NMR Imaging and Spectroscopy
- 1983
These captions and transcript were generated by a computer and may contain errors. If there are significant errors that should be corrected, please let us know by emailing digital@sciencehistory.org.
Transcript
00:00:30 Paul C. Lauterbur received his Ph.D. in chemistry from the University of Pittsburgh in 1962.
00:00:53 He is leading professor of chemistry at the State University of New York at Stony Brook,
00:00:59 as well as research professor of radiology at the same institution.
00:01:05 He is an expert in the fields of physical chemistry and of medical imaging.
00:01:11 His research has included work on nuclear magnetic resonance studies of structure and
00:01:16 properties of molecules, crystals, and biological systems, imaging by magnetic resonance zeugmatography,
00:01:27 including biological and medical applications.
00:01:31 He is a fellow of the American Physical Society, and from 1981 to 1983 was president of the
00:01:38 Society of Magnetic Resonance in Medicine.
00:01:42 He received the gold medal of that society in 1982 and the APS prize in biological physics
00:01:49 in 1983.
00:01:50 Dr. Lauterbur will speak on NMR imaging and spectroscopy.
00:02:04 Thank you.
00:02:07 Very grateful to have this opportunity discussing this work with an audience that brings here
00:02:15 many different viewpoints.
00:02:18 Perhaps first I should mention the strange word that came up in the course of the introduction
00:02:25 and was incidentally perfectly pronounced, zeugmatography.
00:02:32 I attempt to call this field NMR zeugmatographic imaging, but as has often been the case where
00:02:46 someone gave a funny middle name to a child, it tends not to use it anymore.
00:02:52 But I think I will try to point out to you that it was not merely a whim, but actually
00:02:59 that there is something of substance in that attempt.
00:03:05 Also what I'm going to tell you is in some sense a cautionary tale.
00:03:15 It's not so much a matter of a story of finding ways to solve problems, but more a matter
00:03:24 of discovering that there were no problems to be solved in order to achieve a certain
00:03:31 result and that by and large the problems were entirely in our own heads.
00:03:41 Why, for example, did it take more than a quarter of a century after the first demonstration
00:03:51 of nuclear magnetic resonance in condensed matter for anyone to begin using nuclear magnetic
00:04:00 resonance signals to make pictures?
00:04:04 And as it has turned out during the past 10 years, NMR signals are almost ideally suited
00:04:12 to the making of images.
00:04:14 That is one of the most straightforward, at the same time subtle and versatile methods
00:04:20 available for making images of many different things.
00:04:27 And the principle that underlies the formation of images from NMR signals is extremely simple.
00:04:37 It's nothing more than the Larmor equation.
00:04:40 If the magnetic field is different at different locations of the same nuclear species, the
00:04:48 resonant frequency will be different.
00:04:51 So that if you have a non-uniform magnetic field within an object, the nuclei in different
00:04:59 locations will have different resonance frequencies.
00:05:05 And if you simply measure the resonant frequencies and the amount of signal present at each resonant
00:05:13 frequency, you have a representation of the object.
00:05:17 And by appropriate manipulation of that information, you can get images in one dimension, two,
00:05:26 three, even four dimensions, as we shall see.
00:05:34 In fact, the work of George Paik on structures of crystals using NMR is an excellent early
00:05:46 example of the use of the idea that if you know the distribution of magnetic field in
00:05:55 terms of the coordinates within the crystal, that you can then determine from the frequencies
00:06:02 of the resonance phenomena, the frequency shifts, the coordinates of the nuclei giving
00:06:07 rise to the signals, and hence can learn something about the structures of crystals.
00:06:14 In that case, of course, the sources of the inhomogeneous magnetic fields within the crystal
00:06:22 are other atomic nuclei, but there is really nothing terribly different in all that.
00:06:30 So the experiment that one has to do in order to make an image from NMR signals, or in fact
00:06:41 from any kind of spectroscopic line, is to shift that line by the use of some external field.
00:06:53 It could, in principle, be a magnetic field or an electric field, a velocity field, a
00:06:58 MacDoppler shifts, anything that one can apply overall to some object that can produce a
00:07:04 frequency-dependent shift in a spectral line.
00:07:11 So this process, when described in that way, obviously has nothing whatsoever to do with
00:07:27 the methods of imaging that are familiar from all the history of science and all the
00:07:35 history of everything before science, and that is a method that's traced back to descriptions
00:07:39 in physical optics.
00:07:41 There is no mention of wavelength at any point in this process, has nothing to do with diffraction.
00:07:48 We are merely labeling the photons that give rise to the image with frequencies that correspond
00:07:57 to locations in space.
00:08:00 And here is one of the reasons that this particular method of using spectral signals in general,
00:08:08 NMR signals in particular, was not recognized for a long time, because in everyone's mind
00:08:17 there was the equation of imaging is optical imaging.
