WO1990012329A1

WO1990012329A1 - Region selection in nuclear magnetic resonance inspection

Info

Publication number: WO1990012329A1
Application number: PCT/GB1990/000494
Authority: WO
Inventors: Alexander James De Crespigny; Thomas Adrian Carpenter; Laurence David Hall
Original assignee: Alexander James De Crespigny; Thomas Adrian Carpenter; Laurence David Hall
Priority date: 1989-04-03
Filing date: 1990-04-03
Publication date: 1990-10-18
Also published as: GB8907493D0; JPH05500169A; EP0466756A1; AU5346590A

Abstract

A method and apparatus is described for selecting a region of interest within an object whose nuclear magnetic resonance is to be investigated. Selection is achieved by locating the object within a zone within a generally uniform intense magnetic field, adding two or more magnetic fields so as to introduce a magnetic field gradient across the zone so that different regions of an object located within the zone are influenced by different magnetic field intensities so as to possess differing magnetic resonance frequencies. The magnetic fields are typically produced by electric currents in appropriately positioned coils and by varying the relative values of the currents so the resulting magnetic field gradient can be positioned and rotated of otherwise displaced within the zone. By following each incremental repositioning of the magnetic field gradient with an RF signal having a broad spectrum of finite energy frequency components and a defined band of zero of very low energy frequency components, so a region of the material within the zone can be affected so that the nuclea spin orientations in that region are aligned whilst those outside the region are random and by stepwise rotating the gradient direction and applying RF signals at each step so a region is defined with its axis coinciding with the axis of rotation of the magnetic field gradient.

Description

Title: Region selection in nuclear magnetic resonance inspection

Field of invention

This invention concerns nuclear magnetic resonance and is particularly concerned with method and apparatus for selecting regions of interest from a volume which is considerably greater in size than the region of interest.

Background to the invention

As the use of volume selective nuclear magnetic resonance has become more widespread, both for "high resolution spectroscopy" and for "zoom-imaging", so techniques for extracting signal from a reduced volume of spins within an extended sample have proliferated. The most versatile and generally applicable are those involving magnetic field B gradients in conjunction with selective Rf pulses and they fall broadly into two groups: techniques which achieve localisation by selectively exciting spins inside the volume of interest and those which saturate all the magnetisation except that within the volume of interest.

From the former group the STEAM (1) sequence, involving stimulated echoes, is probably the most easily implemented. As with all echo techniques, the results are

T weighted, which rules out its application to short -, species such as 31-P metabolites, but proves favourable for

1 water and lipid suppression in H metabolite detection. DRESS (2) achieves good sensitivity with surface coils but also suffers from ₂ limitations, while the OSIRI (3) technique is useful for observing fast relaxing magnetisation although its suffers from pulse bandwidth limitations and unlike STEAM, it is not feasible to shim on the selected volume. New techniques have recently been developed (4,5) which achieve 2D selective excitation in the single radiofrequency/gradient pulse and are much less limited by 2 relaxation. However they also suffer from problems such as poor outer volume suppression, difficulty in moving the selected volume under computer control and off resonance effects.

The present invention has as its object a new technique for selective saturation, involving noise modulated RF pulses, which overcomes many of the problems outlined above while producing a well defined, non-rectilinear, volume of interest which can be moved anywhere within the sample and which is variable over a wide range of shapes and sizes.

Random noise generation algorithms are now available for many computer systems,- and it is easy to generate a frequency response function which is random except for a small region of zero amplitude (see Figure 7a). Fourier transforma ion of this yields a pseudσ-random modulation function (see Figure 7b), which may be used to modulate an RF signal so as to produce a modulation envelope similar to Figure 7(b) (see Reference 10). If an Rf pulse having such an envelope is applied to a homogenous sample located in a magnetic field gradient, (which alters the magnetic resonance frequency of the material forming the sample profile in the direction of the gradient) the nuclear spins will be rotated through random angles about random axes except for nucieii within any region along the gradient at which the nuclear magnetic resonance frequency of the material has no corresponding energy component in the RF pulse. Noise pulses have been used previously in volume selection, in OSIRIS, where they are used to reduce the subtraction errors in the ISIS experiment (see Reference 11). In that case however the same noise pulse is used, in order to give the same frequency response, for each stage in the subtraction cycle.

A single noise pulse cannot achieve complete saturation of the required magnetisation, for, whilst an ideal noise pulse may distribute the magnetisation vectors evenly over a sphere, (with random phase and flip-angle over the whole bandwidth) it will do so reproducibly; the pseudo-random pattern of magnetisation produced relates directly to the particular noise pulse used. This is why such pulses work with the ISIS subtraction experiment (see Reference 11).

