GB2445759A - Magnetic resonance imaging scanner - Google Patents

Abstract

A magnetic resonance scanning apparatus has a first magnet having a ring shape occupying an x-y plane and configured to generate, within a sampling volume defined within, and in the plane of, the ring, a Bo magnetic field in the z-direction. Axial inhomogeneity outside the x-y plane is also provided. This is a means of slice selection. The Bo field has linear radial inhomogeneity in the x-y plane, which is a means of selecting a contour slice (a circumferential portion of the sample extending around the z-axis and in the x-y plane of the sampling volume) within the sampling volume via variable frequency rf. A second magnet is configured to generate, within the sampling volume, a gradient field having a field in the z-direction and gradient dBz/du where u is an axis in the x-y plane, such as the x or y axes.

Description

MAGNETIC RESONANCE IMAGING SCANNER

The present invention relates to magnetic resonance imaging scanners and related methods of making and using such scanners for medical and non-medical applications.

It is well known that whole-body MRI scanners are expensive, immobile units that require their own shielded room and skilled operators. For these reasons, waiting lists tend to be long and the technology is not widely distributed outside the hospital environment. In response to these limitations, a number of MRI instrument manufacturers have developed intermediate cost IvIRI scanners suitable for imaging body extremities such as the head, limbs and joints. Representative scanners include Hitachi's "AIRIS" scanner, which uses permanent magnets creating a 0.2 T field and Esaote's "ARTOSCAN C" scanner. These units are immobile and use large permanent magnets, but have been designed with "open access" for patients' limbs. Their imaging protocols still require a conventional

stationary homogeneous magnetic field.

Development of the "MAGNEVU" scanner represents a further generation of Iv[R.I instrumentation in that it is relatively small, light-weight and mobile. However, it has a small field of view which limits its use, e.g. mainly to the study of inflammatory and erosive joint disease in rheumatology. Although the scanner itself is on wheels and so is mobile, the imaging protocol still requires stationary magnet and patient. Moreover, it differs from the apparatus to be described in the present disclosure in that it is based on spread spectrum imaging and phase encoding. It also develops a two-dimensional image of a planar constant-field surface through the sample, unlike the apparatus to be described herein, which builds a planar image by deliberately making the field inhomogeneous within the plane.

It is an object of the present invention to provide a low cost, easy-to-use MRI scanner.

According to one aspect, the present invention provides a magnetic resonance scanning apparatus comprising: a first magnet configured to generate within a sampling volume defined within an x-y plane, a B0 magnetic field in the z-direction the B0 field having radial inhomogeneity in the x-y plane and axial inhomogeneity outside the x-y plane; a second magnet configured to generate, within the sampling volume, a gradient field having a field in the z-direction and gradient cLB/du where u is an axis in the x-y plane; and an RF coil configured for generating a B1 magnetic field transverse to the

Bo field within the sampling volume.

According to another aspect, the present invention provides a method of gathering magnetic resonance data comprising the steps of: generating, within a sampling volume occupying an x-y plane, a B0 magnetic field in the z-direction, the Bo field having radial inhomogeneity such that the B0 field varies monotonically as a function of radial distance within the x-y plane from a central z-axis of the sampling volume for all azimuth angles about the z-axis, and axial inhomogeneity outside the plane of the sampling volume; generating, within the sampling volume, a gradient field having a field in the z-direction and gradient dB/du where u is an axis in the x-y plane; generating an RF B1 magnetic field transverse to the B0 field within the sampling volume; and obtaining magnetic resonance data in respect of a contour slice within the sampling volume, the contour slice being a circumferential portion of the sample extending around the z-axis and in the x-y plane of the sampling volume.

Aspects of the MRI scanner as described herein may include one or more of the following features.

The scanner may use a circular, thin ring-shaped or "polo-mint" shaped magnet assembly (hereinafter referred to as a ring magnet) oriented in the x-y plane of a Cartesian coordinate system which creates a static inhomogeneous magnetic field having both radial inhomogeneity in the z = 0 plane and outside the z = 0 plane.

Throughout the present specification, the ring axis is defined as the axis passing through the centre hole of the ring and orthogonal to the plane of the ring.

The scanner may use the principle of motional relativity which constructs images not by rapidly switching magnetic field gradients, but by moving the fields over the sample in the z-direction orthogonal to the plane of the ring magnet assembly.

Alternatively the sample could be translated through the plane of the stationary magnet.

The scanner may use a protocol for acquiring a three-dimensional image of a sample in the inhomogeneous magnetic field, based on z-slice selection, radial contour slice selection and one-dimensional projection imaging.

The scanner may use a protocol for acquiring three-dimensional images in which only the orientation and not the magnitude of a transverse gradient is changed.

The scanner may use pulsed pre-polarisation of an electromagnet to increase signal / noise ratio in the case that permanent or electromagnet magnet designs are chosen.

The MRI scanners as described herein may be used in medical, veterinary, retail and laboratory applications. The MRJ scanners as described herein may have applications for routine use in hospitals for clinical diagnosis andlor could be suitable for widespread availability in doctors' surgeries, veterinary clinics and even sports centres. Other uses may include in the laboratory for research in the biological and food sciences, materials science and process engifleering.

