primary coil comprises a single current loop, it can be shown that the unshielded loop carrying current I produces magnetic ﬂux Figure 1. Photograph of home built magnetic resonance imager based on a 0.5 T super-conductive magnet.
lines which form a series of elliptic like magnetic ﬁeld loops with displaced centres (7). However if the current loop or the primary coil is now magnetically screened with the double screen arrangement, the magnetic ﬁeld lines within the inner screen are exactly the same as produced by the primary coil in free space. However, between the screens the magnetic ﬂux is conﬁned and the ﬁeld inverts. The magnetic ﬁeld at the center of the coil assembly takes the same form and magnitude as the unscreened coil. That is to say, the magnetic ﬁeld B within a current loop is as expected for an unscreened loop.
Outside the outer screen the ﬁeld B = 0. If the inner magnetic screen is removed leaving the outer screen, ﬂux conﬁnement between the screens can no longer occur and the magnetic ﬁeld now leaks beyond the outer screen.
268 Figure 3. Diagram of a slice through the mediastinum showing the two lung ﬁelds and heart mass, also shown is the Fourier transform of this real-space image to the k-space map.
(Reproduced with permission from M K Stehling, R Turner and P Mansﬁeld, SCIENCE 253, 43–50 (1991).) and where k and r may both be one-, two- or three-dimensional. The density (r) describes the image in real space and S(k) describes the image in kspace. In our case G(t) is an externally applied time dependent magnetic gradient, is the magneto-gyric ratio which is constant for a particular nuclear spin species and t is the evolution time.
Using the Fourier transform expression in Eq.  allows a reversible transformation from k-space to real or r-space. This is exempliﬁed in the forward and inverse Fourier transforms shown in Figure 3.
In fact the two dimensional k-space image is built up from a series of one dimensional free induction decays, (FID’s), suitably stacked to give an image which is in effect the diffraction pattern of the object. The inverse FT of the k-space image produces the r-space image. In this case a transverse cross-sectional image through the mediastinum showing a diagram of the heart mass and lung ﬁelds.
In order to obtain the k-space map in a single experiment a specially designed pulse sequence is applied as shown in Figure 4. In general terms this comprises an initial spin preparation phase followed by a transverse slice selection pulse. This creates an active magnetic signal or FID which is allowed to decay away in the presence of the spatial encoding gradients Gx and Gy. For EPI these are applied in the form of a square wave or a trapezoidal waveform for Gy and either a long low level pulse or a train of short blipped pulses for Gx. In either case the areas under the long low pulse or the string of blips must be equal. The effect of these gradient waveforms causes the FID follow
ing slice selection to dephase and rephase in a series of spin echoes. The amplitude of these spin echoes is initially low but grows to a maximum and then decays provided an initial dephasing gradient pulse is arranged immediately before the blipped or low level Gx sequence starts.
Also included in Figure 4 is the k-space trajectory. This starts at kx = ky = 0.
The pre-pulse described above displaces the trajectory from 0 to -kxmax at which point the locus starts from ky = 0 moving right to kymax. At this point the trajectory is displaced upwards with the ﬁrst positive blip. The transverse scan then proceeds from right to left traversing the full ky scan from kymax to -kymax.
In this way the whole of the k plane is scanned. During this journey the signal is regularly sampled to produce the ﬁrst step in obtaining the k-space scan as sketched in Figure 3. The remaining operation is an editing function which reverses the order of all data in alternate lines of the k-space image. At this point there are two options to choose from in order to produce the r-space image. The ﬁrst is to perform a two-dimensional FT on the k-space map thereby producing the r-space image. However, to do this in general requires an overhead in time for data manipulation in addition to the two dimensional FT. The alternative approach is to take the ﬁrst and last points in the reordered k-space scan and, metaphorically speaking, pull the whole array out, like a string of beads, to form a one dimensional array. The whole string is Fourier transformed using a one dimensional FT with less computing overhead.
270 Figure 5. Snap-shot EPI images through the heart obtained with use of a surface coil. (1) Transection during systole shows left ventricular myocardial wall thickening. (2) Rapid ventricular ﬁlling in late systole. (3, 4) Transections obtained during diastole show thinner myocardial walls. The spatial resolution of these images is less then 2 mm. (Reproduced with permission from M J Stehling et al., RADIOLOGY, 170: 257–263, (1989).) General Two Dimensional Imaging Results at 0.5 T The above described EPI sequence was used to image a series of patients and volunteers during the early eighties through to the nineties. Whole body scans may be performed comprising 64 transections commencing in the upper thorax and moving down the torso in 5 mm steps through the mediastinum, liver, kidney into the lower abdomen and ﬁnishing just below the bladder. They are an example of a quick sighting scan, the whole imaging process taking approximately one minute.