It has been proposed that a small volume within a large object be defined by using a small radio frequency coil to limit the volume of the object which is influenced by the radiofrequency field. However, the radiofrequency field is not always homogeneous, nor is it always closely matched in shape to the region of interst.

It is an object of the present invention to provide a slicing (or selection) technique with no limitation on the range of shapes generated which define the volume of interest.

The invention also has as an object a new technique for selective saturation, involving noise modulated RF pulses, which overcomes many of the problems outlined above while producing a well defined volume of interest which can be moved anywhere within the sample and which is variable over a wide range of shapes and sizes.

Summary of the invention

According to one aspect of the invention there is provided a method of selecting a region of interest within an object whose nuclear magnetic resonance is to be investigated in which the object is located within a zone within a generally uniform intense magnetic field, which zone is influenced not only by the intense field but also by two or more local magnetic fields, so that within the zone a magnetic field gradient exists so that different regions of the object are influenced by different magnetic field intensities to therefore possess differing magnetic resonance frequencies characterised in that the local magnetic fields are varied so that the direction of the magnetic field gradient across the zone is incrementally rotated through a number of discrete positions and an RF signal having a broad spectrum of finite energy frequency components and a defined band of zero or very low energy frequency components, is applied after each incremental step, so as* to cause random nuclear spin orientations of nucleii in the object material located in those parts of the zone for which the RF signal contains energy at the appropriate frequency, but no changes in the orientation of the nuclear spins of nucleii located in those parts of the object material for which the RF signal does not contain energy at the appropriate frequency, so that as the field gradient rotates a volume of material is defined (the selected region) in which no changes in the orientation of the nuclear spins occurs, whilst in general the nucleii in the remainder of the material will have random spin orientations.

As the number of gradient steps is increased, the region of zero saturation becomes more clearly defined.

It is an advantage of the invention that each RF signal is only required to achieve partial saturation of the magnetisation so that much less RF power is required amp-1-i-fier than if saturation is required to be obtained from one burst of RF signal. This is an important factor in clinical applications of the invention where RF power must usually be limited. In addition, since the method does not depend on pulse flip-angles, a surface coil may be employed to give increased sensitivity.

The invention thus provides a method oϋ single shot localisation by selective saturation, using a pre-pulse which may be placed at the start of any pulse sequence. This method uses "rotating" field gradients and noise modulated RF pulses, the latter with a low power and RF field homogeneity requirement which should allow its use with surface cells. The method appears to be applicable to work on both protons and phosphorus.

Movement of the selected volume by altering the field gradient and/or the RF signal zero energy components may be under computer control and the fact that the non- rectilinear shape is variable over a wide range, generally allows much closer tailoring of the selected region to the region of interest in the sample.

The invention can be further understood by considering two tubes of water. If the magnetic field (B ) is homogeneous over the entire sample, then all the water protons will resonate at the same frequency. Spatial information can be encoded by applying a linear variation of magnetic field across the sample (a field gradient), so that

* protons in different regions of the mass of water resonate at different frequencies.

For imaging an RF signal is applied having finite energy components across the entire spread of frequency variation.

"Slice-selection" in accordance with the invention is achieved by using a frequency-selective RF signal (ie one in which some of the frequency components in the Rf signal are at zero (or very low) energy so that protons having those resonant frequencies are not energised and their spin axes remain aligned). Such a signal will only influence part of the sample. The slice position can be varied by changing the band of frequencies within the RF signal. The slice thickness depends on the bandwidth of the zero (or very low) energy frequency components in the RF signal.

According to a preferred feature of the present invention, an RF carrier is modulated using noise pulses which are optimised by in every case having zero energy components over the same part of their frequency spectrum and a random selection of high energy frequency components outside that part of their spectrum.

The method isolates a volume of spins in 2 or 3 dimensions, within a larger sample, by selectively saturating spins outside the said volume. Selective saturation is relatively common for removing resonance lines in spectroscopy [J.Chem.Phys. vol59, number 4, p1775 1973] and has been used in an early imaging procedure [J.Phys C. vol17, L457 1974]. It has been used for volume selective spectroscopy such as in [J.Mag.Res. 70 319-326 1986] where saturation outside the volume of interest is achieved by creating coherent magnetisation and then dephasing it with gradients. This requires both good RF homogeneity and a high power amplifier. OSIRIS [J.Mag.Res 78 519-527 1988] uses noise pulses in volume localisaton but not for saturation and can only inspect rectilinear regions of interest. By contrast the present invention allows arbitrary regions to be defined by superimposing multiple projections.

Localisation methods which work by removing signal contributions from outside the volume of interest through saturation, are particularly attractive because of their

31 ^2 independence, an important factor for . P m-vivo work.