Embodiments of the present invention will now be described by way of example and with reference to the accompanying drawings in which: Figure 1 is a perspective view of a possible configuration of an MRI

apparatus of the present disclosure;

Figure 2 is a schematic diagram illustrating: (a) a partial side cross-sectional view of the apparatus showing slice selection along the z-axis; (b) a

-I

partial view on the z-axis showing radial contour' slice selection with an enlargement inset, giving one-dimensional projection imaging; Figure 3 is a perspective schematic diagram illustrating translation of the scanner assembly along the z-axis or, alternatively, translation of the sample through the magnet; Figure 4 shows a pulse sequence for acquiring a two-dimensional image in thez=Oplane; Figure 5 is a schematic diagram illustrating the relationship between distorted radial contours of figure 2(b) and the one-dimensional image projections; Figure 6 is a schematic view on the z-axis of a B0 ring magnet formed as concentric rings of permanent magnets, each polarised in the z-direction, i.e. perpendicular to the plane of the paper; Figure 7 is a schematic cross-section through the B0 magnet rings showing

calculated Bo field inhomogeneity;

Figure 8 is a graph showing B field strength as a function of radial distance from the axis of the B0 ring magnet, where R is the radial distance from the centre ofthex-yplane atzO; Figure 9 is a schematic plan view of an alternative design of B0 ring magnet; Figure 10 is a graph illustrating radial inhomogeneity in the B field calculated for the three coil configuration of figure 9 with currents optimised for greatest radial homogeneity; Figure 11 is a schematic plan view of an alternative design of B0 ring magnet based on five concentric current-carrying coils; Figure 12 is a graph showing the radial profile of the B0 field calculated from the coil-based ring magnet geometry of figure 11 with current flows and directions optimised to minimise the quadratic and quartic radial gradient in the centre; Figure 13(a) is a graphical representation of the x-y positions at z = 0 of coil loops of a short actively-self-shielded RF saddle coil; Figure 13(b) is a graphical representation of the resulting B1 field at z = 0 in the x-y plane illustrating field strength and direction by arrow length and direction, from the coil of figure 13(a); Figure 13(c) is a graphical representation of the resulting B1 field at z = 0 in the x-y plane illustrating field strength in grey scale, from the coil of figure 13(a); Figure 14 is a graphical representation of the theoretical Q-curve for an RF probe; Figure 15 is a schematic perspective view of part of a transverse gradient coil having two opposing quadrants for generating a transverse G gradient; Figure 16 is a schematic perspective view of the transverse gradient coil of figure 15 showing the current flow into a single quadrant of the coil; Figure 17 is a graphical representation of the calculated G gradient along the y-axis at z = 0 fromthe coil geometry of figure 15; Figure 18 is a schematic view along the z-axis of an alternative transverse gradient coil design based on two overlapping quadrant layers having 1000 and 800 arcs overlapping; Figure 19(a) is a schematic cross-sectional view of a pair of Helmholtz coils located on the two faces of the ring magnet of figure 6 and carrying current in

opposite directions to create a G gradient field;

Figure 19(b) is a graphical representation of the positions of the Helmholtz coils of figure 19(a) in x-y-z space; Figure 20 is a graph showing the G gradient along the z axis as a function of z-position calculated from the coil geometry of figure 15; and Figure 21 is a graph showing an expanded view of the G gradient of figure showing the near-linear gradient close to the z 0 plane and extending a centimetre either side of the plane.

Figure 1 shows a possible configuration of an MRI apparatus 10 defining a circular cylindrical cavity 11 having an axial length (z-axis) and a diameter (x and y axes).

The cylindrical cavity is preferably open access from both ends to allow passage of objects to be imaged therethrough. A suitable transport mechanism (not shown) may be provided to convey objects to be imaged through the cavity 11, or alternatively the magnet arrangements to be described that surround the cavity may be motorised to sweep a sampling volume over a stationary object within the cavity 11. The MRI apparatus may be portable with an integrated data input / output device 12 and control systems 13.

Figure 2 shows the imaging concept. The Mk1 apparatus 10 is based on an imaging protocol which exploits motional relativity and mechanical slice selection in which the magnet arrangements are moved over the sample 23, or conversely the sample 23 is moved through the magnet arrangements. The apparatus 10 uses the magnetic field created in the central hole of a ring shaped magnet 20. This magnetic field is designed to be radially inhomogeneous in the central plane 21 of the ring magnet and preferably also inhomogeneous outside the plane of the magnet. The magnetic field can even be azimuthally inhomogeneous to a small degree, i.e. as a function of azimuth angle around the z-axis. This deliberate field inhomogeneity obviates the need for expensive shim coil sets. The apparatus 10 may also use non-switched transverse gradients during each angular (azimuthal) image acquisition step. Moreover, perfectly linear field gradients are not required.

The imaging protocol is based more on the line-scan technique, except that the "lines" are distorted concentric contours in the x-y plane. Computing power is utilized to relieve hardware design specifications, compensating for field imperfections and poor signal / noise ratio at the acquisition and processing stages.

Furthermore, interpretation is now more feasible via the Internet, thereby reducing the need for on-site highly skilled interpretive personnel.

If the ring magnet is constructed from permanent magnets it is very low-cost but also low field (<0.1 T). However, alternative designs based on cooled current-carrying resistive electromagnets or superconducting magnet coils can also be considered. The latter may be more expensive, but may also be higher field.