00:08:23 And the question would naturally arise, well, there is a wavelength associated with the
00:08:29 frequency that one is measuring in this phenomenon, and therefore, in some way or other, that
00:08:38 must limit the resolution.
00:08:41 This is obviously not true in the NMR case, typically getting resolutions of the orders
00:08:46 of millimeters or better, with wavelengths of the orders of meters, tens, hundreds of
00:08:52 meters, there's obviously no relationship, and in the description I've given you, there
00:08:56 is no relationship whatsoever.
00:08:58 So this is a method of imaging that has nothing whatsoever to do with the usual methods of
00:09:04 imaging that we're familiar with, and that's where the funny name came from.
00:09:08 The zeugmatography, the zeugma comes from a Greek word meaning something that joins
00:09:13 two things together.
00:09:15 And it is the simultaneous interaction of the nuclear spins with an applied field gradient
00:09:24 that tunes them, in effect, to different frequencies, and a second field that reads out the signal
00:09:32 intensity at each of those frequencies, it gives rise to the information that produces
00:09:36 an image.
00:09:38 So that the usual functions of a single radiation field or stream of particles or whatever that
00:09:44 gives rise to an image are split into two.
00:09:47 And it is one field that defines location, another field that provides you the means
00:09:51 of observing that interaction.
00:09:55 OK.
00:09:57 All this seems very simple after the fact.
00:10:02 In the early days when I described this work, it was not at all uncommon for very accomplished
00:10:06 physicists in the audience to point out that they weren't quite sure what, but there must
00:10:11 be something wrong with it.
00:10:14 Expressions were used like mistake, swindle, violates the uncertainty principle, probably,
00:10:22 and things of that nature.
00:10:24 But again, that is merely a matter of saying that optical imaging is the only way that
00:10:29 exists to find out where things are.
00:10:33 Now, another much more technical factor that arose in interpreting this kind of technique
00:10:44 was the attempt to take a rather messy phenomenon observed in the laboratory and to give it
00:10:52 a simple form that could be handled analytically.
00:10:55 The peculiar line shapes that arise when an object of arbitrary shape is placed in
00:11:03 a magnetic field with some arbitrary non-uniformity tended to be replaced by Gaussians, which
00:11:11 could easily be handled, or even worse, by Lorentzians.
00:11:15 And so that one would say, for example, if the magnetic field was inhomogeneous by a
00:11:20 Gauss over an object a centimeter in size, that that would cause for a proton resonance
00:11:26 about a 4 kilohertz spread of the signal.
00:11:31 And instead of looking at it in detail, one would say, well, that corresponds to a decay
00:11:39 time of the order of, let's say, 80 microseconds.
00:11:42 And then you would turn right around and say, therefore, if we assume that this can be represented
00:11:48 by an exponential decay, then you are looking at a Lorentzian line width, a Lorentzian line
00:11:55 with about that width at half height that I previously mentioned.
00:11:59 And obviously, there is no useful information in that line.
00:12:05 This is not a hypothetical example.
00:12:08 I have had come to me not only chemistry undergraduates and graduate students, but a physics postdoc
00:12:17 who broke out into a cold sweat one night after he had actually made images in our laboratory
00:12:23 and said, I've just proved that it cannot be done.
00:12:28 I don't know what is happening, but we can't be doing it.
00:12:31 And this was, of course, a result of using this common procedure of smuggling into the
00:12:35 equations an assumption of a simple line shape, which, of course, wipes out all of the interesting
00:12:41 detail that we are looking for in the first place.
00:12:45 Another problem that I observed to constantly arise arose from the fact that if you apply
00:12:53 a non-uniform magnetic field to an object, in the usual case, what one does is set up
00:13:00 a situation in which there will be surfaces of constant magnetic field intersecting the
00:13:05 object.
00:13:06 And the signals you observe are double integrals of the signal intensity over the intersection
00:13:11 of a surface of constant magnetic field with the arbitrary object.
00:13:14 And hence, they are one-dimensional representations of the object.
00:13:18 And it was usually suspected, or at least a gut feeling, that there probably didn't
00:13:25 exist any well-defined inversion procedure for deducing from these doubly integrated,
00:13:32 one-dimensional representations what the real object was.
00:13:36 And in fact, I know also one accomplished physicist who, after having seen images made
00:13:41 by this technique, was wondering what kind of procedure we had used to, again, smuggle
00:13:49 into the data the final answer, because it didn't seem possible.
00:13:56 The final bastion of those who felt uneasy with the whole thing was that, well, it probably
00:14:08 didn't matter anyway, because the only interesting thing that it looked like people might find
00:14:15 to do with these techniques was to look at human beings, medical purposes, and so on.
00:14:24 And you couldn't really expect to build magnets large enough and homogeneous enough in a practical
00:14:30 way to do that kind of an experiment.