Figure 5 shows four EPI snap-shots taken through the heart using a surface coil on the chest wall (8). Each image is acquired at a different phase of the cardiac cycle. Figures 5.1 and 5.2, for example, correspond to different phases during systole, when the heart is in contraction and pumping. Figure 5.2 shows loss of signal (black) due to turbulence within the left ventricle. Figures
5.3 and 5.4 were acquired during the relaxed phase or diastole. In all images the myocardium gives a less intense signal allowing differentiation of the muscle tissue from blood.
Figure 6 shows four transections through the liver of a patient with a series of hydatid cystic lesions (9). These show as bright regions within the darker liver. An imaging variation on the same patient using the inversion-recovery or IR-EPI sequence uses an initial spin inversion pulse as the preparation phase. After a short delay or inversion time TI, the EPI sequence follows.
When TI = 0 we have the normal sections as in Figure 6. By varying TI, short
T1 signal components within the image may be effectively removed thereby delineating the normal tissue. This procedure can be used to indicate clear margins between the lesions and the normal tissue. By increasing TI long relaxation time components may be eliminated thereby revealing the relatively fast relaxing hydatid cystic lesions.
Foetal Imaging Results at 0.5 T An important area of application of EPI has been foetal imaging during the 3rd trimester. This has been valuable in assessing cases of foetal growth retardation (10–12).
Figure 7 shows a maternal coronal view of a foetal saggital section at 37 weeks gestation. The foetal section, taken from a set of scans, clearly shows the head, brain and spinal column and also the right leg. The darker signal from within the amniotic sac is the placenta. The brighter signal regions surrounding the foetus come from the amniotic ﬂuid.
In other transectional images at 37 weeks gestation, highlighting of the lung ﬁeld can be used to measure foetal lung volume. The same technique can be used to measure the subcutaneous fatty tissues surrounding the foetus.
Foetal birthweight versus the EPI estimated foetal volume has also been measured for 12 babies with birthweight spanning the range 1.5–3.5 kg 272 Figure 7. Saggital section taken from a three-dimensional data set of a foetus in utero at 37 weeks gestation. The foetal brain and spinal canal can be clearly seen. Also shown to the right and slightly darker is the placenta. The bright region between the placenta and foetus corresponds to amniotic ﬂuid.
(11,12). The foetal volume is assessed from a series of contigious EPI slices spanning the foetus. Using the same technique as described above, the foetal cross-sectional image is traced out and highlighted. The elemental volume is the highlighted area multiplied by the slice thickness. The total foetal volume is then the sum of all cross-sectional slices. The results indicate a good linear correlation between volume and weight.
Paediatric Imaging at 0.5 T In the course of imaging young children suffering from cyanotic heart disease, (13–16), snap-shot transectional images have been taken through the mediastinum for a child with a normal heart. They all show the lung ﬁelds and heart mass at the base of the heart, moving down through the left and right ventricles towards the apex. Also shown is the classiﬁcation system for a normal heart. This is contrasted with an example taken from a set of images for a child with a truncus arteriosus. Towards the apex the left and right ventricles can be clearly seen as dark regions together with the intact ventricular septum. The left and right atria show as bright regions. The classiﬁcation diagram indicates the coalescence of the pulmonary artery and the aorta into a common truncus.
EPI Results at 3.0 T A number of snap-shot images of various objects have been obtained at 3.0 T (17), including a snap-shot image of a phantom comprising 256 256 pixels.
The slice thickness was 0.5 cm. The image acquisition time was approximately 90 ms. Other examples include images through the brain of a normal volunteer. The posterior horns of the ventricles together with part of the anterior horns are just visible in these images.
ECHO-VOLUMAR IMAGING (EVI) Imaging Sequence Details of the pulse-timing sequence for EVI (18) are shown in the upper part of Figure 8. The top trace shows the Gy gradient modulation waveform. The next modulated waveform is the gradient Gx. The third waveform Gz includes the slice selection gradient Gz together with a negative pre-positioning pulse immediately followed by a long low level gradient. In the bottom trace the slice selection pulse is indicated and slightly later the modulated spin signals arising from the sequence are sketched.