Although signal / noise ratio is limited by the low main magnetic field in a permanent magnet design, there are nevertheless several advantages to working at low field that should be noted. First, it is not usually signal / noise ratio that clinicians are concerned with but rather the ability of the image to discriminate regions of interest such as diseased areas or turnours. But this is related much more to contrast than to signal / noise ratio, provided there is sufficient spatial resolution in the image. It is therefore important to note that there is usually greater T1 relaxation time contrast at low field because of differences in the T1 f frequency dispersion. Moreover, because T1 is usually shorter at lower field, acquisition repetition times requiring relaxation of longitudinal magnetisation back to equilibrium can be faster, allowing more acquisitions per unit time. The 12 relaxation time is usually longer at low field because susceptibility effects are negligible and proton exchange effects greatly reduced. This permits more echoes per unit imaging time so RF power deposition decreases.

These features should enable the development of an ultra low cost, mobile M1RI scanner that is not only useful for preliminary examinations in hospitals but also should find widespread application throughout the wider community, including veterinary practices, sports clinics, research laboratories and for quality control in the manufacturing sector and at retail outlets.

The MRJ apparatus, in one embodiment, uses the principle of motional relativity to build a three-dimensional image of the object from a series of two-dimensional slices. To clarify the discussion, we use the Cartesian coordinate system as shown in figure 2. Unlike conventional M1RI, which is based on creating a homogeneous main magnetic B0 field over a defmed volume, the MIRJ apparatus described here seeks to create an image only over a substantially two-dimensional (x-y) plane 22 located at z = 0, best seen in figure 2(a). The B0 field is preferably highly inhomogeneous outside the z = 0 plane, so that an IvfRJ image is only selectively acquired for the slice 22 at z = 0. This B0 field is preferably achieved using concentric rings of either permanent magnets or electromagnetic current-carrying coils or combinations thereof as will be described. The finite slice thickness at z = 0 effectively defines the spatial resolution of the imaging apparatus in the z-direction. If no, or insufficient, inhomogeneity outside the z = 0 plane is achieved by the ring magnet 20, then an additional magnet may be provided to create such inhomogeneity. In the preferred apparatus, a Helmholtz coil arrangement whose two coils are located on the two faces of the thin ring magnet 20 creates a constant z-gradient that allows control of the z-slice thickness by adjustment of the magnitude of a G gradient (i.e. dBIdz) as will be described in connection with figure 19.

As shown schematically in figure 3, a three-dimensional image can then be built up by mechanically moving the whole scanner assembly of a fixed B0 magnet 20, a gradient magnet 31 and an RF coil 32 over the sample (e.g. 23 as shown in figure 2) in the z-direction and acquiring two-dimensional images at different values of z. Of course, the same effect could also be achieved by translating the sample along the z-direction through a stationary scanner assembly.

In a preferred embodiment, it is not even necessary that the magnetic field B0 is independent of x and y in the z = 0 plane. Instead, the B0 field is radially in.homogeneous, in the sense that it increases (or decreases) monotonically as a function of radial distance r from the centre (r = 0, on the z-axis) outwards at any given azimuthal angle, where r is the distance (2 + y2). Such a radially inhomogeneous field permits contour slice selection 24 with a suitable selective RF pulse as shown figure 2. It is not even necessary that the field in the z = 0 plane is angularly homogeneous, in the sense that it is independent of the azimuthal angle theta, defmed as tan'(y/x). In other words, the "contours" 24 that are slice-selected could be distorted out of the ideal circular shape because of an azimuthal dependence in the z = 0 plane.

The mechanical slice-selection along the z-axis has the advantage that the operator has full control over the number of z slices and their location. The operator can choose to image only a single slice of the sample merely by sliding the z 0 plane of the scanner into the chosen z-position of sample 23. All mechanical operations are preferably driven pneumatically, or with stepper motors or by any equivalent means using technology widely used in robotics.

An examination of figure 2 shows that the imaging protocol is akin to three-dimensional line-imaging in the sense that there are two levels of orthogonal slice selection followed by a one-dimensional image projection using a read-out gradient The first slice selection is along z and is achieved by B0 field inhomogeneity along the z-axis and slice-selective radiofrequency pulses. The second level of slice selection is the creation of concentric (possibly distorted) contours of transverse magnetization by frequency-selection in the radial (and r possibly azimuthal) inhomogeneity of the B0 field within the z = 0 plane. As we shall see, this protocol also relaxes, to some degree, the requirement for highly linear transverse magnetic field gradients (G and G) and a highly homogeneous radiofrequency B1 field because non-linearities and inhomogeneities in these fields.

can in principle be compensated by field mapping and correcting the image at the data processing stage.

Preferably the magnet 20, RF coil 32 and transverse gradient coils 31 are as thin as possible along the z-direction so as to permit easy mechanical translation of the whole assembly slice-by slice over a sample 23 (such as a human or animal head, arm, leg, hand or a non-human workpiece to be examined).

The imaging protocol Figure 4 shows a preferred imaging pulse sequence 40, which is a constant-field variant of STEAM (Stimulated Echo Acquisition Mode) imaging. A stimulated echo is created by a train of three 90 pulses 41a, 41b, 41c. These pulses are slice selective. They select the z = 0 plane because of B0 field inhomogeneity and because a gradient G (i.e. dBz/dz) is created with a pair of Helmholtz coils 31, to be described later. They also select a (possibly distorted) angular contour within the z = 0 plane because of the radial (and possible azimuthal) inhomogeneity of the

main magnetic field B0.

