APPLICATIONS OF FUNCTIONAL IMAGING IN NEUROPSYCHOLOGY

 

CONFERENCES
TOPIC: MISCELANEOUS

---

 

Abstract

Reports on Functional Brain Imaging have become ubiquitous in the
neuropsychological and cognitive neuroscience literature today. Yet,
evaluation of the merits of such reports requires a certain degree of
familiarity with the fundamentals of the various functional imaging
technologies and with the basic structure of the procedures employed for
constructing brain activation images with each. Accordingly, the aim of
this tutorial is to acquaint neuropsychologists with such fundamental
aspects of imaging, so as to enable them to form reasonably sound judgments
about its actual and potential contributions in their field.

To accomplish this aim, I will first review those concepts that apply to
all imaging methods. Second, I will explain the process of image
construction using each of the following methods: Positron Emission
Tomography (PET), functional Magnetic Resonance Imaging (fMRI) and Magnetic
Source Imaging (MSI). Finally, I will review the main procedures used, in
the context of each method, for establishing the correspondence between
brain activation images on the one hand and neuropsychological functions on
the other, placing special emphasis on ways of assessing the reliability
and the validity of both the images themselves and their association with
specific functions.

In order to facilitate understanding of the admittedly complex technical
aspects of the imaging methods, several drawings and illustrations will be
used, and those technical terms that appear frequently in the clinical and
research literature, will be systematically defined. It is intended that at
the conclusion of this tutorial, those who have studied it will be able to
1) explain what, precisely, is represented in functional images, 2)
describe how functional images are constructed with each of the most
frequently employed methods (PET, fMRI, MSI), and 3) evaluate with
reasonable accuracy the merit of functional imaging studies purporting to
address neuropsychological issues.

The purpose of this tutorial  is to outline the basic aspects of three representative imaging methods, Magnetoencephalography (MEG), Positron Emission Tomography (PET) and functional Magnetic Resonance Imaging (fMRI), and to explain how we go about establishing the correspondence between images of brain activation patterns obtained with each of these methods, and psychological functions.

Baseline Activation

Activation, the object represented in functional images, refers to many different biochemical and physiological events which may be grouped into two categories corresponding to the two fundamental neurophysiological processes:  Metabolism and Neural Signaling.  The events that constitute metabolism include changes in the regional rate and volume of blood, relative quantities of oxygen in the blood of rates of perfusion, rates of consumption of oxygen or glucose, and many others.  The events that constitute signaling among neurons include the release of neurotransmitters in the synapses, as well as the flow of ions within cells and outside of cells, that is, electrical currents.
Some of these biochemical and physiological events are associated with electromagnetic signals which radiate from their point of origin inside the brain to its surface, where they can be recorded by special instruments.  These electromagnetic signals do not interact causally with the biological events that constitute activation and, because they do not affect them, functional imaging is considered non-invasive.

The processes of metabolism and of signaling are continuous, but the rates at which they transpire vary from one brain region to the other. In fact, each region has its own characteristic level of baseline activation.  Obtaining the baseline activation profile is the easiest form of functional imaging because such profiles reflect the total amount of activation of each brain region, which remains more or less constant over time. Consequently, if we could view and measure directly the rate of metabolism or the rate of signaling of neurons in each part of the brain and plot it over time we would likely arrive at a figure like the one below.  Moreover, if we were to obtain, at different points in time, functional images of this activation profile, the images would be very similar, as shown in the same Figure.

This tutorial is based on the book "Fundamentals of Functional Brain Imaging" by A.C.Papanicolaou. (Lisse: Swets and Zeilinger, 1998)
 
 

Fig 1 A schematic rendering of a hypothetical activation profile of a section of the brain.  Each line represents the degree of activation of each small region of the brain, as it changes slightly from moment to moment.  This baseline profile of activation is represented in stylized images that could have been obtained at two different points in time.  Different colors represent different degrees of activation of the various brain structures. Since the profile remains more or less constant over time, the functional images are almost identical.

The main use of functional images of baseline activity is to reveal malfunctions, or pathological deviations from the expected baseline rates of metabolism and signaling of particular areas of the brain, either abnormally low (hypo-activation) or abnormally high (hyper-activation).  There are two kinds of such deviations or malfunctions.  The first is chronic or constant over time and the other phasic, appearing intermittently at particular times.  An example of the former is Parkinsonís disease where a particular brain area, the internal globus pallidus, is constantly hyperactive and its constituent neurons are signaling (and metabolizing) at high rates.

An example of a phasic malfuntion may be an epileptic discharge resulting in transient deviations from the normal level of activation of the affected structures. Such deviations may be captured in images taken when the deviations occur.  The brain regions that function abnormally can then be identified by comparing those images to others taken before or after the abnormal episode. The uses of functional images of baseline activation will be further discussed in later sections of this tutorial.
 

Function-specific activation patterns

Of far greater interest for neuropsychologists are the functional images of activation patterns, embeded in the global, baseline activity, that are related to behavioral and psychological functions. This is because we have reason to believe that the correspondence between a given activation pattern and a given function is not accidental and superfluous, but that each activation pattern reflects aspects of the neurophysiological mechanism which is a necessary condition of each function.

A function may be viewed as the process of production of a set of similar phenomena (whether subjective, like perceptual experiences or objective, like movements) that serve a common purpose. A brain mechanism is a set of events that take place in particular brain areas in a particular order, resulting in the generation of the phenomena that define each function.