Such methods include; VSE(6),SPARS(7),SPACE(8) and DIGGER (9). All of these selectively create transverse magnetisation outside the volume of interest at some point in the pulse-sequence which is then dephased by the applied field gradients, while magnetisation inside the volume of interest is stored along the Z axis. A problem common to all these existing techniques is the high RF power level required to saturate all the spins in the outer volume. Other problems associated with pulse bandwidth and pulse imperfections, mean that spins near the edge of the sample are not fully saturated while some magnetisation is excited inside the volume of interest. Also, since these techniques all depend on accurate pulse flip angles, good B^ field homogeneity is required which rules out their implementation with surface coils.

Other techniques which aim to define arbitrary volumes are described in [7th SMRM 1988] [J.Mag.Res. 81 43-56 1989]. These operate by selective excitation of the desired region but suffer from T₂ relaxation effects in many species, unlike the present invention which is T₁ dependent.

A noise window pulse applied with a constant magnetic field gradient may produce points along the gradient axis, away from the central window, where magnetisation is hardly excited at all and others where it is flipped about 180 degrees, so that even multiple applications of this pulse will leave some Z magnetisation outside the selected volume. In order to select a plane of unperturbed spins, it is necessary to use many different noise pulses in succession, with the gradient held constant.

Alternatively, a "rod" of unperturbed spins can be selected by using the same noise pulse successively, but with the field gradient applied in a different direction each time. The cross section of the "rod" is dependant on the variation of the magnetic field (ie the gradient) which can be kept constant or varied between incremental repositioning thereof.

The invention is of particular merit in the following applications:

a) Volume selective spectroscopy with near arbitrary volumes in 2 and 3 dimensions;

b) Zoom imaging; c) Suppression of phase encode aliasing;

d) Saturation transfer experiments to measure flow, diffusion, perfusion etc;

e) Used in conjunction with surface coils, columns of spins parallel, oblique or perpendicular to the plane of the surface coil can be defined and sampled using rotating frame;

f) Reducing the dimensionality of spectroscopic imaging, and

g) Selectively saturating a volume within an extended sample.

This invention enables any arbitrarily shaped region of interest to be defined within an object. Thereafter, that volume can be studied by magnetic resonance (MR) in a variety of different ways, of which examples are specified below.

The invention can be added as a "module" ahead of, during, or after, any other MR sequence already in use. In essence the method selectively saturates the magnetisation of all the spins outside the volume of interest whose MR- signals are thereby suppressed, leaving that of the spins within the volume available for study. Saturation is achieved by use of optimised, pseudo-noise, modulated radiofrequency pulses applied in the presence of tailored gradient waveforms.

The volume of interest can be interrogated by any class of NMR spectroscopic measurement, whether one-dimensional, two-dimensional or three-dimensional.

The region defined by the volume of interest can be subjected to imaging by any known method.

There is a general problem associated with MR measurements of small regions within a large object. This arises when "zoom" imaging is used (a technique which gives higher resolution than if the same number of imaging increments were spread across the entire width of the object). The problem is that the MR-signals from the outer volumes "fold-over" and superimpose on the image of the inner volume (phase-encoded aliasing). This can be suppressed by using this invention to saturate the spins of the outer volumes.

MR can be used to measure quantitatively the movement of liquids - either flow, diffusion or perfusion. The invention can be used to saturate all spins except those of a small volume, and then to monitor the movement of those spins from that small volume. (This is known as saturation transfer).

The invention can be used to define and sample the magnetisation of the region of interest, by saturation of other regions. For example, a column of spins can be defined which is either perpendicular, parallel or oblique, with respect to the plane of a surface coil, and then studied using rotating frame zeugmatography, or gradient phase encoding.

The invention makes feasible a variety of measurements that would otherwise be either impossible, or very difficult. The following are given by way of example:

1. Normally the B -field is arranged to be homogeneous over the entire diameter of the object, although the homogeneity may fall off along the length. There are occasions when the choice of magnets means that this relationship cannot be achieved and the static magnetic field (B_Q) is inhomogeneous. This will mean that MR signals from the region of interest will be degraded by those from those portions of the object which experience the low-grade field.

Using the invention all signals can be saturated except those from a defined region of interest which can be "tuned" and positioned to coincide with either the appropriate region within the object, or within the magnet. Importantly, this relationship can be varied from one study to the next.

2. The defined region of interest can have any arbitrary shape defined by convex surfaces. (This is to be compared with most available methods which are restricted solely to rectilinear shapes).

3. With many conventional magnetic resonance imaging measurements image artifacts arise from the physical motion of remote parts of the object. For example in NMR inspection of the human body, images of the spine often show motion artifacts arising from the respiratory movement of the fat of the "tummy". The magnetisation and resultant "visible" movement of that fat can be suppressed by the present invention.