For this to be effective, the B0 field strength needs to increase monotonically from the centre (r = 0) outwards, which is made possible by the circular ring geometry of the B0 magnets (permanent or electromagnetic) which maintains constant field 43 for the selected slice. The stimulated echo 42 is encoded in the frequency domain with a transverse read-out gradient 44 in G where p is, in the simplest case, either x or y. Fourier transformation of this echo, followed by a magnitude operation creates a one-dimensional projection of the angular contour in the direction of the transverse gradient. To distinguish points within the contour it is* necessary to acquire a minimum of two projections, e.g. one projection with a G gradient and a second projection with a G gradient. Additional projections at

I

intermediate orientations are optional but may be desirable to remove ambiguities in the image processing stage.

To avoid possible problems with induced eddy currents created by rapid switching of the transverse gradients, the preferred acquisition sequence uses constant, non- switched transverse gradients. However this requires that the 900 pulses are non-selective with respect to these transverse gradients. The transverse gradient fields 44 are therefore weaker than the radial field gradient 43 from the B0 magnet 20 and

the G (z = 0) field gradient 45.

Because both the B0 and the RF field B1 project out of the z 0 plane there is the possibility that magnetic resonance signal (stimulated echo intensity) is excited in regions of the sample located outside the z = 0 plane. To remove this out-of-plane contribution, the G1 gradient field 45 created by the Helmholtz coils 31 is slowly switched off in the time between the last slice-selective 90 pulse (corresponding to the Nth annular contour slice) and the second 90 degree pulse of the first annular contour slice. Switching off the Hehmholtz G gradient in this way makes it act as a "spoiler gradient" for out-of-plane magnetization because it changes the frequency of all transverse magnetization outside the z = 0 plane and therefore removes its contribution to the stimulated echoes. In this way, only stimulated echoes from magnetization in the z = 0 plane are detected. This, of course, requires that the G gradient 45 is positioned so that the zero field plane of G coincides with the z = 0 plane of the magnet 20 and RF coil 32. Note however, that the G gradient does not need to be linear outside the z = 0 plane, nor highly linear inside the z 0 plane.

This protocol permits rapid imaging because it is not necessary to wait for longitudinal magnetization relaxation between ring contours. Instead the G gradient is turned on and successive contours are excited in a single-shot by changing the frequency of the soft RF pulses in the protocol used in STEAM imaging. This creates a set of projections for concentric contours in the G gradient direction. The G gradient is then turned off and the G gradient is turned on and, after a time 5T1 for relaxation of the longitudinal magnetization, the sequence is repeated. This provides G and G projections for all the concentric contour slices in the z = 0 plane which, after suitable processing (to be discussed later), allows a two-dimensional image in the z = 0 plane to be constructed. The scanner is moved to the next z-position and the whole procedure repeated. Note that the projections are not limited to the x and y directions in any z plane but can be made at any angle to the x axis in the z plane. The preferred image processing protocol uses at least four directions at 0, 45, 90 and 135 degrees to the x axis for unambiguous image reconstruction.

Thus, in a general sense, the pulse sequence selects a first transverse direction with a transverse gradient in a first direction (e.g. the x-direction) and then, by rapidly switching the frequency of the RF pulses in the pulse sequence, it excites magnetisation (and signal in the form of echoes) from all the ring contours in a single shot. This is represented by the loop with suffix N (for N contours) in the pulse sequence of figure 4. The transverse gradient direction is then changed (e.g. to the y-direction) and the whole single-shot sequence is repeated. It can therefore be very fast.

It is noteworthy that because projections can be acquired sequentially from inner to outer rings it is possible to adjust the amplitude of the RF pulses used to excite each contour to ensure that the tip angle is 90 . The duration of each RF pulse can also be adjusted to vary the slice thickness of each contour individually and to ensure a 90 excitation. This relaxes the need for B1 field homogeneity over the whole imaging area in the z = 0 plane. Moreover, B1 homogeneity outside the z = 0 plane is also no longer needed.

Image reconstruction Figure 5 illustrates the image reconstruction concept. To extract a two-dimensional image from the set of x and y projections for each concentric contour in the z = 0 plane requires that the main B0 field is mapped, i.e. that each point in the z = 0 plane inside the imaging area is associated with a known water proton resonance frequency. In addition the frequency shifts induced by the G and G, gradients in the z = 0 plane (which may be slightly non-linear) also need to be mapped. Knowing the B0 field map and the frequency of the slice-selective 900 jj pulses and the excitation bandwidth it is in principle possible to draw a contour 51, 52 of each concentric slice in the x-y plane corresponding to z = 0. Knowing the contour and the field map of the G and G gradients it is then possible to calculate the shape of the x and y projection profiles 53, 54 from each concentric contour for a uniform sample. These predictions can be compared with experiment if the sample comprises a homogeneous material such as water or a gel. It is therefore apparent that there is a one-to-one correspondence between the projection profiles and the distribution of magnetization around each contour. With a suitable numerical algorithm it is, in principle, possible to work backwards from the experimental x-and y-projections for each concentric contour to deduce the distribution of intensity around each contour. Once this has been done for one contour it can be repeated for all the concentric contours and, with interpolation, a two-dimensional image can be reconstructed from the sets of profiles. Appendix 1 shows how this can be done with a maximum entropy algorithm.

Magnet design To implement this system, a non-switched magnetic field B0 is required in the z-direction in a thin slice in the z = 0 plane. This can be achieved with a permanent magnet design as shown in figure 6.