When the mechanism is operating, it gives rise to a pattern of activation which is embeded in the global profile of the baseline activity.  In order to identify the activation pattern that corresponds to the brain mechanism of a specific function, we must arrange things so that the naturally occurring function may occur on demand, within the laboratory, while we are recording the activity of the subjectís brain.  We must, in other words, provide the occasion that would trigger the phenomena that define the function.  To do so we may present stimuli of the same kind as those that naturally trigger the function, and we may instruct the subjects to deal with them as they would under normal circumstances.  The stimuli and the instructions constitute the experimental task.  If the phenomena that define the particular function are subjective, specific instructions may be given to the subjects to indicate their occurrence by some pre-specified discriminant response which can stand as objective proof that the phenomena have, in fact, occurred.  If the stimuli and the task instructions do occasion the function, the task is said to be ecologically valid and capable of activating the brain mechanism of the function

To establish the connection between activation patterns and brain mechanisms of functions is quite different from recording baseline or abnormal activation profiles.  This is because when we look at the baseline activation profile, we usually do not see any major changes in it that would allow us to recognize when and what brain areas are especially activated, for the following three basic reasons: First, at any given point in time many functions, besides the one we are interested in, are concurrently performed.  Second, the temporal variations in the baseline activation profile that are seemingly random are due to the changing energy requirements for the execution of all functions that the brain performs concurrently, and not just the one we are interested in. Third, the energy requirements for each of these concurrent functions, including the one of interest, are relatively low as compared either to baseline differences from one anatomical structure to another or to obvious abnormal deviations from baseline.  That is, the activation pattern specific to each function, at each point in time, is buried in the seemingly random temporal variation and can only be extracted and imaged if, first, we know exactly when the function begins and ends and second, if we can induce repetitions of the phenomena that define the function.  Both of these requirements presuppose that the phenomena that define the function are objectively specifiable and repeatable.  If these two requirements are met, we can then extract the activation pattern corresponding to them through the procedures of either averaging or of integration and subtraction.  The first of these procedures is used in the context of MEG; the second in conjunction with the other functional imaging methods.
 
 

 
The nature of activation Imaged

Signaling among neurons constitutes one of the basic forms of activation that can be imaged with the current methods.  It consists of electrochemical events that take place at the synapses, in the axon and the dendrites of neurons.  With the exception of the phenomenon of neurotransmitter release and uptake, which does not involve directly electrical activity, all others involve flow of electrically charged particles, or ions, which results in electrical current.

Were we to view directly the variation of the electrical currents at each and every set of cells in the brain, which are referred to as current sources, and were we to plot these variations as a function of time, we would obtain the typical picture of activation that we have considered before ( Figure 1).  That is, we would find that the amount of signaling each source is producing changes from moment to moment in an apparently random manner but within certain limits.  We consider the randomness in that variation only apparent because we do not know what is the purpose of each ripple of activation, or to what end each source is signaling at each point in time.  We assume, however, that signaling always serves to mediate some function or other, or that the pattern of signaling throughout the brain that corresponds to each of the many functions that are taking place simultaneously, is contained in this apparently random variation and that special procedures are necessary to isolate it, extract it and image it.  But we have also commented on the fact that, at times, abnormal deviations in activation that clearly exceed the normal range of its variation take place, and that these do not require any special procedures for their isolation and extraction.  In the following paragraphs we will describe how we record and image , with the method of MEG, such deviations in signalling that are visible in the baseline activation profile and then we will describe how we extract function-specific signalling patterns.

Let us then assume that a set of cells that are typically not synchronized, begin to signal in unison.  Their combined electrical currents will create a large deviation, much beyond the typical range.  Such a phasic deviation could well be an epileptiform discharge as shown in Figure 2.
 

Fig 2 A schematic rendering of the electromagnetic signals recorded on the head surface echoing the electrical currents inside the brain.  A transient deviation in electromagnetic signal intensity over a particular region of the head surface reflects the coordinated signaling activity of a large set of neurons somewhere in the brain.

 In such cases, using MEG, one can answer questions like: where is the source of this deviation (i.e, what area of the brain is epileptogenic)?  Needless to say, the pattern of activation of the brain itself is hidden from our view.  We have no direct access to the source currents themselves.  We only have indirect access to the degree that these currents give rise to another form of electromagnetic energy which can travel outside the head where it can be captured and recorded, as shown in Figure 2, namely the magnetic flux.
 

Recording the Magnetic Flux

The magnetic flux is recorded by means of magnetometers.  These are superconducting loops of wire positioned over the head surface.  As the flux lines thread through the loop, they create in it current by induction.  The strength of the current is proportional to the density of the flux at that point, so that knowing the value of the induced current, we have a measure of the flux strength at that point.  If a sufficient number of magnetometers are placed at regular intervals over the entire head surface, then the shape of the entire distribution created by a brain activity source can be determined.

On the basis of the surface flux distribution, the position and strength of the brain source that produced it can be estimated. Once the estimates are made, the estimated source (i.e. the activated brain region) is identified using the following procedure: Three fiducial points are defined on the subjects head surface. Usually they are clear anatomical landmarks like the two pre-auricular points and the nasion.  These three points define the coordinate system that includes the brain and the position of the magnetometers relative to it.  The line between the pre-auricular points defines the y-axis of the coordinate system.  The line between the nasion and the mid point of the x-axis and perpendicular to it, defines the x-axis and the line perpendicular to the x-y plane, passing through  the intersection of the x and y axes, defines the z-axis of the coordinate system, as shown in Figure 3.