00:14:34 Again, it turned out that it was absurdly easy to build large homogeneous magnets, resistive
00:14:41 magnets, superconducting magnets, even permanent magnets, with quite enough homogeneity to
00:14:47 do perfectly satisfactory experiments of this kind.
00:14:51 So all of the problems that had to be faced were entirely imaginary.
00:14:58 And even to some extent, the problem that these techniques might require a tremendous amount of
00:15:04 computation to perform the inversion procedure is, to some extent, imaginary.
00:15:08 Of course, it's easier with modern computers.
00:15:11 The first images that I made were made by doing hand calculations on a pad of paper
00:15:18 during very dull seminars.
00:15:21 I don't know what's going on out there now.
00:15:23 But one can do, for simple low-resolution images, procedures literally on the back of
00:15:31 an envelope that can demonstrate that these matters work and that can give one images
00:15:36 of some sort.
00:15:37 People have also developed analog techniques that can generate such images.
00:15:41 So really, there were nothing but imaginary problems.
00:15:46 Now let's try seeing what shows up on the slides.
00:15:52 Let me get this pointed the right direction and start pushing the buttons.
00:16:01 This was taken with the first publication on the subject, the example of taking an object
00:16:07 consisting of two cylinders, applying magnetic field gradients in different directions so
00:16:13 that the signal strength of the two locations was slightly different, and projecting out
00:16:17 a one-dimensional representation like these with different orientations of the magnetic
00:16:22 field gradients.
00:16:23 Linear gradients are the easiest to use.
00:16:25 The transformation is very simple.
00:16:26 In this direction, in the magnetic field gradient, both of these objects are in the same magnetic
00:16:31 field, and the signals are then superimposed.
00:16:35 And the original method that we used for generating an image of the object from these one-dimensional
00:16:41 representations was the reconstruction from projections that was actually, at that time,
00:16:51 that was originally in about 1971 to 2, x-ray computer tomography was not widely known.
00:17:00 And it was unfortunately one of these cases where it was necessary to reinvent a wheel
00:17:06 and to invent an algorithm for a projection reconstruction.
00:17:13 Actually it turned out that Radon had done all of this in the fundamental mathematics
00:17:16 in 1917, and everybody was reinventing that wheel.
00:17:20 But at any rate, it's a very straightforward process to modify these projections or the
00:17:24 image itself to get a true image from a set of one-dimensional representations.
00:17:30 Here is one of the first kinds of one-dimensional representations of an object consisting of
00:17:35 two 1-millimeter capillaries, and surrounded by a mixture of H2O and D2O to give a lower
00:17:41 proton density, a weaker hydrogen NMR signal here, about 4 millimeters across.
00:17:49 And from four of these and an iterative algorithm, the first image was produced like this on
00:17:54 a 20-by-20 array using a very primitive computer, and by hand outlining the contours of constant
00:18:01 density so as to see this outer low-density region and the two 1-millimeter capillaries.
00:18:06 And so that was first produced, oh, 11 years or so ago.
00:18:14 And we then proceeded on to small animals, inventing a process that up until then had
00:18:22 not been recognized as essential in NMR work, and that was introducing the sample to the
00:18:28 sample tube in order to make it feel more comfortable with the subsequent experiment.
00:18:33 So here we had our first mammalian sample sniffing about the sample tube.
00:18:39 This was the equipment we used in the early days, being inserted by a physicist here,
00:18:44 Dr. Lai.
00:18:46 And now this is a one-dimensional representation of a mouse.
00:18:49 And you can sympathize with the people who doubted whether from signals like this, this
00:18:54 is about 4 kilohertz across, people who doubted whether from signals like this one could
00:18:58 ever rediscover the structure of the mouse.
00:19:02 But from 12 projections and an iterative algorithm, this was the first example of what is in effect
00:19:08 a cross-section of the thoracic region of a living mouse, the low-density regions of
00:19:13 the lungs, and so on, a very under-determined, low-resolution reconstruction, but nevertheless
00:19:20 beginning to give an inkling of what could be done.
00:19:24 We also, by the way, since this is not a medical meeting, I should mention a few things, some
00:19:29 of them really unpublished, that we did about that time.
00:19:32 This is a one-centimeter cube of opal.
00:19:35 And opal has, within its structure of silica, trapped very small amounts of water as saturated
00:19:43 silicic acid.
00:19:44 And so we made an image from the relatively mobile water phase within this apparently
00:19:49 hard rock.
00:19:51 And that is an image of the opal showing the distribution of water phase within it.
00:19:57 This is not a gem opal, which would have been much more uniform, not as interesting from
00:20:02 this point of view.
00:20:05 We also made images from true solids, the silicon-29 resonances in quartz.
00:20:11 And here we have an image of two samples of amethyst in terms of the silicon-29 signal.
00:20:19 And the spin lattice relaxation time for silicon in quartz is very strongly dependent upon
00:20:25 impurity concentrations.