The embodiment of figure 6 uses concentric rings of permanent magnets 61, 62 as illustrated in figure 6. In this example, the concentric magnet rings 61, 62 are composed of a permanent magnet material such as neodymium-iron-boron and polarized in the z-direction (orthogonal to the plane of the drawing). The inner ring magnet 62 has the opposite polarization to the outer ring magnet 61 and is smaller. its function is to reduce the radial inhomogeneity of the Bo field in the central hole 63 to values, appropriate to imaging, e.g. of the order of 1 Gauss/cm.

Figure 7 shows a cross-section through the magnet rings 61, 62, together with the calculated field map showing the calculated B0 field inhomogeneity. Figure 8 shows the calculated radial dependence of the B field, with the vertical axis showing B field strength and the horizontal axis showing radial distance r from the centre of the ring magnet. The following points are noteworthy.

Unlike conventional NIvIRJMRI magnets, the magnet 20 produces a B0 field directed along the z-axis that is deliberately inhomogeneous in the radial direction (i.e. as a function of x and/or y) over a thin slice centred in the z = 0 plane. The Bo field does not need to be highly homogeneous in the angular (azimuthal) orientation. The B0 field is inhomogeneous outside the z = 0 plane so as to facilitate z 0 plane slice selection. In all these requirements, the field therefore differs from conventional magnet arrangements which aim to create a homogeneous field over an extended volume of space or within a plane or measurement surface.

In the preferred embodiment, the B0 field is created by two concentric rings of permanent magnets 61, 62 polarized in the z-direction and having opposite polarization, as illustrated in figures 6 and 7. The inner ring 62 is smaller and carefully designed to reduce the large radial inhomogeneity (dB/dr) created by the larger outer magnet ring. The preferred design creates a maximum radial gradient (dB/dr) of the order of 1 Gauss/cm which is of the same order of magnitude as conventional imaging gradients.

To make fine adjustments to the r-dependence of the radial field gradient (dB/dr), a "radial shim coil" 64 may be incorporated between the inner and outer permanent magnet rings 62, 61. The radial shim coil is preferably a ring shape and disposed coaxially with the B0 magnet 20. The radial shim coil 64 may comprise loops of copper conducting wire carrying a constant but adjustable current. The shim coil may be cooled by blowing air over the coil.

Apart from the optional radial shim coil, the apparatus does not require a shim coil set In particular there need be no azimuthal shims. Sufficient azimuthal field homogeneity can be ensured by creating the magnet rings from three overlapping layers of magnets. Each magnet layer can be made by gluing magnet segments together to make a circular ring structure. f

In other embodiments, the permanent magnet rings 61, 62 can be replaced with a number of resistive electromagnetic coils. An example is shown in figure 9 which uses three concentric resistive current-carrying loops orcoils 91, 92, 93. The current direction in the rings does not need to be the same. In the illustrated three-ring system the current in the innermost coil 93 is reversed relative to the other two coils 91, 92. For the avoidance of doubt, throughout the present disclosure, the expression coil' is intended to encompass any electrical path structure that generates an electromagnetic field by passage of electrical current along the path whether that path comprises one or more adjacent series wound or electrically parallel current loops. A suitable current driver is provided (not shown) to drive an appropriate level and direction of current in each of the loops or coils 91, 92, 93.

In the present specification, unless the immediate context dictates otherwise, general references using the expression magnet' encompass both permanent magnet arrangements and electromagnet arrangements or combinations thereof.

As will be appreciated by those skilled in the field, resistive electromagnetic coils require significant current input and therefore water-cooling of the coils. Further, copper coils may be heavier than permanent magnet rings. A further alternative is the use of superconducting coils cooled in liquid helium (or high temperature superconducting coils cooled in other cryogenic liquids). This is more expensive and requires liquid helium andlor liquid nitrogen cooling but does not require a power supply. In the superconducting embodiment, much higher field strengths are utilizable, thereby increasing the signal/noise ratio.

Figure 10 shows the radial dependence of the B0 field calculated for the three-coil arrangement of figure 9, with current directions, amplitudes and coil geometries optimized to minimize the radial gradient. The graph illustrates radial inhomogeneity in the B0 field for the three-coil configuration when the currents are optimised for greatest radial homogeneity.

Higher (or lower) numbers of coils could be used. Figure 11 illustrates a design based on five concentric current carrying loops 101 -105 which could be either resistive or superconducting. The current in the innermost loop 105 is reversed r relative to the other loops 101 -104. This design and the current directions and amplitudes have been designed to minimise the quadratic (d2B/dt2) and quartic (d4B/dz4) gradients at the centre. Figure 12 shows the radial dependence of the field. It will be appreciated that alternative concentric ring geometries based on one or more rings could be used.

An alternative magnet design uses only a single magnet ring instead of two concentric rings. This has the advantage that the magnet ring can be made smaller in radius and lighter in weight and the field strength in the centre can be made much stronger because the opposing polarizations in the two-magnet ring design no longer exist. However the radial field inhomogeneity for a single magnet ring is now much greater so that even a "hard" microsecond radiofrequency pulse would be slice selective. In this sense the protocol would now resemble stray field imaging (STRAFI), but a similar contour selection imaging protocol should still, in principle, be possible.

The radiofrequency coil To minimize mutual inductance between the RF coil 32 and the surrounding transverse gradient coils 31 it is desirable that the RF coil is self-shielding.