Fig 3 The system of Cartesian coordinates, anchored on fiducial landmarks, defines the space that contains the brain activity sources.  The position of each source can therefore be specified with reference to these coordinates.

   Thus, the exact position of the recorded distribution or the exact distance of each point of the head surface from the origin of the z-y-z coordinates is known.  Also, the position of the source is defined with reference to this coordinate system and so is its relative orientation.  Usually, lipid markers (e.g. vitamin pills) are attached on these three fiducial points, and a structural MRI is taken, either before or after the MEG recording session.  The positions of the markers are visible on the MRI scans.  Therefore, the relative position of all brain structures with respect to the position of the source of activity is also known.  Given this fact, co-registration of the MEG-derived active source and the structural MRI is possible: The position of the source or sources can be projected onto the appropriate MRI slices resulting in composite images of the type shown in Figure 4.

Fig 4 A typical image representing a cluster of sources of phasic abnormal events (like epileptiform discharges) on a patient's MRI.

The Averaging Procedure

As mentioned in the first section of this tutorial, additional procedures are necessary to extract activation patterns specific to particular brain functions, embeded in the global, baseline activation profile. Also mentioned, was the fact that in the case of MEG, such extraction is accomplished with the "averaging procedure." This procedure will be explained using, as an example, a simple sensory function like vision, occasioned by light stimuli.
Averaging is applied to the flux recorded on each and every surface point during the phenomena that define either a motor or a sensory function, in this case, during presentation of light flashes (and the resulting visual experiences).

Fig 5 A schematic representation of the on-going record of magnetic flux over the head.  Each trace represents a record of magnetic flux from a single scalp location.  At successive points in time, a sensory event is repeated n number of times.  Yet no obvious changes in the record of the flux attend each repetition of the event.

As in Figure 1, representing the baseline activation profile, so in Figure 5, we cannot see any appreciable change in the flux when a phenomenon happens, unlike the case of epileptiform discharges considered before.  This is because the amount of additional neural signaling, due to the sensory responses to the light stimuli, is minute as compared to the background signaling which corresponds to all concurrent functions of the brain.  We assume, however, the following: First, that each time the flash occurs there must be an additional pattern of neural signaling that is embedded in the seemingly random variation of the background flux that we record.  Second, that this pattern remains essentially the same every time the flash is presented since the stimulus is identical in all its repetitions.  Third, that whereas at those particular times that the stimulus occurs, the signaling specific to the stimulus is almost the same, all other signaling reflected in the flux that  corresponds to other, concurrent functions, cannot be the same but must vary randomly since, at those particular times, it is highly unlikely that other functions occur in synchrony with the reaction to the stimulus.   Fourth, that at any given time, the several superimposed patterns corresponding to the several functions that jointly constitute the global flux are independent of each other and, as such, it is their sum that constitutes the recorded global, baseline flux.

Averaging is successful only to the degree that these assumptions are correct, and this appears to be the case with functions such as simple movement and sensory reaction, as well as more complex ones like language or memory.

The process of averaging involves the following steps:  First, the flux is recorded from each point of the head surface, during several presentations of the same sensory stimulus.  Each epoch of flux, that is, each portion of activity beginning a few milliseconds before and extending several milliseconds after each repetition of the stimulus, is separately stored.

The epochs are digitized by converting the intensity of the flux at each successive time point into numbers.  Averaging of all the epochs collected at each surface location is then accomplished by adding the digitized epochs and dividing by their total number.  If the four above mentioned assumptions are correct, what emerges as the average epoch is a waveform, or an evoked response of a particular shape, as shown in Figure 6.

Fig 6 Average evoke responses to auditory and visual stimuli. Responses consist of early and late components.

 The evoked responses consist of early and late components. The former correspond to activation of the sensory cortex specific to each type of stimulus and the latter, typically, to activation of the association cortex. That is, the former enable identification of the mechanism of simple sensory and the latter of higher functions.

 Identification of such mechanisms is accomplished through the following steps: The surface distribution of averaged flux at each and all successive time points during the evolution of the evoked response is successively analyzed such that sources that produced each of the successive distributions over the time interval of the response can be calculated and projected onto the structural images of the head as described above.  Figure 7 shows some examples of functional images for different simple sensory functions.

Fig 7 Typical MEG images displaying signs of the mechanisms of the visual auditory and somatosensory functions.  Each symbol represents the computed source location of the averaged magnetic flux recorded at a given point in time after the presentation of a visual, auditory or somatic stimulus.  As expected, sources of visual responses are located in mesial occipital cortex, and the sources of auditory responses in the superior temporal plane, bilaterally.  Stimulation of each of three fingers in each hand is associated with anatomically distinct sources in the contralateral parietal lobe.

To extract the activity specific to a higher function, the same averaging procedure may be used to obtain the late components of evoked responses, as follows. Let us assume that we wish to image the mechanism of verbal memory. To do so, we may present word stimuli to subjects with instructions to, say, identify words that occur more than once in a session. Evoked responses to these stimuli may be recorded and averaged as before. If the study is successful, the early components of the evoked response ought to be accounted for by sources in the primary auditory cortex, whereas the sources at the late components ought to outline the mechanism of the function that was occasional by the task, in this case verbal episodic memory.
 

 
The Nature of Activation Imaged

We have distinguished two basic types of brain activation; neural signaling and metabolism.  The rate of metabolism varies as much as the rate of signaling.  It varies first, from one type of tissue to another and second, it varies in a given tissue sample depending on the amount of work the sample performs.  Since the main type of work neurons perform is signaling, increases or decreases in the rate of signaling are correlated with  increases and decreases in their metabolic rates.