00:20:29 And doing a partial saturation experiment, the signal from the slightly darker amethyst
00:20:33 shows up stronger, and that for the weaker amethyst shows up, for the lighter amethyst
00:20:40 shows up weaker because it remains more saturated.
00:20:43 And it is conceivable that one could use images of this sort with more modern equipment to
00:20:51 do some interesting studies of mineralogical objects, art objects, objects of technological
00:20:57 interest as well.
00:21:01 This is an example of some work we did in collaboration with an oil company looking
00:21:06 at cores containing a mixture of oil and water that you get when you try to force oil
00:21:15 out of a cylinder of sandstone, essentially, with water.
00:21:21 And these are slices from a three-dimensional image going successively down from the top
00:21:27 through the middle through the bottom.
00:21:30 And because of differences in relaxation time, we see this blobby character.
00:21:33 We can see separately the distribution of the water and the oil.
00:21:37 This was a very early experiment of low resolution, but indicates one of the possibilities for
00:21:43 looking at the distribution of fluid phases within opaque objects.
00:21:49 We even did some experiments with electron spin resonance, a little piece of paper here
00:21:53 soaked in a stable free radical, put in an X-band cavity to which were applied magnetic
00:21:58 field gradients appropriately and reconstructing the distribution of the unpaired electrons.
00:22:05 Other experiments of this sort have also been done, some of them by the South Africans looking
00:22:10 at impurity centers in diamonds.
00:22:16 Another kind of experiment that was done in those early days had to do with combining
00:22:20 zoogmatography and spectroscopy.
00:22:23 That is, obtaining NMR spectral information about molecular structures and the distributions
00:22:28 of different molecular species within a complex object and getting all that information at
00:22:35 the same time.
00:22:36 This was an object which we had, actually, sulfuric acid here and water here and organic
00:22:41 liquid here, para-tertiary-butyl-nitro-benzene, the free induction decay following a pulse
00:22:48 and the Fourier transform of that giving the very low frequency, high resolution NMR spectrum
00:22:54 showing the signals of the various components.
00:22:58 I don't have any slides to demonstrate it, but within the past year or so, other groups
00:23:03 have with modern high field superconducting magnets suitable for use with humans and animals
00:23:12 have begun to produce images showing the distribution of different proton-containing compounds such
00:23:18 as water and fat within the living animals and humans.
00:23:23 One method, by the way, that is used without any real imaging techniques that is used in
00:23:31 order to obtain NMR information about living things is simply to place a radio frequency
00:23:37 transmitting in a receiver coil over a muscle or on top of the head or over the liver or
00:23:45 something like this and to observe the total integrated NMR signal from the region in the
00:23:51 immediate vicinity of the coil.
00:23:53 This has been used for the study of a number of diseases and physiological events and is
00:23:58 now being combined with the imaging so as to obtain all these kinds of information at
00:24:06 once.
00:24:07 Ultimately, three dimensions of spatial information and at least one dimension of spectral information,
00:24:12 making what I mentioned before at least four-dimensional images.
00:24:16 Anyway, the first task here was to devise techniques for getting both spatial information
00:24:22 and spectroscopic information, and this shows that by applying a magnetic field gradient
00:24:30 initially in this direction, large compared to the spread of spectral frequencies, that
00:24:37 one could essentially define regions of constant frequency in planes intersecting the object
00:24:45 as shown here, 1, 2, 3, and then to do a selective excitation by using a radio frequency exciting
00:24:52 pulse that was relatively long, so it had Fourier components covering only a narrow
00:24:56 frequency range, hence a narrow magnetic field range, hence a narrow range of this
00:25:02 coordinate, and to excite then only the material along this line as shown here, or only the
00:25:08 material along this line as shown here, or only along this line as shown here, and by
00:25:13 repeating this process and going through a reconstruction algorithm and so on, one can
00:25:19 eventually make an image.
00:25:22 This again is very low resolution from six projections in the early days, very limited
00:25:26 computing available, but this shows where the water was, where the sulfuric acid was,
00:25:31 and where the organic liquid was, and these techniques are now being applied with much
00:25:37 more power and resolution in living things.
00:25:41 Another example of the way in which NMR spectroscopy is being used to study living organisms is
00:25:50 this actual fake example here.
00:25:55 This is a sample in a set of test tubes made up from the materials that are actually observed
00:26:01 in living tissue by laying a coil over a muscle, for example, the three phosphorous resonances
00:26:06 in ATP, a phosphorous resonance in creatine phosphate, and a phosphorous resonance in
00:26:13 inorganic phosphate.
00:26:15 So that we have here a resolved spectrum, and these similar spectra are commonly observed
00:26:21 now, as I said, for studying exercise metabolism, disease, and being used in practical medical
00:26:29 diagnosis in several situations.
00:26:31 By applying magnetic field gradients, we could broaden these signals by amounts that depended
00:26:36 and in patterns that depended upon their locations within the object, and by the use of enough
00:26:41 magnetic field gradients, eventually could reconstruct the distributions here.