Thus, to summarise, the NMR technique, in accordance with the invention, is a method for defining a reduced region within an object under investigation so as to reduce or eliminate the signal from other areas. It involves applying a series of slice selective RF pulses in conjunction with a magnetic field gradient which changes direction after each pulse. Each of these pulses acts to randomise the spins in the sample except within a plane of well defined width and orientation (determined by the gradient direction) where spins are undisturbed. The region of coincidence of all these null slices defines a volume of intact magnetisation which may be interrogated without interference from spins outside this region all of which are saturated.

The advantage of the invention can be seen by considering just four of the known alternative techniques.

a) Selective excitation sequences (STEAM), uses 3 selective excitation pulses in a spin-echo sequence to excite a rectilinear region. The fact that it excites the region makes it sensitive to T₂ relaxation (which is not the case for the invention as it leaves the region unperturbed and is sensitive to T- relaxation) and limits its use to slow relaxing species. Rectilinear regions are not always ideal, particularly in-vivo, although conformal versions of the technique are being developed in conjunction. ith d). The invention can easily define conformal (non-rectilinear) regions.

b) 2D Selective excitation pulses. 2D K-space pulses can excite non-rectilinear regions without the need for a spin-echo. They still cannot be used for the shorter ₂ species and have some chemical shift sensitivity. The invention can define 3D as well as 2D regions and has a more useful chemical shift artefact (see below).

c) Selective saturation sequences (DIGGER), uses 3 pulses to selectively excite magnetisation by 90 degrees and then dephase it with field gradients. The pulses excite magnetisation across the bandwidth except for a central slice of zero excitation. Conceptually this is quite similar to the invention except that magnetisation is excited coherently and exact flip angle is important. It is also difficult to achieve good slice profiles.

d) Cycling methods such as (ISIS) use inversion pulses to invert orthogonal slices and an 8 step add/subtract cycle to produce a net signal only from the volume of interest (VOI). The multiple shot nature of this technique makes it impossible to shim with it, a big advantage of the invention. It does not use a spin echo and so is not T₁ or T₂ weighted and is much used for short T₂ measurements.

The ability to generate any chosen convex 2D or 3D volume is a feature which none of these techniques possess. In addition, the chemical shift artefact for the invention is more useful than in the above cases. For an off resonance spin, each slice is shifted an amount in the gradient direction. This results in a blurring of the region of interest (ROD for the other species, and for narrow bandwidth pulses could be eliminated altogether.

Possible variations on the selection sequence of the invention are as follows:-

1) The need to use a large number of RF pulses as required by the sequence of the invention could cause a problem with power dissipation in living systems. Noise pulse optimisation is thus important for the invention and noise pulses could be improved with this in mind. Currently pulses are optimised according to an iterative interpolation scheme which is partly in the literature. A preferred optimisation technique uses simulated annealing to improve pulse profiles. Simulated annealing is described by Kirkpatrick S., Journal of Statistical Physics, vol 34, P 975, (1984) and as applied to NMR pulse design by C.J. Hardy, P. . Bottomley, M. O'Donnell, P. Roe er, Journal of Magnetic Resonance, 77, P233, (1988). Better definition of pulse profiles is possible, such as one which produces less excitation away from the slice, giving just as sharp slice edges as before but requiring slightly less RF power. Tailoring the bandwidth and number of points in the noise pulse to improve the randomisation of magnetisation would also result in reduction of required power.

2) Remnant coherence in the transverse magnetisation produced by the method of the invention can appear in an FID or simulated echo as contamination. Including side lobes on the cosine gradient waveform can dephase this and avoid the problem. Other gradients in the succeeding sequence may have the same effect.

3) Improvement of randomisation can be achieved by the use of different noise pulses at each step in the saturation sequence. This may also be applied to the case where gradients are just in 2 directions to select a square ROI„ Different noise pulses may also be used for each phase encode step in an imaging experiment.

4) Arbitrary convex volumes may be optimised by varying the angle between successive gradient steps according to the shape being selected.

5) Arbitrary concave shapes may be possible but with some signal from parts of the required volume.

6) If the noise pulses produce uniform coherent excitation inside the slice, instead of zero escalation then at the end of the method of the invention, spins in the ROI would be, partially at least, in a steady state and could be examined with some steady state sequence for imaging or localised measurements.

7) The invention may be performing using a surface coil to receive, and possibly to transmit also, if the noise pulses are capable of exciting magnetisation to sufficient depth without nulls.

Applications of the invention in imaging are as follows -. -

1 ) Optimum use of receiver bandwidth, in imaging may be obtained by following the method of the invention with an imaging sequence, thus potentially improving signal to noise by just collecting signal from the ROI.