The sequence of physiological processes that result in fMR images can be summarized as follows: If a particular sample of tissue (set of neurons) engages in increased signaling rate, its metabolic rate will increase, resulting in relatively higher rates of oxygen consumption, therefore in lower amounts of hemoglobin (oxygenated blood) in its vicinity.  This change in the concentration of hemoglobin in the sample, becomes apparent about two seconds following the increase in the signaling rate. Next, the reduction of hemoglobin triggers a vascular reaction resulting in an over-supply of oxygenated blood to that tissue sample, 5 to 8 seconds later.
 

Recording fMR Images

If we could view directly the relative content of hemoglobin in every sample of tissue throughout the brain, and the variations in the content of each sample over time, we would obtain the typical resting or baseline activation profile (as shown in Figure 1) and, the functional images of that pattern, taken at different times, would look essentially identical given that the variations corresponding to the sum total of functions performed are minor as compared to the variation among structures, with respect to their signaling and metabolic demands. If a particular region were intermittently malfunctioning (as, for example, in the case of epilepsy) once again, the large deviations in hemoglobin content of the structure after the discharges, as compared to that before the discharges, might be, in principle, visible.

Hemoglobin, relative concentration of which constitutes the type of activation to be depicted in fMR images, is not associated with any electromagnetic signals that could reveal its presence.  Yet it affects in specific ways the behavior of hydrogen atoms which, as it will be explained below, do under some circumstances, emit recordable electromagnetic signals.  And, it is from those signals, indirectly affected by the relative amounts of hemoglobin, that the distribution of the latter throughout the brain is estimated and represented in images.

The special circumstances under which recordable signals are emitted by the tissues in hemoglobin are created by a radio frequency (RF) magnetic pulse delivered by the MRI apparatus to the head which is already placed into a static magnetic field B. This constant magnetic field induces hydrogen protons to precess around the axis defined by its direction, with a frequency which is proportional to its strength. The RF pulse induces the protons to precess in phase around the B axis and adds energy to them. Once the RF pulse stops, the protons begin to precess at random again and to emit their excess energy in the form of electromagnetic signals at the frequency of their precession which, once again, depends on the strength of the magnetic field B. When the hydrogen protons precess in an environment rich in hemoglobin, they emit signals of higher intensity than otherwise. Thus, the presence of excess oxygenated blood in a tissue sample (as a result of activation of that tissue sample) makes that sample distinct from neighboring ones.

But to construct an image from such signals, we need to know the locus of origin of each of the multitude of simultaneously recorded signals.  If all the recorded signals are identical in frequency, it would be impossible to know the point of origin of each.  However, the frequency of the signals can be manipulated by manipulating the static field B as follows. Specifically, instead of using a static magnetic field B that is of uniform strength throughout the area it covers, we use one that is graded, that is, one that has different values at each point in space, as shown in Figure 8.  Let us then imagine any two signals arriving simultaneously from two unknown spots from within the area covered by the graded magnetic field B.  Whether the two signals are of equal or different intensity, they are bound to have slightly different frequency with the one coming from the spot where the B field is slightly stronger having a higher frequency and the other a lower frequency.

Fig 8 When the static magnetic field B is graded along all three dimensions, it has a slightly different value in each part of the space inside the magnet (voxel), represented here by cubes of different shades of color.  Consequently, resonant signals originating in the different voxels will have different characteristic frequencies.

Notice that since we arrange the values of the static field, we know in advance the precise value of field B at each and every point of space in which it is applied. Consequently, we can determine precisely the point in space (and the point in the brain) that the signal came from, since the frequency of the signal is proportional to B at each point in space.  In principle, then, we can arrange the shape of the magnetic field B (using combinations of magnetic coils) such that every region of space or element of space (voxel) inside the MR magnet has a slightly different and unique value.  Therefore, the resonant signals emanating from each voxel will also have a different value, which value we will know in advance.  Thus, the relative intensity of the recorded signals contains information about the degree of activation of each spot in the brain and the frequency of the signals contains information about the location of each activated spot.  That way, spots in the brain having abnormally high or low baseline activation can be identified. However, activation that is specific to particular functions requires additional procedures in order to be extracted from the global, baseline activation.
 

The Integration and Subtraction Procedure

Extraction of function-specific activation patterns through averaging is not suitable for fMRI (or PET or any other kindered imaging methods). Instead, pattern extraction, with these methods, involves the procedure of integration and substraction:

Stimuli may be presented at the same rate as in the case of MEG, but recording of the electromagnetic signals must be made over a greater temporal span that includes a time period sufficiently long for obtaining the requisite number of electromagnetic signals for constructing an image during, before and/or after the repeated execution of several tokens of the function in question, as shown in Figure 9.

Fig 9 Electromagnetic signals recorded over time intervals before (1), during (2) and after (3) the presentation of stimuli (the onset of which are indicated by blue lines), with fMRI or PET.  Signals recorded during each interval are integrated in order to construct a functional image.

Viewing the recorded electromagnetic signals over these temporal spans, we will not be able to discern any systematic difference in the segments of recorded pattern during, before or after the performance of the function.  The images obtained during these three time periods, though they may look slightly different, will not allow us to identify the pattern of activation specific to the function, because that pattern is contained in, and fused with, the on-going global activation that corresponds to all functions concurrently performed by the brain, in addition to the function of interest.