00:26:46 This is where the creatine phosphate is located.
00:26:48 This is where the ATP is located, and this is where the inorganic phosphate is located
00:26:53 in this small test object.
00:26:54 Again, people have recently begun carrying out these methods with much more powerful
00:26:59 modern equipment on living organisms and are beginning to produce separate images of individual
00:27:05 phosphorous metabolites, thus combining the spatial localization and the spectroscopy.
00:27:15 Another kind of thing that one gets into, just to throw out a tantalizing bit that has
00:27:19 never been followed up, NMR signals depend upon motion in a magnetic field, and in particular,
00:27:30 the nature of the NMR signal that you observe can depend upon the direction of motion, such
00:27:35 as flow, within a liquid relative to the direction of the magnetic field gradient.
00:27:40 The result is that a particular volume element in an image has an intensity or a phase or
00:27:47 some other characteristic that depends not only upon its location, but upon its velocity
00:27:53 vector, its direction, and that makes it necessary of using reconstruction algorithms to find
00:28:00 a way to reconstruct not a scalar set of numbers, but a set of vectors.
00:28:05 And we had made some start a few years ago on this.
00:28:08 Here is an example where we were looking at bifurcated flow coming in here, going out
00:28:12 here, and here, and the result of carrying out a reconstruction algorithm from simulated
00:28:16 data to indicate that the techniques of image reconstruction might possibly be extended
00:28:25 to higher orders.
00:28:26 In fact, there seem to be ways of extending it to all orders of tensors.
00:28:30 Here's another example with a simulated phantom showing both vector and scalar components,
00:28:38 just the scalar component, and just the vector component.
00:28:41 These are simulated examples, but one can imagine translating them with the aid of appropriate
00:28:46 experiments into real examples.
00:28:49 But this is something that has not yet occurred, and I hope one of my students follows it up
00:28:54 again someday.
00:28:55 If not, maybe someone else would enjoy it.
00:28:59 The methods that are then used for getting the images from the one-dimensional signals
00:29:07 are many and varied.
00:29:09 One method that is still used in many circumstances is that of projection reconstruction.
00:29:15 These are, this is for example, a one-dimensional projection of a uniform sphere, and in order
00:29:23 to make an image of that sphere from a set of these one-dimensional projections, one
00:29:30 method is to use a filtered back-projection algorithm.
00:29:33 And this merely involves convolving this projection with another function, and then smearing the
00:29:41 resulting functions back across a three-dimensional volume from all possible directions, and,
00:29:48 with a limited set of course, and then the result is a build-up of the signal intensity
00:29:54 in the center that represents the sphere.
00:29:57 It turns out that in three dimensions, image reconstruction is an even simpler problem
00:30:02 than in two dimensions.
00:30:04 In two dimensions, the filter is not local.
00:30:08 In fact, this is one over here.
00:30:10 In three dimensions, it is local.
00:30:12 It is merely a matter of evaluating the Laplacian at the point.
00:30:15 So you can really just do a second derivative, or just a three-point second difference filter
00:30:19 here.
00:30:21 This is what happens to the projection.
00:30:23 Then you smear these functions with their negative wings back over the three-dimensional
00:30:27 volume.
00:30:29 The positive portions add up.
00:30:30 The negative wings wipe out all the things that shouldn't be there, and this is a trace
00:30:34 through the center of that sphere, zero intensity out here.
00:30:38 Uniform intensity through the center, zero intensity on the other side.
00:30:42 So this, again, is actually a very simple algorithm to put into a computer to implement.
00:30:48 A method that is very widely used now also is one in which you actually do direct multidimensional
00:30:55 Fourier transforms and make use of the phase behavior of the signal, which you can directly
00:31:01 observe because these signals last for a long time.
00:31:05 You can stimulate the NMR signal, for example, with a 90-degree pulse here.
00:31:09 You can put on a gradient in the x-direction, which causes the signal to evolve with a frequency
00:31:17 that corresponds to its coordinate in the x-direction, to its x-coordinate.
00:31:21 Then you can switch to a gradient in the y-direction.
00:31:24 The evolution of the signal occurs now with a different frequency corresponding to its
00:31:27 y-coordinate.
00:31:29 You can do this Fourier transform in order to get the distribution in y.
00:31:35 And then you can change the length of time during which the x gradient is allowed to
00:31:42 operate, and in that way, in effect, trace out point by point the phase change that occurs
00:31:49 after a given period of time in the signal evolving in the x-coordinate, and do a second
00:31:54 transform with these, and get directly the object from a two-dimensional Fourier transform.
00:32:01 You can also do this in three dimensions.
00:32:03 You can also do it in four dimensions, where you put on three gradients out here, and then
00:32:09 observe the signal in the absence of a gradient when the signal that you're observing is only
00:32:15 evolving in the spectral dimension.