2) Zoomed images of the ROI, without aliassing, can be obtained by following the method of the invention with an imaging sequence with increased gradients or reduced sweep width.

Having selected a volume using the invention, a multislice imaging sequence can then follow. The tendency of the nucleii in the material surrounding the region of interest to revert to their initial spin state during the multislice imaging sequence can cause contaimination in later ones of the multislice images. One way to avoid this is to use a 2 shot process. This involves waiting after the first multislice imaging sequence for the system to fully revert to its initial spin state and then repeat the selection process of the invention before repeating the multislice imaging sequence, applying an RF inversion pulse so as to rotate the spin axis of all nucleii (both inside the reion of interest a well as outside) though 180°. The results of the subsequent multislice imaging sequence may then be subtracted from those of the first multislice imaging sequence to eliminate contamination effects due to the previously mentioned reversion (relaxation) of the nucleii in the surrounding material.

4) Volume selective snapshot images can be obtained by following the method of the invention with a fast imaging technique. EPI is favourable because its low RF power dissipation makes up for the power dissipated by the invention. Other possibilities include FLASH or CEFAST.

5) A 3D image of the selected volume can be made by following the method of the invention with a suitable imaging sequence.

6) Volume selective maps can be made of Chemical shift/Perfusion/Diffusion/relaxation/flow.

7) Motion artefacts in vivo can be reduced by suppressing regions most responsible for the artefact, e.g. fat layers with respiratory motion or the heart in upper spine studies.

8) Multiple echo/CMPG sets can easily be acquired by following a method according to the invention with a train of pulses.

Applications of the invention in spectroscopy are as follows:-

1 ) A spectra from an FID may be obtained by applying a hard or soft excitation pulse immediately after the inventive selection method to avoid the need for a spin echo. A prefocussed pulse can be used for excitation so that the slice gradient also acts as a crusher for unwanted transverse magnetisation.

2) Signal-to-noise can be optimised by making the selected region conformal to the ROI using a conformal method according to the invention in 2 and 3 dimensions.

3) Selective measurements of quantities^' may be made:

(a) T-_| by preceding the invention with a 180 degree pulse.

(b) ₂ by using a multiple echo sequence or CPMG.

(c) Multiple quantum of J resolved spectra can be acquired by appending suitable sequences.

(d) Diffusion/perfusion/flow can also be measured selectively.

Throughout the foregoing text mention has been made by number to various References. A key to those References is set out below:

Key to References 1. J. FRAHM, K-D MERBOLDT, and W. HANICKE, J. Magn. Reson. 72, 502 (1987)

2. P.A. BOTTOMLEY, T.B. FOSTER, and R.D.DARROW, J. Magn, Reson. 59, 338 (1984)

3. A. COϋNELLY, C. COUNSELL, J.A.B. LOHMAN, and R.J.ORDIDGE, J. Magn. Reson. 78, 519 (1988)

4. P.A. BOTTOMLEY, C.J.HARDY, . Magn. Reson. 74, 550 (1987)

5. J. PAϋLY, D. NISHIMURA and A. MACOVSKI, J. Magn. Reson. 81, 43 (1989)

6. W.P. AUE, S. MULLER, T.A. CROSS and J. SEELIG, J. Magn. Reson. 56, 350 (1984)

7. P.R. LUYTEN, A.J.H. MARIEN, B. SIJTSMA, and J.A.DEN HOLLANDER, J. Magn. Reson. 67, 148 (1986)

8. D.M. DODDRELL, W.M. BROOKS, J.M. BULSING, J. FIELD, M.G. IRVING and H. BADDELEY, J. Magn, Reson. 68, 367 (1986)

9. D.M. DODDRELL, J.M. BULSING, G.J. GALLOWAY, W.M. BROOKS, J. FIELD, M. IRVING and H. BADDELEY, J. Magn. Reson. 70, 319 (1986)

10. B.L. TOMLINSON and H.D.W. HILL, J. Chem, Phys. 59, 1775 (1973)

11. R.J.ORDIDGE, A. CONNELLY and J.A.B. LOHMAN, J Magn, Reson . 66 , 283 ( 1 986 )

12. R.J. ORDIDGE, Magn, Reson, Med. 5, 93 (1987).

The invention will now be described further by way of example, with reference to the accompanying drawings, in which:

Figure 1(a) shows a simple three-step saturation sequence;

Figure 1(b) illustrates the idealised 2D response to the sequence of Figure 1(a);

Figure 2 illustrates an example of a selective saturation prepulse;

Figure 3 shows a simulated 2D effect using a 32 step version of a pre-pulse, and is a profile along the diameter of the selected region;

Figure 4(a) and (b) show the amplitude and phase of the response to a simple noise-window pulse when (c) and (d) show the response to the optimised pulse used in the experimental work; (a) and (c) are thus the frequency spectra obtained by Fourier transformation, whilst (b) and (d) show the transverse magnetisation of a uniform spin system, obtained by integration of the Bloch equation;

Figure 5A shows a slice through an agar phantom described in the text, with no selective pre-saturation pulse;

Figure 5B shows the same slice as Figure 5A acquired using the pre-pulse but with no RF power in the noise pulses so that any loss is due solely to eddy current.