We assume, however, the following: First, that during the period before and/or after the performance of the  function of interest, the corresponding images represent the activation related to all other concurrently performed functions, but not the activation related to the function we wish to image.  Second, that during the time period that the specific function was repeatedly performed, the brain activity, captured in the corresponding image, contains, in addition to all others, the pattern corresponding to that specific function of interest.  Third, we assume that the background activation pattern found in the "before" and "after" images will be practically the same and that it will remain the same during the time period of the performance of the function.  That is, it will also be contained in the image obtained during the performance of the function.  Finally, we assume that the background global activation and the function-specific pattern are independent of each other and they could be separated by simple subtraction.  We subtract, therefore, from image #2, obtained during the performance of the function, either image #1 or image #3, as shown in Figure 10.

Fig 10 Subtracting images recorded either before (1) or after (3) the repetitive performance of a function from the image recorded during the function (2) results in similar "difference" images.  The latter represent the activation pattern specific to the function or the sign of the mechanism of that function.

Usually, the "before" and "after" images are not exactly the same because the state of the subject rarely stays the same over long periods of time.  Consequently, the subtracted image which presumably represents the function-specific pattern may also be different.  It is therefore obvious that, unless the procedure is repeated a number of times and results in practically the same "difference" image every time, we cannot maintain that the image is valid representation of anything, since it is not reliable.
The issue as to how to obtain a stable background activation image (also known as the problem of obtaining the appropriate "control" image) is of great importance, especially where complex or higher functions are concerned. We will comment on it again in a later section of the tutorial.
 
 

 
The Nature of Activation Imaged

Neurons, just like all other cells, utilize a variety of molecules and compounds in order to subsist and to function. Some of these molecules and compounds, molecules of water, for instance, are brought to the cells by the bloodstream. Since the overall activity of cells is not uniform throughout the brain, the distribution of any and all molecules is also not uniform.  Rather, their relative concentration varies across the different tissues and structures of the brain.  Therefore, were we to visualize directly the concentration of any of these substances we would, once again,  see the typical baseline activation profile shown in Figure 1, above.

It is precisely such distributions of particular substances throughout the brain that constitute the object that PET images are intended to capture.  The nature of activation, however, that such distributions represent, differs depending on the type of substance visualized.  For example, if the distribution of a neurotransmitter is visualized, the type of activity represented will most likely be that of signaling, or of potential signaling.  If the distribution of water molecules or of glucose is visualized, the specific type of activation imaged could be either the metabolic rate of neurons, or the local blood flow rates in the different areas of the brain. This is because each molecule or compound is delivered by the blood, is taken up by the tissues, is metabolized, and the products of its metabolism are excreted at different rates specific to each.  Consequently, when and how each type of molecule is visualized, determines whether its distribution would reflect blood flow or metabolic activity.  The choice of particular recording parameters best suited for imaging metabolic or blood flow rates is based on tracer kinetic models which describe the rates at which particular substances are transported to the different tissues and the rates at which they are metabolized.  Therefore, different recording parameters are chosen for visualizing the distribution of particular substances when the intention is to image local blood flow activity and different ones if the intention is to capture local rates of metabolic activity.

The above mentioned molecules and compounds do not emit electromagnetic signals which would reveal their position and relative concentration in the different areas of the brain.  This is because their constituent elements are not radioactive.  Yet it is possible to introduce into the brain, through the blood, equivalent organic molecules that contain atoms that are isotopes of the natural ones and which emit positively charged particles, called positrons.  These, in turn, interact with electrons and produce photons which can be detected over the head surface.

Isotopes are also called tracers or probes because they allow us to trace or to probe the neurophysiological processes that constitute activation. The process of emission is random.  Each positron escapes the nucleus at unpredictable times but, on the average, the time required for all positrons to be emitted is known, and it differs from one type of isotope to the other.  The common measure of the temporal course of positron emission is the half-life which is defined as the amount of time it takes for half of the positrons to be emitted in a sample of a particular isotope.

Molecules containing positron-emitting atoms are introduced into the blood through intravenous injection.  In a short time period after injection, they are dispersed throughout the brain at rates and amounts that depend, as a rule, on local metabolic needs.  In each locality, they are utilized by the neurons at rates and amounts that also depend, as a rule, on the same local metabolic needs.  The pattern of their distribution inside the brain, revealed by the pattern of signals they emit over a particular time interval, becomes the basis for inferring the relative degree of activation of the different brain regions during the same time interval.  How these signals are emitted and how their surface distribution is formed, through interaction with our recording instruments, is explained below.
 

Formation of the Surface Distribution

Let us follow the fate of a single, positron-emitting atom of oxygen (15O) which, as a part of a water molecule, may be carried from the point of injection to a specific part of the brain, by the blood stream.  At unpredictable moments, positrons will escape from its nucleus, one at a time.  The emissions start as soon as the isotope is prepared and they continue for minutes until all excess positive charge is shed from its nucleus.