00:32:20 Another one of the objections, if you recall, that I indicated was you could never build
00:32:23 a magnet big enough to do people.
00:32:26 This was what our graduate students proposed during a Christmas party some years ago as
00:32:30 my potential solution to this problem.
00:32:33 And that was to make the people smaller.
00:32:35 And they chose one of their favorite faculty members, who lectured in organic chemistry
00:32:38 as it turned out, and showed how we might put him in a standard-sized machine.
00:32:44 In fact, this did not turn out to be necessary, as I've indicated.
00:32:52 And the problem of both making big enough magnets and observing NMR signals in them
00:32:58 turned out, again, to be one where untutored intuition was far off.
00:33:03 And an exceedingly simple calculation showed that there was basically no problem.
00:33:07 Long ago, we did the calculation in which essentially what we said is, what is the signal
00:33:13 strength, what is the signal-to-noise ratio that you can observe with an RF coil big enough
00:33:17 to go around a body, and with a given little volume of water?
00:33:23 And we said that if you can observe a high enough signal-to-noise with a reasonable averaging
00:33:31 time for a certain volume of water, we assume that the various processes for deducing the
00:33:40 locations in space of these small volumes can be made to operate with a negligible degradation
00:33:45 in the signal-to-noise ratio, which is, in fact, true.
00:33:49 And therefore, if you simply calculate the potential resolution on the basis of how small
00:33:55 a volume can you observe to give a given signal-to-noise at what we thought then was a practical value
00:34:02 of magnetic field for a large magnet, you can construct such a trade-off here, which
00:34:06 tells you that if you average signal over a time of the order of a few minutes, that
00:34:11 you can get down to a few millimeters resolution.
00:34:14 And enough signal-to-noise to resolve the signal-intensity differences that we expected
00:34:20 to see in human beings.
00:34:21 And this provided all the motivation that was necessary to go ahead with the development
00:34:24 of medical applications, because it showed that if you did everything right, that you
00:34:30 were going to get results that would be medically useful.
00:34:34 And the first system we built up here was a 1 kilogauss, 0.1 tesla magnet, with a resistive
00:34:42 magnet generating magnetic field uniform to a couple of parts in 10 to the fifth over
00:34:46 a volume like this.
00:34:48 Magnetic field gradient coils, which stuck out of our magnet.
00:34:52 The newer magnets, they're all inside.
00:34:55 And the RF transmitting and receiving coil in here.
00:34:59 And you only need three gradient coils, of course, because linear magnetic field gradients
00:35:02 combine like vectors.
00:35:04 So you simply vary the currents and point your gradients in any direction whatsoever.
00:35:09 The first machine we built looked like this, sitting in the basement of the chemistry building,
00:35:13 where it still sits doing various experiments.
00:35:16 And well, the magnet, the gradient coils, the power supply, pulsed NMR spectrometer,
00:35:22 pulsed NMR postdoc, a departmental secretary, all that you need to put together a publicity
00:35:27 picture.
00:35:30 The commercial machines now have a tendency to look more like this.
00:35:33 This is a large superconducting magnet in one of the companies that is being used for
00:35:39 some experimental imaging.
00:35:41 And these are now in use, giving magnetic fields up to about 1.5 tesla for resonant
00:35:48 frequencies of over 60 megahertz.
00:35:56 And by the way, they are being housed sometimes in temples, such as this one at Cleveland
00:36:00 Clinic, to house two of those large 1.5 tesla systems for medical imaging.
00:36:07 This is becoming a new object of worship in the medical profession.
00:36:15 And I'll show you some of the reasons why now.
00:36:17 Without going into all the details of NMR pulse sequences and so on, one of the most
00:36:23 important reasons for the tremendous growth of popularity of these techniques is that
00:36:29 the intrinsic contrast between one tissue and another, one organ and another, damaged
00:36:34 tissue and normal tissue, is very high if you look at the relaxation times.
00:36:40 Because NMR spin lattice and spin-spin relaxation times for water and fat in tissues often vary
00:36:47 by factors of two and three and more, so that one can see very sharply the differences that
00:36:53 appear, even at relatively low signal-to-noise ratios.
00:36:57 And if you use various pulse sequences, for example, observing echoes here, if you use
00:37:04 a time interval between pulses and pick off the magnetization at this point, for example,
00:37:09 a brain signal will give a stronger signal than cerebral spinal fluid, and you will get
00:37:14 dark for this fluid in the ventricle here and lighter for the brain.
00:37:17 If you pick off at this point, you may get nothing interesting in your head.
00:37:24 This is why the radiologists are requiring the consultation with physicists about these
00:37:30 problems now, of course.
00:37:32 And then if you look out here at this pulse interval, you will pick up cerebral spinal
00:37:36 fluid with a stronger signal than the signal in the surrounding brain.