Accordingly the invention incorporates an actively self-shielded short saddle coil whose design is based on known designs and which is illustrated in figure 13.

Figure 13(a) is a cross-section through the z-axis and indicates the radial (x and y) positions of the axial portions 130 -133 of the saddle coil 135 that extend parallel to the z-axis. The saddle coil 135 comprises two layers of axial portions of the coil loops extending along the inner and outer surface of a cylindrical former (respectively axial portions 130, 131 and 132, 133). The length of the saddle coil along the z-direction is minimized to increase the filling factor, i.e. to increase the fraction of the volume inside the RF coil occupied by the sample transverse magnetization. In this case the relevant sample volume is that of the thin slice in the z = 0 plane. In other respects, the geometry of the saddle coil accords with standard practice.

Figure 13(b) shows the calculated direction and strength of the B1 field for the coil of figure 13(a). Figure 13(c) shows the calculated field strength of the B1 field in grey scale for the coil of figure 13(a). The self-shielded aspect also has the advantage that, by the reciprocity theorem, the coil should be less sensitive to external noise sources, so that it should not be necessary to restrict the scanner to within a specially constructed RF-shielded room.

Other types of transverse RF coil arrangements known in the art can be used.

Tuning the RFcoil It is apparent from figure 14 that, whichever RF coil design is used, some retuning of the RF coil may be required. The Q-curve for the RF coil has the idealised form F(cD) = (lhr)(coo/2Q)/[(U)o/2Q)2 + (U) -(00)2] where 0 is the resonance frequency. The horizontal line in figure 14 shows that at 10% deviation from the minimum point on the Q curve, the frequency off-set is given as (o' -COo) = (1/3)(w()/2Q) This corresponds to a gradient Gma.,net(max) = (w' -coo)IFOVR where FOVR is the radius of the circular field of view.

Assuming the resonance frequency o = 2.116 MHz, Q = 30 and FOVR 4.25 cm, this gives a maximum magnet gradient of 0.65 G/cm before retuning is required.

However the radial gradient is of the order of 1 G/cm or more so it is probable that RF probe retuning will be required.

The transverse gradient coil design To create a linear transverse field gradient (i.e. dB/dx = dB/dy, or more generally dB/dr, or still more generally dB/du where u is any axis within the z = 0 plane not r necessarily passing through x = y = 0) in the z 0 plane, a unique design is used based on opposing arcuate coil loops. This is shown in figures 15 to 18.

The partial coil loops 200 as shown in figure 15 form part of a magnet that comprises a pair of opposing current loops 201, 202 each including an arcuate portion 203, 204 extending around a part of the circumference of the sampling volume in cavity 11 defined by the B0 magnet. A diagram showing the complete loop formation for one of these arcuate portions 203 is shown in figure 16. Current is directed to and from the arcuate portion 203 along radially extending current paths 210,211 and leads 212, 213.

This design of gradient coil creates a uniform, near-linear transverse gradient over a slice at z = 0 rather than a uniform linear gradient over an extended distance in the z-direction as in conventional gradient coil designs. For this reason it has a length in the z-direction roughly equal to that of the B0 magnet rings.

Because the gradient coil of figure 15 uses only opposing arcuate portions (which may be quadrants or near-quadrants), it is straightforward to have two sets of coils in the same circle, to produce a transverse gradient in any direction by adjusting the magnitude of the current through the two sets of coil loops 201, 202.

Alternatively, the circular arrangement of opposing quadrants makes it easy to reorient the gradient direction by mechanically rotating the two sets of coils 201, 202 around the sampling volume. In this way a non-switched, constant current supply can be used, obviating the need for expensive gradient controllers.

Because the preferred image acquisition protocol uses non-switched fields during the stimulated echo pulse sequence, the NMR signal is not perturbed by induced eddy currents. For this reason, unshielded transverse gradient coils can be used, greatly simp1if'ing the design and construction of the coils and permitting stronger gradients for a given current-turn density. r

Although a linear gradient over the whole imaging area in the z = 0 plane is desirable, it is not necessary in the apparatus described here. This also relaxes the design constraints.

Figure 17 shows the magnetic field component B as a function of radial distance on the y-axis, as calculated for the G) gradient coil arrangement of figure 15.

Figure 18 shows an alternative arrangement of gradient coil assembly viewed along the z-axis using four outer arcuate loop portions 231 -234 each extending around a part of the circumference of the sampling volume, and four inner arcuate loop portions 235 -238 each extending around a part of the circumference of the sampling volume but radially inward of the outer arcuate loop portions. As indicated earlier, each arcuate portion may extend around a quadrant, i.e. 90 of the sampling volume. However, in the arrangement of figure 18, a departure from exact 90 segments is possible, the quadrants' being alternately 80 and 100 with the inner and outer loop portions being effectively rotated relative to one another by 90 .

With the arrangement of figure 18, it is possible to strengthen the gradients using the two concentric layers 231 -234 and 235 -238 respectively. Although the optimum gradient linearity is obtained with overlapping 100 and 80 arcuate portions, other combinations are possible.