Once a positron is emitted from the nucleus of the atom, it will speed towards an unpredictable direction and before it traverses a distance of a couple of millimeters it will collide with one of the electrons in its environment.  As a result of the collision the positron and the electron will both be annihilated.  That is, they will be converted into a pair of high frequency photons that will fly with equal speed in diametrically opposite directions.  The energy of these photons is sufficient to propel them without distortion clear through the brain tissues and the skull to the surface of the head. These photons constitute the electromagnetic signals that form a surface distribution which contains information about the distribution of the positron-emitting water molecules inside the brain. The apparatus that records these emissions consists of an array of scintillation detectors arranged around the head.  Each detector consists of a crystal (the scintillator) coupled to a photomultiplier tube. When a photon hits the crystal, visible light is emitted.  The light interacts with a cathode plate inside the photomultiplier tube causing the emission of electrons which end up producing an electrical pulse that is recorded and timed.  The precise timing and counting of the pulses generated by the scintillator detectors is the basis for determining where each positron emission originates, consequently where the tracer molecules are, and in what concentrations they exist in the various regions of the brain. The detectors, it has been said, are symmetrically arranged around the head.  Therefore, to the degree that the head remains motionless throughout the recording interval, its exact position with respect to each detector is known and it is constant. Consequently, a photon pair resulting from a positron-electron collision is likely to interact with a pair of detectors almost simultaneously.  On the other hand, photons that do not belong to the same pair are likely to arrive at any two detectors with a greater time difference.  The arrival time of a pair of photons to a pair of detectors will be exactly the same if the origin of the photons was mid-way between the two detectors.  Otherwise, the arrival times will differ slightly.  In either case, to the degree that the exact time of arrival of each photon is registered, their origin can be determined since the velocities of the two photons are known and the distance of the head from each detector is also known.  In principle, the duration of the time-of-flight of each photon in a pair could be used to estimate the position of the tracer molecule inside the brain.  In practice, however, a different procedure, called back projection, is used for the same purpose.

Given that the number of photons originating in a particular area reflects the number of molecules concentrated in that area over the recording interval, the relative degree of activation of the different areas can be inferred from the relative number of photons originating in each.  At the end of the recording session, thousands of coincidental or nearly coincidental photon pairs are registered by the detectors and the relative timing, as well as the specific detectors that registered each pair, are recorded by the systemís computer.  This set of recordings constitutes the surface distribution of the electromagnetic signals used to construct the images of the profile of brain activation during the recording interval.

As metioned, the most frequently used procedure for developing the surface distribution into activation images is called back projection. The basic principles of back projection as applied to PET may be described briefly as follows: Given a coincident detection of two photons, their common origin is assumed to be, with equal probability, anywhere along the line between the detectors as shown in Figure 11.

Provided that at each spot in the brain a number of positron-emitting molecules are concentrated, each emitting a number of positrons, several positron-electron collisions will occur in its immediate vicinity resulting in several photon pair emissions.  To each pair detected, corresponds a trajectory of coincident emissions and to each point along each trajectory the same probability of it being the origin of the pair is assigned.

Fig 11  It is assumed that a coincident photon pair may originate anywhere in the brain along the emission trajectory.The points where the coincidence lines cross most often indicate the most likely place of origin of the corresponding photon emissions.

Were we to add the probabilities of a photon emission origin at each point, we would find that some spots have a much higher probability of containing the origin of emissions than others.  In Figure 11 two separate sources of emission are shown, each representing two areas of unequal concentration of positron emitting molecules, therefore, two areas of different activation levels.  Applying the same procedure on all trajectories of coincident photons during the recording interval and transforming the sums of probabilities in each pixel into colors, each representing a different range of values of these sums, we create the functional image of a given cross-section of the brain and of all cross sections.

The developed PET images resemble clouds of different hues or shades of gray where each hue or shade represents a different degree of activation of the underlying brain structures.  To identify which are the structures more or less activated, it is often necessary to superimpose these cross-sectional functional images on structural ones, as in the case of fMRI and MEG, described previously.  To superimpose or co-register functional and structural images of the same cross-section of the brain, it is necessary to use some landmarks of the brain or the head that are seeing both in the functional and the structural (MR or CT) scans and align those. When structural scans are not available, the functional images are superimposed on drawings of the brain to facilitate recognition of the activated regions.

As in the case of fMRI to discern deviations in the baseline activation profile (either hypo- or hyper-activation) due to the malfunctions of particular brain areas, the procedures thus far described are sufficient. However, to extract an activation pattern, specific to a particular function, the "integration and subtraction" procedure, described in the previous section, is necessary.

As it can be recalled, that procedure involves subtracting an image recorded when no specific task is performed (control image) from one recorded during a task that occasions the function of interest. However, more complex experimental designs involving more than just one subtraction are commonly used for two reasons: First, because what constitutes a good "control condition" is not always clear, and second, because any experimental task usually occasions more than one single function. An example of such designs will be give in the following section.
 
 

 
A Typical Neuropsychological Study

It is commonly asserted that in the context of a typical speech comprehension task, a hierarchy of linguistic functions, is occasioned. Or, that speech comprehension is not a single brain function but a series of interconnected functions, although how many is a matter that depends on the particular theory one adheres to.
For example, it can be claimed that to understand the meaning of words heard, first, an acoustic function is necessary for determining the physical features of the word stimuli, then a phonological function for identifying the language-relevant features and finally a semantic function (also involving memory) for identifying the meaning of the set of phonological features that constitute each spoken word. Now, to find the activation pattern specific to each function, an experimental design involving more than just one control task is necessary.

An activation pattern may be imaged while subjects listen to words and indicate by means of some discriminant response that they comprehend their meaning.  That pattern is supposed to contain activity specific to the semantic (S), the phonological (P) and the acoustic (A) functions, as well as activity due to the rest of the concurrent functions (R) performed by the brain.  Therefore, the pattern obtained during the task would be a composite of sub-patterns S + A + P + R.

A second activation pattern may then be recorded while the subjects listen to the same words but are now told to ignore what the words mean, to attend to some of their phonological features instead, and indicate through the same discriminant responses that they did so.  This second pattern is supposed to contain all the component function-specific patterns the first one did, except for the  pattern specific to the semantic function; consequently it would be a composite of sub-patterns P + A + R.