00:37:40 So you have a true multi-parameter image here, and you can get out many different results,
00:37:47 and you can tune the procedure to match the problem.
00:37:51 Here is what is essentially a hydrogen concentration image of the brain, a so-called coronal slice
00:37:56 like this.
00:37:58 This is the image obtained with a series of 90-degree pulses showing some differentiation
00:38:05 between white and gray matter.
00:38:07 Here is a signal obtained with an inversion recovery technique using inverting pulses
00:38:14 in which we see the signal from the gray matter almost entirely disappearing and seeing very
00:38:19 clearly the structure of the white matter.
00:38:21 And here is a spin echo signal in which we see essentially T2 variations and bright signal
00:38:28 for cerebral spinal fluid and not much other structure.
00:38:31 But it turns out these are often very useful in showing up damage to the brain.
00:38:37 For example, here is an example of an image obtained showing some damage to the brain
00:38:44 here using a T1 image.
00:38:47 So this dark region represents long T1.
00:38:52 And an image of the same patient of the same circumstances showing the region here where
00:38:57 this represents T2, and the bright shows longer T2, which in fact shows some surrounding edema
00:39:03 as well as a central lesion that you picked up in the T1 image.
00:39:06 So there's a tremendous amount of flexibility here.
00:39:09 One way of representing these complex multi-parameter images that is being tried is to represent
00:39:17 the proton concentration as signal intensity and the T1, for example, as color on a color
00:39:25 scale.
00:39:26 So here we have a slice taken from the head of one of the people working on the project
00:39:33 showing the scalp with a short T1, cerebral spinal fluid with a long T1, and so on.
00:39:43 Here are some other examples of the kind of things that can be done with contemporary
00:39:46 or near-contemporary devices.
00:39:48 You can use your heart to control the NMR spectrometer.
00:39:52 You can take the electrical signal from the electrocardiogram and feed it into the computer
00:39:57 and have it then emit a radio frequency pulse at the constant point in the heart cycle and
00:40:04 build up over a number of heartbeats enough information to reconstruct an image, for example,
00:40:10 of a slice through the heart.
00:40:11 And this is one done at a time when the left ventricle was open.
00:40:14 And by adjusting the separation of the pulses here, one can see when the left ventricle
00:40:21 is closed.
00:40:22 These are the arms on either side.
00:40:23 This is a slice side to side through the body, the liver down here and the lungs in here
00:40:30 and so on.
00:40:32 So one can actually control one's apparatus directly with one's own heart, which I think
00:40:38 is something that all of us have wished we could do from time to time.
00:40:45 This is an example of some of the relatively good representation of the quality of images
00:40:51 that can be obtained with contemporary devices.
00:40:53 This was obtained at about 0.5 tesla, 5 kilogauss, again of one of the people working on the
00:41:00 project.
00:41:01 You can see the brain, the spinal cord coming down here, the tissue layers within the tongue,
00:41:08 all sorts of fine detail that's familiar to a radiologist.
00:41:12 Some of it is such fine detail that radiologists don't even know, haven't yet found what it
00:41:17 is or have to go back to their anatomy books to restudy.
00:41:20 Here's a more recent one done at 60 some megahertz, 1.5 tesla, showing, I think this will be officially
00:41:28 shown for the first time in a couple of weeks in Chicago, so don't tell anyone.
00:41:35 They done at this very high frequency of a cross section of the brain, and one can see
00:41:39 small blood vessels and all sorts of detail.
00:41:42 These can be done in just a few minutes.
00:41:47 Here is a section from another company, I'm being fair, this is from GE, sliced through
00:41:52 the neck showing the cord coming down here, shoulders coming out here, vertebral bodies,
00:41:59 muscle bundles and so on.
00:42:00 A tremendous amount of detail, totally unobtainable by any other technique.
00:42:04 Here's an abdominal image, also taken at 0.5 tesla.
00:42:08 You can see the vessels coming down from the region of the lungs, you can see the bladder
00:42:12 down here, the kidneys, and so on, and magnification of part of this image, you can see a tremendous
00:42:20 amount of detail, the vasculature and intestinal loops and so on, all kinds of things.
00:42:31 That was a quick run through of the present state of the art commercially, and a common
00:42:38 opinion now among radiologists and other diagnosticians who have had a lot of acquaintance
00:42:43 with this technique over the last couple of years with more or less modern instrumentation,
00:42:49 is it will probably replace x-ray computed tomography for most applications in the head,
00:42:56 and the odds look good on its replacing x-ray CT for many conditions in the pelvis.
00:43:05 Other abdominal applications are as yet somewhat uncertain and up in the air.
00:43:12 But in any case, there is a considerable informed body of opinion now that says that NMR techniques
00:43:17 may largely displace x-ray techniques for sophisticated medical imaging diagnostic procedures,
00:43:24 thus relieving us all from a considerable burden of ionizing radiation.