Figure 5C shows the same line as in 5(b) but using the noise pulses to select a small circular region in the centre by saturating all the magnetisation outside;

Figure 5 also contains a profile through the centre of the three images, and it is to be noted that any variation in intensity from left to right across the image is due to B_* inhomogeniety in the probe;

Figure 6 illustrates how a volume of interest selected from a phantom may be moved around the sample;

Figure 7(a) shows the Fourier transform of the frequency response of a typical noise pulse; and

Figure 7(b) shows the response of a uniform spin system to such a pulse.

The invention lies in a method of using Nuclear Magnetic Resonance equipment and in such equipment when programmed to operate in accordance with the invention. For a general discussion of NMR apparatus, reference is made to the textbook by E. Fukushima, S.B.W. Roeder, "Experimental Pulse NMR". Addison-Wesley, (1981).

In this new technique, the B field gradient vector is incrementally rotated so as typically to describe a circle, in a number of discrete steps, while the RF is pulsed after each adjustment of the magnetic field gradient vector direction. Figure 1a shows a simple three-step case, while Figure 1b illustrates the ideal response from a uniform plane of spins; the shading represents partial saturation, the darker the shading the greater the degree of saturation. As the number of gradient steps is increased, the region of zero saturation becomes more clearly defined and in that regard this technique is conceptually analogous to the back-projection (12,13) method for image reconstructions.

A simple algorithm, based on the geometrical principles demonstrated in Figure 1b, has been implemented on a Micro VAX II computer to simulate the effects of arbitrary gradient waveforms with an RF pulse (producing an ideal, uniform staturation profile outside a central window of zero excitation) applied after each gradient step.

Figure 2 shows a typical presaturation pulse where the gradient vector describes a circle and Figure 3 shows the simulated idealized response to such a prepulse with 32 gradient steps. The RF power is set so that 99% saturation occurs away from the circular region of interest, when each individual pulse reduces the Z magnetisation by only 14%. As is shown later (Figure 5), this simulation is in excellent agreement with experimental results and also illustrates one of the advantages of this technique in that since each RF pulse is only required to achieve partial saturation of the magnetisation, much less RF power is required than if saturation is required every one RF burst.

One of the practical problems associated with the implementation of this localisation technique is that the excitation profile of a simple noise window pulse is far from the ideal "square" profile assumed for the simulations, and some magnetisation is excited inside the noise window. This is illustrated in Figure 4a which shows the frequency response of the noise pulse described in Figure 7b, obtained by Fourier transformation, but in which the time data were zero-filled from 256 to 512 points to show the effects of the finite pulse length. The Fourier transform of the desired spectrum (Figure 7a) is an infinite- function which we approximate to be a finite pulse of 256 points. This is equivalent to multiplying the infinite pulse by a square "top hat" function of width 256 points, which has the effect of convolving each point in the frequency response with a sine function of period equal to the sampling interval in the time domain. Fourier transforming the 256 points defining the pulse (Figure 7b) gives a Nyquist frequency exactly equal to that of the convolving sine function, so that it is sampled at the bottom of each oscillation and the resulting spectrum is that of Figure 7a, with an apparently perfect null inside the central window. Zero- filling the pulse to 512 points doubles the Nyquist frequency so that the sine function is sampled at the top and bottom of each oscillation, as can be seen in the central window of the spectrum in Figure 4a. Figure 4b shows the transverse response of a uniform spin system to this pulse, calculated to the same resolution as in the frequency spectrum by iterative integration of the Bloch equation using rotation matrices, with the pulse scaled to give zero average Z magnetisation outside the selected region. A significant amount of magnetisation is excited within the central region due to a combination of the effects of pulse truncation (giving the high frequency oscillations) and spin system nonlinearity. Since this reduction in longitudinal magnetisation will cause signal loss from the volume of interest when several noise pulses are applied successively, some pulse optimisation is clearly necessary. Previous work on noise-modulated RF pulses (14) has concentrated on improving "randomisation" and bandwidth, whereas the invention aims to improve the excitation null in the central window. The pulse truncation effect can be reduced by multiplying the noise function used by a Gaussian envelope so that the convolving function in the frequency domain is another smooth Gaussian. However this takes no account of the nonlinearities.