Slice thickness in the z = 0 plane With reference to figure 19, a pair of Helmholtz coils 241, 242 is provided respectively on the front and back faces 243, 244 of the ring magnet 20. These coils 241, 242 carry current in opposite directions thereby creating a G gradient with a null field in the z = 0 plane. This serves two purposes. Firstly, it controls the slice thickness in the z-direction. Secondly it permits the removal of signal from outside the z = 0 plane. This is achieved by switching the Helmholtz coils off after the first two 90 pulses 41a, 41b (figure 4) in the stimulated echo sequence 40. Figure 20 shows the B field strength gradient along the z-axis calculated for the coil geometry of figure 19. Figure 21 shows an expanded view of the gradient ( of figure 20 showing that it is nearly linear over a distance of a centimetre either side of the z 0 plane.

Pulsed pre-polarisation The maximum field created at x = 0, y = 0, z = 0 by the two concentric permanent magnet rings 61, 62 using the design specifications resulting in the B2 field of figure 8 and neodymium magnets is of the order of 500 Gauss (or 0.05 Tesla).

This is a very low field so signal / noise ratio after two degrees of slice selection could be problematic. Pulsed pre-polarisation is one possible way of overcoming this problem, should it arise. Alternative strategies are to use superconducting magnets or a single permanent magnet ring with a higher radial gradient.

The principle of "pulsed pre-polarisation" involves transiently raising the main magnetic field for a few seconds to superpolarise the sample under analysis at a higher field before image acquisition in the lower field of the non-pulsed magnet.

This is achieved by transient passage of a high current through a copper coil wrapped around the outside of the outer permanent magnet ring 61 in such a way that it is rapidly (e.g. in a few milliseconds) switched off before the NMR pulse sequence is started. Operating fields (without pulsed pre-polarisation) are designed to be between 3 and 5 tvfHz. The control system for pulsed pre-polarisation has already been published and tested, but does not appear to have been implemented

in any low-field imaging system.

Suitable drive mechanisms for the gradient and RF coils (and B0 coils, if used) are within the reach of the person of ordinary skill using existing technologies.

Suitable analysis and control systems for implementing the pulse sequences and detecting magnetic resonance data therefrom are also within the reach of the person of ordinary skill using existing technologies.

Other embodiments are intentionally within the scope of the accompanying claims.

APPENDIX 1. A Maximum Entropy procedure for image reconstruction using the proposed scanner.

The protocol will be illustrated for the case of two projections using a transverse gradient oriented along the x and y directions (0 and 900 respectively). To avoid reconstruction artifacts it is necessary to use at least four projections along 0, 45, and 1350, but the principle of the MIEM (Maximum Entropy Method) image reconstruction is the same.

The image space for a particular contour-slice in the z=0 plane is taken to be an N by M array of intensities f (i = ito M column number = x-coordinate; j = ito N = row number = y-coordinate).

1 2 3.... M 2 1) 3 intensity

N

The total intensity obtained by summing down the columns is the x-projection, denoted by the vector X, i = 1. . . M. The total intensity obtained by summing across the rows is the y-projection, denoted by the vector Y, j = 1. . .N.

The total intensity over all pixels is denoted F = E f1 The contribution made by the intensity level f1 at the (i,j)th pixel to the projected value X1 denoted by a weighting factor x.

The contribution made by the intensity level at the (i,j)th pixel to the projected value Y denoted by a weighting factor Yji The arrays and yj will typically be equal and comprise a pattern of 0's and l's.

The l's will typically be in a (distorted) contour, with a background of 0's.

The problem, to be solved by a maximum entropy approach is to fmd the array of intensities f1 which maximizes the entropy of the image: S = -E E1 fjLnfj Subject to the M+N+1 linear constraints on the intensities, Zfj1x1=X1fori= 1...M EfjY = Y forj 1. . .N This problem can be solved by the method of Lagrange multipliers. It can then be shown that the solution is in the form of an algorithm comprising 4 steps: Step a) Estimate the total intensity F. Usually we will be able to assume that the total intensities in the x-and y-projections are both approximately equal to F: ZX1=EY3=F Noisy data will spoil this, but to a reasonable first approximation we can estimate F by averaging the two sums.

Step b) Defme the function 0 such that G=FLnZ+E1a1X +E3Y where Z = LZ exp(-aixji -13y) Step c) Minimise G by varying the M+N independent variables (trial parameters) a and fj. At the minimum of (3 defme the optimum Z with the optimized set of parameters a and 3y denote this Zopt Step d) Evaluate the final intensity matrix (the image) f1 as = (F/Z0).exp(-axj -f3iyJj) with the optimized values of a1 and This completes the required solution.

This protocol is performed for each contour in the z = 0 plane and the two-dimensional image in the z = 0 plane is constructed by addition or interpolation of these contour images.

Delerinining the weighting factor arrays x1 andy, The protocol requires that we already know the weighting factors x and Yji. These can be determined for each contour-slice selection with the following steps: Step 1. Frequency map B0 using a probe coil in the x-y plane at z 0.

Step 2. Knowing the excitation bandwidth and the resonance frequency of the contour-slice-selection radiofrequency pulses, construct an iso-frequency contour map in the x-y plane. This will be the shape of the selected contour and will enable crude Xji and Yji matrices to be constructed.

Step 3. The crude x and Yji matrices can be refined by imaging a homogeneous sample, such as a uniform gel (or water) for which all the = a constant. The maximum entropy procedure can then be applied to the water projections in the x-and y directions to deduce the optimum values of the x, and yp matrices. r

Claims (30)

1. A magnetic resonance scanning apparatus comprising: a first magnet configured to generate within a sampling volume defmed within an x-y plane, a B0 magnetic field in the z-direction the B0 field having radial inhomogeneity in the x-y plane and axial inhomogeneity outside the x-y plane; a second magnet configured to generate, within the sampling volume, a gradient field having a field in the z-direction and gradient dB/du where u is an axis in the x-y plane; and an RF coil configured for generating a B1 magnetic field transverse to the

B0 field thin the sampling volume.