Then a third activation pattern can be obtained while the subjects hear the same words but are told to ignore everything else about them, attend exclusively to their acoustic features (e.g. relative loudness) and indicate that they did so by means of the same discriminant response.  In this case the pattern is supposed to consist of only the activity specific to the acoustic and the rest of the concurrent functions, namely A + R.

Assuming that all three experimental tasks required the same level of alertness and effort and that they were completed to the same degree of accuracy, we may subtract the second from the first (i.e., [S + P + A + R] - [P + A + R]) to get the pattern of the semantic function S, and the third from the second (i.e., [P + A + R] - [A + R]) to get the pattern of the phonological function, P.  To obtain the sign of the acoustic function we would need an additional task involving all concurrent functions R but not the acoustic function and perform the subtraction once again (i.e., [A + R] - R) to get the pattern specific to the acoustic function A.

It is obvious that for this scheme to work, the different functions postulated must be independent of each other, involving clearly separate mechanisms, the activities of which are combined through simple addition.  These, along with all other assumptions mentioned in the foregoing description constitute the structure of most functional imaging studies.
 

Uses of baseline activation profiles

We have often commented on the use of baseline activation as a means of identifying chronic and phasic malfunctions of the brain associated with neurological disorders like epilepsy or Parkinsonís disease.  But it is also possible that baseline activation profiles vary depending on subject characteristics, and on the behavioral state of each individual.  These variations may be useful for clinical diagnosis as well as for answering questions about maturation and development of the central nervous system, and about the neurophysiological basis of distinct behavioral traits.

The benefits of these applications of functional imaging, the expected validity of their results, and the technical requirements for obtaining them vary depending on two factors.  First, they depend on the a priori likelihood that the behavioral state, trait or disorder involves easily discernible peculiarities in brain physiology and on the degree of our prior knowledge regarding the nature of those peculiarities. Second, they depend on the correct assignment of individuals in different categories on the basis of their symptoms or behavioral characteristics.

Specifically, most neurological disorders, like epilepsy or dementia, are known to involve alterations in brain physiology and much is already known about the nature of these alterations.  Also, extreme variations of behavioral states, like sleep and arousal, and definite differences in biological characteristics, like gender or developmental stage, are equally well known to involve variations in cerebral physiology. Therefore, the validity of the images can be judged against our prior knowledge or anticipation of what the peculiar activation profiles should be like in each case.  Psychiatric disorders, on the other hand, the brain mechanisms of which are less well known, present special difficulties and require more precautions in assessing correctly the fidelity of functional images that purport to capture signs of their mechanisms.  Finally, when baseline brain activation is imaged with the intention of discovering neurophysiological markers of behavioral or personality traits rather than diseases with known neurophysiological correlates, the technical demands for establishing the validity of the images become tremendous.  But, while the likelihood of erroneous interpretations of the results increases, the possibility of unexpected and truly novel findings increases, at the same time, commensurately.  To verify, for example, that dementia indeed involves hypo-activation of certain brain areas, is certainly useful, but to discover that aggression, let us say, or obsession or creativity are associated with definite and unique profiles of brain activation is to gain insights of unprecedented scope and implications.

As mentioned above, the second factor on which the expected validity of such functional images depends, is the correct assignment of individuals in particular diagnostic categories on the basis of behavioral symptoms or characteristics.   When the behavioral symptoms or the biological characteristics are clear, as for example, age or maturational level, clear and reliable differences in images are typically obtained.

On the other hand, if the behavioral symptoms or characteristics are not sufficiently clear such that individuals are falsely assigned to a particular category, it becomes more difficult to establish the brain activation profile unique to that category if, in fact, it exists.  For example, if many depressed individuals are assigned to the diagnostic group of dementia, establishing the neurophysiological signs of the latter disorder becomes difficult or impossible.  Correct grouping, once again, is easiest for neurological disorders with unequivocal behavioral symptoms and signs and also for clearly definable biological states and traits.  It becomes more problematic for psychiatric disorders and most difficult for personality traits, the presence and relative prominence of which, in a given individual, is a question of culturally-conditioned judgments.  It is, therefore, not surprising that most credible results of functional imaging involve the former cases.

The fact that reliable distinctions among brain activation profiles consistent with expectations may be obtained, does not necessarily imply that such profiles can be used to identify a particular disorder.  That is to say, differences among activation profiles of normal individuals and Alzheimerís patients or among demented and Parkinsonís patients do not necessarily amount to signatures of each disease.  To claim that a particular abnormal activation profile depicts the pathophysiological mechanism of a particular disease one must first demonstrate that individual patients can be correctly assigned to diagnostic categories solely on the basis of their functional images.  Though at present no claims to that effect have been made directly, there is no reason to doubt that, with constant improvements in the fidelity of images, this feat may soon be accomplished.

Recordings of baseline activation profiles are used extensively to assess malfunctions of brain mechanisms of sensory, motor and higher functions.  Often, though, damage to specific brain mechanisms resulting in specific behavioral deficits can be readily assessed with routine diagnostic procedures including structural brain imaging.  Other times, structural imaging fails to disclose the damage that would account for all of the symptoms observed.  Moreover, in some cases, specific behavioral deficits appear, yet, structural imaging fails to identify any brain damage.  The role of functional images in the former situations is primarily to confirm and provide supplementary information regarding the nature and extent of the damage; in the latter, it is to explore the possibility of more subtle alterations in brain activation that may account for the observed symptoms.  In that capacity functional imaging is unique and its contributions to clinical diagnosis truly revolutionary.