00:43:31 Now I'm just going to finish up with a couple of recent things from my laboratory that suggest
00:43:35 new directions in which these techniques can go.
00:43:38 These are some images of the distribution of a gas within the lungs of a dog.
00:43:44 Of course, the NMR signals from a gas are very much weaker than those from a liquid,
00:43:47 roughly a factor of 1,000 corresponding to the density difference.
00:43:52 One can compensate for that simply by relaxing the resolution requirements in volume by a
00:43:57 factor of 1,000, which means only a factor of 10 in linear resolution in three dimensions.
00:44:02 So instead of trying to resolve things at the millimeter level, we try to resolve things
00:44:05 at the centimeter level, and we get about the same number of nuclei in the volume element.
00:44:11 And secondly, by using inert fluorinated gases, it is possible to separate the signal from
00:44:16 the gas from the much stronger signals from the surrounding body structures.
00:44:20 So here is, in effect, a front view and a side view and a head-on view of the distribution
00:44:26 of gases of this fluorinated gas within the lungs of a living dog who is inhaling it.
00:44:31 And in fact, one can, well, somehow this got sideways, that's all right, don't worry.
00:44:38 The dog is now lying down, and so here we have the chest wall, the head end, the diaphragm
00:44:44 and liver, the heart.
00:44:46 And all shown in white is a proton two-dimensional shadow image, and in blue is a gas distribution
00:44:53 superimposed by a primitive process of laying two pieces of film on top of one another.
00:44:58 With better imaging electronics, that can be avoided.
00:45:02 And so one can see the anatomy and also see the functional behavior of the lungs more
00:45:08 or less simultaneously.
00:45:11 Another thing that we have been doing relatively recently is making microscopic images.
00:45:16 There's no limit to the resolution of these techniques imposed by optics, for example.
00:45:22 There's no reason that you could not see two atoms separated by an atomic distance
00:45:29 if you had a large enough gradient in some field that was able to perturb the frequencies
00:45:38 of one relative to the other.
00:45:40 Of course, if they start out different, there's no problem.
00:45:42 But if there were two identical atoms near one another, and one could have a Stark effect
00:45:46 or a Zeeman effect or whatever so as to shift the spectral lines a little bit, you could
00:45:52 resolve atomic distances.
00:45:55 The only limitation on resolution is when you try to look at too small a volume, the
00:46:03 volume of the material will not contain enough nuclei so that you will not have a large enough
00:46:08 signal-to-noise ratio to detect with any available apparatus.
00:46:14 And ultimately, if you go down to small enough dimensions, you will begin getting into diffusional
00:46:20 problems as well.
00:46:21 But those will not be easily reached.
00:46:24 So here is an example.
00:46:25 This is one millimeter across here.
00:46:27 The space within here is 50 microns.
00:46:30 The walls of this flattened capillary here are 50 microns.
00:46:33 And this is an image made in a few minutes by a physics student in my laboratory, sections
00:46:39 from a complete three-dimensional image.
00:46:40 The whole three-dimensional image is made in a few minutes, and it's just sliced up
00:46:44 in the computer for display.
00:46:46 So the size of the picture element is about 20 microns.
00:46:51 We're limited by the resolution in the computer program and display at this point.
00:46:57 But this is entirely consistent with the calculations that you can make that you should be able
00:47:01 to get down at least to something of the order of 5 to 10 microns resolution, and hence make
00:47:07 a three-dimensional microscope untroubled by opacity in the visible region that would
00:47:15 make it possible to make images of all sorts of objects with their structures being represented
00:47:24 by different parameters than one is usually used to.
00:47:30 Whether this will prove to have a lot of uses remains to be seen.
00:47:38 This is an example from the same three-dimensional image of slicing that object in a perpendicular
00:47:43 direction starting from close to the side of the tube.
00:47:47 So you start slicing down through the object and seeing another view of it.
00:47:54 Finally, I think that this was a very appropriate quotation that some colleagues of mine in
00:48:04 London called my attention to.
00:48:06 Back in 1885, T.H. Huxley said, what an enormous revolution would be made in biology if physics
00:48:14 or chemistry could supply the physiologist with the means of making out the molecular
00:48:17 structure of living tissues comparable to that which the spectroscope affords to the
00:48:22 inquiry into the nature of the heavenly bodies.
00:48:24 At the present moment, the constituents of our own bodies are more remote from our kin
00:48:27 than those of Sirius in this respect.
00:48:30 Well I think the physicists and chemists more or less backed into this one, and almost without
00:48:36 knowing what they were doing, developed all the techniques that sort of came together
00:48:41 and made it possible to look into the human body without any known damage, and to pick
00:48:48 out the locations of various substances of interest, to see the variations in the signals
00:48:55 that result from changes in the local biophysics and biochemistry, and all in all I think perhaps
00:49:02 to come as close as one ever can to satisfying this wish of Huxley's.
00:49:06 Thank you.