A simple iterative process may therefore be used by which integration of the Bloch equation yields the response of a uniform spin-system to an initial RF pulse shaped as shown in Figure 7(b). The response is first modified by setting all points within the 5% central window to zero, this being the most desired feature of the response, and then Fourier transformed to give an approximate RF pulse shape. This can be inserted into the Bloch equation integration routine and the process repeated. This algorithm gives pulse optimisation which is oscillatory. One of the best results is illustrated in Figure 4c which shows the much improved excitation null within the central window. Figure 4(d) shows the transverse response to the same pulse but with the scaling set to reduce the averge Z magnetisation by 14%, as used in the simulation, and illustrates the corresponding reduction in excited magnetisation with the central window. The magnitude of the transverse magnetisation created by this optimised pulse is more uniform than that created by the un- optimised pulse but the phase is still quite random and there is no problem with the creation of unwanted coherence.

In one experiment, work was carried out using an Oxford Reseach Systems Biospec 1 console with a 2T, 31cm bore Oxford Instruments cryomagnet and a home built, 21cm i.d, gradient set. The flood phantom used contained 0.5% agar in a O.lmM solution of McC1₂ with T₁ and T₂ values of 880 ms and 100 ms respectively. Figure 5A shows a proton image of an 8 mm slice through this phantom, acquired in one scan using a standard spin-echo imaging sequence with no RF gradient pre-pulse, while Figure 5B shows an image acquired using the same sequence but with selective saturation pre-pulse and zero RF power for the noise pulses. The simulation of these two images illustrates that the loss in signal intensity due to eddy currents caused by the changing field gradients in the pre-pulse was small, measured by integrating pixel levels of the image region at 6%. Figure 5C was acquired using the same pulse-sequence as for Figure 5B but with the RF power turned on during the pre-pulse to isolate the selected volume. The diameter of the phantom was 4.5cm while the selected volume measured 7mm in diameter. All receiver amplification values and display parameters were kept constant for the three images and as the profile across images 5B and 5C shows, excellent outer volume suppression is achieved, while virtually all of the Z magnetisation available after eddy current effects is retained inside the selected region. The mean signal loss in the region of interest, caused by noise pulse imperfections, was measured as 28% and the structure seen within this region is consistent with the noise pulse excitation profiles in Figure 5.

The suppression ratio for signal from the outer volume can only be reliably measured by spectroscopy.

The pulse sequence used for Figures 5b and 5c consisted of a selective pre-pulse with 32 gradient steps (an upper limit dictated by the spectrometer, due to memory constraints), followed by a standard slice-selected spin- echo imaging sequence of echo time 21ms. A stabilisation delay is left between the pre-pulse and the imaging sequence, in this case

30ms, which should be kept as short as possible to minimise the effects of T^ relaxation of spins in the outer volume. Any magnetisation in the outer volume which relaxes back along the Z axis after the 90 degree pulse at the start of the imaging sequence will simply be flipped along the -Z axis by the 180 degree refocussing pulse and so will not contribute to the spin-echo signal. Thus controlled T₂ weighting may be applied simply by altering the spin-echo time.

A 3ms noise pulse, defined by 256 points, was used for each gradient step in the pre-pulse and the magnitude of the gradient was 6 mT/m. Thus the total pre-pulse length was 96ms. T₁ relaxation during the selection sequence may be a problem for some faster relaxing species, however, preliminary calculations indicate that this effect may be counteracted by increasing the RF power, although possibly at the expense of some signal loss from the region of interest.

An important practical point in the implementation of the method of the invention for volume selection is the effect of eddy currents induced in the magnet by the gradients used in the pre-pulse. The fields due to these currents can decrease the signal/noise ratio in the resulting image/spectrum by dephasing the transverse magnetisation created by the sequence used following the pre-pulse. This effect can be seen in a comparison of the images in Figures 5A and 5B. Fortunately both the smooth shape and symmetric nature of the gradient waveforms employed minimise these effects. Those eddy currents that do arise are caused mainly by the edges of the small steps which make up the gradient shape, and it is possible to greatly reduce their effect with a small amount of pre-emphasis in the gradient producing signal generators (drives).

Pre-emphasis may be obviated by greatly increasing the number of. steps defining the gradient shape, so that the length of each step roughly corresponds to the gradient rise time.

The use of shielded gradient coils may be particularly advantageous to reduce the effect of eddy currents.

A useful feature of the method of the invention is that non-rectilinear regions of interest may easily be defined by changing the field gradient producing waveforms in the pre-pulse from simple sine/cosine shapes. In fact, any non-reentrant shape may be defined with- 100% of the magnetisation available from within the selected region, introducing potential signal-to-noise ratio enhancements over the application of rectilinear sampling windows.