There, functional imaging not only contributes to the understanding of the pathophysiology of the observed behavioral or cognitive deficits, but it contributes to the discovery of the mechanism of the function compromised.  For example, though several proposals have been offered regarding the mechanism of episodic memory, there is no certainty regarding the nature of that mechanism.  Consequently, when the function collapses, as in cases of transient global amnesia, for example, no predictions can be made about whether, or where and what brain abnormalities ought to be observed.  In fact, deficits of this type, unattended by any other behavioral signs of brain pathology, have often been called "psychogenic," that is, of psychological as opposed to neurophysiological etiology but with the advent of functional imaging, many such "psychogenic" deficits will receive a neurophysiological explanation.
 

Uses of function-specific activation patterns

Besides baseline activation profiles, function-specific activation patterns can and have been used to explore possible differences in brain mechanisms of particular functions in groups of individuals that differ with respect to some prominent physiological or psychological characteristic, for example, gender, age, or presence or absence of psychopathology.  Here again, the validity of the functional images and the relative difficulty in establishing it, depend on whether individuals can be assigned unequivocally to a particular category, the degree to which the mechanism of the particular function is known, and whether individuals in the groups compared, can perform the same function with equal ease.  Two examples of this type of application of functional imaging will be discussed in some detail to illustrate its utility as well as the requirements for establishing its validity.

There are reasons to hypothesize that the brain mechanisms of language involve left hemisphere structures in men and both left and right hemisphere structures in women.  Such a hypothesis, and others of the same type, can be readily evaluated through functional imaging.  The requirements for evaluating them correctly, that is in a manner that would allow reasonable interpretation of the results, are as follows: First, that all members of each group possess the characteristic on the basis of which they are classified together.  In this example, this requirement is perfectly satisfied since gender is as unequivocal a characteristic as we can reasonably expect.  The second requirement is that, beside this defining characteristic, no other prominent characteristic distinguishes the members of the two groups.  If that requirement is not met, if, for instance, several males happen to be mentally deficient and some of the women creative writers, any difference that may be found in the brain activation patterns of the two groups could not be uniquely attributed to gender.  This requirement can be met only approximately and we may never be absolutely certain that the groups are perfectly matched on all other characteristics.  It will always be possible to find characteristics that differ among any two groups of individuals besides the characteristic that defines the group.  It may turn out, for example, that the members of the two groups differ in height, weight, possibly in eye color, possibly in general intelligence or some other trait.  Though there is no conceivable way of selecting group members who are identical in everything except the group-defining characteristic, it is possible to select them such that they do not differ appreciably in a characteristic relevant to linguistic competence.  This brings us to the third requirement: Members of both groups must be able to perform the same language task with equal ease and efficiency.  Obviously the odds for doing so are better if the general level of their linguistic competence is approximately equal.  If it is not, then, in order to perform the same task, they will have to exert different degrees of effort or they will be more or less alert during the task, both of which factors may affect their brain activation profile and either obscure, exaggerate, or modify possible differences in the language-specific activation patterns.
Each individual in each group may undergo imaging during a speech task that is believed to represent effectively the function of language, and using one of the previously described procedures, the function-specific pattern may be determined for each individual.  Assuming that the fidelity of these patterns is deemed satisfactory we can proceed to examine whether the hypothesis is correct and that the signs of the language mechanism in men are different from those in women.  A quick way of estimating if this is likely the case is to average all the patterns of men together and all of those of the women to see if there is an overall difference.  Assuming that there is, we have next to establish if the difference that we see in the average images is a real or a circumstantial one because, any time we average images, we will get some pattern.

To reassure ourselves that the patterns are not arbitrary, we must inquire whether or not the activation patterns of most individual men and women is similar to the average patterns of their group.  Or, alternatively, whether the averaged group patterns do reappear in replications of the study with the same or with different groups of individuals.

In some cases, function-specific patterns that characterize particular diagnostic groups are visibly similar across members of the group and clearly different from those of non-members, as in the following example summarized in Figure 12.

Fig 12 Left column: activation profiles specific to the function of reading in a group of normal children. The predominant involvement of left hemisphere structures is obvious in all cases. Right column: activation patterns specific to the same function in a group of age-matched dyslexic children showing, consistently, an aberrent pattern involving greater right hemisphere involvement. (Unpublished data from the authorís laboratory)

The same criteria for assessing the validity of the results of imaging ought to be applied with even greater vigor in cases where group membership is not as easy to establish, where the mechanism of the function in question is not as well known as in the case of language, and where the odds that the task used to represent the function is equally efficiently performed by the members of each group with comparable expenditure of effort, is not reassuringly high.  Consider, for example, the case where we wish to test the hypothesis that schizophrenic patients who presumably have difficulties with "verbal fluency" differ from normals (or some other patient group) with respect to the mechanism of that function.   Assuming that by comparing images of normals and patients, obtained in a verbal fluency task, we found that the images revealed show different amounts of activation in a particular brain area, what are we to conclude?  Should we conclude that different mechanisms mediate verbal fluency in normals and schizophrenics?  Or that the difference is simply due to the difficulty patients have to complete the task and reflects greater effort, greater frustration or less interest to engage in it?  Or should we conclude that the differences found are completely unrelated to the task and simply reflect different baseline activation profiles between the groups?

Sometimes ambiguous results are signs that the issues pursued cannot be settled, in principle, with the methods we pursue them.  Other times, they simply indicate the presence of specific, practical, and easily identifiable difficulties that can be overcome, one by one, in the course of time.

---