2. The apparatus of claim I in which the first magnet comprises a ring shape occupying the x-y plane.

3. The apparatus of claim 2 in which the first magnet comprises: a first permanent ring magnet extending around the ring axis and polarised in a first direction parallel to the ring axis, and a second permanent ring magnet within and axially aligned with the first permanent ring magnet, the polarity of the second permanent ring magnet being opposite to that of the first permanent ring magnet.

4. The apparatus of claim 3 in which the first magnet includes a plurality of permanent ring magnets concentrically disposed and each polarised in the first direction.

5. The apparatus of claim 3 further including a radial shim coil disposed coaxial with and between the first permanent ring magnet and the second permanent ring magnet.

S

6. The apparatus of claim 2 in which the first magnet comprises a plurality of concentric current carrying loops and current control means for driving current flow in at least one loop in an opposite direction to current flow in at least one other loop.

7. The apparatus of claim 6 including at least three concentric current carrying loops, in which the current control means is adapted to supply independent current to each of the loops.

8. The apparatus of claim 7 in which the current control means is configured to drive current through the innermost loop in a direction opposite to all other loops.

9. The apparatus of claim 1 in which the RF coil comprises a saddle coil.

10. The apparatus of claim I in which the RF coil is configured to produce a substantially uniform B1 field as a function of radial distance from the z-axis within the sampling volume within and in the plane of the ring of the first magnet.

11. The apparatus of claim I in which the second magnet is physically rotatable about the z-axis.

12. The apparatus of claim 1 in which the second magnet has a thickness in the z-direction not greater than that of the first magnet and is configured to generate a

substantially linear gradient field dB/du.

13. The apparatus of claim I in which the second magnet comprises a pair of opposing current loops each including an arcuate portion extending around a part of the circumference of the sampling volume within the first magnet.

14. The apparatus of claim 13 in which the arcuate portions of the opposing current loops each extend around a quadrant of the sampling volume within the first magnet.

15. The apparatus of claim 13 in which the arcuate portions of the opposing current loops are equal in length and on opposite parts of the circumference of the sampling volume within the first magnet.

16. The apparatus of claim 13 in which the second magnet comprises a second pair of said opposing current loops each including an arcuate portion extending around a different part of the circumference of the sampling volume.

17. The apparatus of claim 16 in which the second magnet comprises one or more further pairs of said opposing current loops each including an arcuate portion disposed concentrically within the arcuate portions of said pairs of opposing current loops.

18. The apparatus of claim 17 in which the arcuate portions together define two concentric circles with overlapping arcuate portions.

19. The apparatus of claim 1 further including a gradient coil configured to provide a G gradient with a null field in the sampling volume.

20. The apparatus of claim 19 in which the gradient coil comprises two Helniholtz coils disposed on either face of the first magnet and coaxial therewith.

21. A method of gathering magnetic resonance data comprising the steps of: generating, within a sampling volume occupying an x-y plane, a B0 magnetic field in the z-direction, the B0 field having radial inhomogeneity such that the B0 field varies monotonically as a function of radial distance within the x-y plane from a central z-axis of the sampling volume for all azimuth angles about the z-axis, and axial inhomogeneity outside the plane of the sampling volume; generating, within the sampling volume, a gradient field having a field in the z-direction and gradient dB/du where u is an axis in the x-y plane; generating an RF B1 magnetic field transverse to the B0 field within the sampling volume; and obtaining magnetic resonance data in respect of a contour slice within the sampling volume, the contour slice being a circumferential portion of the sample extending around the z-axis and in the x-y plane of the sampling volume.

22 The method of claim 21 further including the step of generating successive RF B1 fields transverse to the B0 field within the sampling volume and at different frequencies thereby obtaining magnetic resonance data in respect of respective successive contour slices within the sampling volume.

23. The method of claim 22 further including the step of rotating the gradient field transverse to the B0 field and subsequently obtaining second magnetic resonance data in respect of a contour slice with the rotated gradient field.

24. The method of claim 23 further including the step of generating successive RF B1 fields transverse to the B0 field within the sampling volume and at different frequencies thereby obtaining magnetic resonance data in respect of respective successive contour slices within the sampling volume with the rotated gradient

field.

25. The method of claim 23 or claim 24 in which the rotated gradient field is at

degrees to the initial gradient field.

26. The method of claim 21 in which the step of generating an RF B1 field comprises generating a sequence of pulses.

27. The method of claim 26 further including the step of generating a gradient field dB/dz during at least one or more of the pulses and reducing the gradient field dB/dz during at least one other of the pulses.

28. The method of claim 21 in which the circumferential portion of the sample extending around the z-axis and in the x-y plane of the sampling volume is non-circular.

29. The method of claim 21 further comprising obtaining magnetic resonance data for at least two different azimuthal angles of gradient field dB/dr for each of a succession of contour slices and computing the projection profile for each one of the different azimuthal angles for each one of the contour slices.

30. The method of claim 29 in which the magnetic resonance data obtained for at least two different azimuthal angles comprises data for at least gradient fields dB2/dx and dB2/dy.