A wide range of other shapes are accessible, albeit with some loss in the total signal intensity obtainable.

The region of interest can be moved easily to any part of the sample by changing the frequency offsets for each of the noise pulses, according to the distance and direction of movement required and the strength of the field gradient. Figure 6(a) shows the image of a whole phantom with no volume selection. Figure 6(b) shows an elliptical volume of interest of elliptical section removed from the centre of the sample. A further change in the transmitter offset for each of the noise pulses in the selective saturation sequence was changed so as to move the selected volume to the edge of the sample as shown in Figure 6(c). In Figure 6(d), the gradient waveforms were modified as well, to rotate the selected volume by 45° in the slice plane.

The duration of the RF signal employed in the experiments to date was 96ms, long enough for significant T^ relaxation and spin diffusion to occur in some systems. However this probably represents an upper extreme for the pulse length and preliminary experimental results indicate that a reduction by a factor of 2 or 4 may be achieved without very significant loss in either region definition or outer volume suppression.

Claims

1. A method of selecting a region of interest within an object whose nuclear magnetic resonance is to be investigated in which the object is located within a zone within a generally uniform intense magnetic field, which zone is influenced not only by the intense field but also by two or more local magnetic fields, so that within the zone a magnetic field gradient exists so that different regions of the object are influenced by different magnetic field intensities to therefore possess differing magnetic resonance frequencies characterised in that the local magnetic fields are varied so that the direction of the magnetic field gradient across the zone is incrementally rotated through a number of discrete positions and an RF signal having a broad spectrum of finite energy frequency components and a defined band of zero or very low energy frequency components, is applied after each incremental step, so as to cause random nuclear spin orientations of nucleii in the material located in those parts of the zone for which the RF signal contains energy at the appropriate frequency, but no changes in the orientation of the nuclear spins of nucleii located in those parts of the object material for which the RF signal does not contain energy at the appropriate frequency, so that as the field gradient rotates a volume of material is defined (the selected region) in which no changes in the orientation of the nuclear spins occurs, whilst in general the nucleii in the remainder of the material will have random spin orientations.

2. A method according to claim 1, wherein the number of different gradient directions is chosen to optimise definition of the selected region.

3. A method according to claim 2, wherein the RF pulses are optimised in by simulated annealing.

4. A method of NMR imaging comprising the steps of selecting a region of interest in accordance with any of claims 1 to 3 and then performing any known nuclear magnetic resonance imaging sequence on the material within the selected region.

5. A method according to any of claims 1 to 4, wherein the RF signal includes a carrier component which is shifted as required after each incremental rotation of the field gradient so as to enable a region of interest to be defined which is not co-axial with the axis or rotation of the field gradient.

6. A method according to any of claims 1 to 5, wherein complex volumes of interest are generated by adjusting any or all of the field gradient, the width of the defined band of zero (or very low) energy frequency components of the RF signal, or the angle between successive gradient directions.

7. A method according to any of claims 1 to 6, in which the RF signal is produced by modulating a suitable carrier with a noise signal from which unwanted components have been filtered.

8. A method according to claim 7, in which the noise signal is randomly generated and a different randomly generated noise signal is employed for each succeeding signal.

9. A method according to any of claims 1 to 8, in which the local magnetic fields are generated by electric currents which possess an essentially cosine waveform profile but in which the steep transitions at the beginning and end of the waveform are replaced by less steep rising and falling edges.

10. A method according to any of claims 1 to 9, in which the directions of the local magnetic fields are orthogonal.

11. A method of nuclear magnetic resonance spectroscopy comprising the steps of any of claims 1 to 4, followed by any known- nuclear magnetic resonance spectroscopy sequence.

12. A method of selecting a region of interest in nuclear magnetic resonance investigation, substantially as hereinbefore described.

13. A method of NMR imaging as claimed in claim 4, substantially as herein described.

14. A method of NMR spectroscopy as claimed in claim 11, substantially as herein described.

15. Apparatus for performing the method of claim 1, comprising

- means for establishing an overall intense magnetic field; - means for introducing two or more magnetic fields in a zone of the overall field within which an object to be inspected by nuclear magnetic resonance is to be located, the introduced magnetic fields co-acting with the overall field to produce a gradient in the magnetic field across the zone;

- means for generating an RF signal having a broad spectrum of finite energy frequency components and a defined band of zero (or low) energy frequency components and applying this RF signal to the said zone;

- means for changing the introduced magnetic fields so as to cause the resultant field gradient in the said zone to rotate incrementally through a series of positions; and

- means for controlling the supply of the RF signal and restricting its supply to periods in the series of steps immediately following each incremental repositioning of the field gradient.