The metabolic and functional mapping of the cortical surface is a particularly important goal of current imaging paradigms. The understanding of the relationship between the measured characteristics of the brain and its behavioral response requires simultaneous information from widely spatially—distributed anatomic areas. This in turn requires the analysis of multiple images and multiple measurement modes, and 3-D display and analysis. The cortical gyri provide important landmarks for the identification and interpretation of brain functions and behavioral deficits. However, current techniques for reconstruction of the cortical topography (gyri and sulci) are generally of low precision. Improved reconstruction, display and analysis techniques using volume—sampled co—registered functional and structural images have been developed.

These techniques provide a new tool for developing behavioral pattern vectors indicative of brain condition and state. The cortical surface metabolic mapping techniques have been used to investigate continuous performance tasks in normal and schizophrenic subjects, on and off neuroleptics.

The investigation of the function and state of the brain is a complex task, combining multi—disciplinary capabilities and techniques drawn from neuroscience, with functionally coupled stable behavioral experimental paradigms and advanced digital imaging techniques. Significant research has been applied to the complex interactions between and among the various functional units of the brain to explain both normal and pathologic behaviors. Mountcastle(1977) postulated that the higher functions of the brain depend on the ensemble action of large populations of nerves in the forebrain. Although phylogenetically older units of the brain play important roles in its functioning, higher brain functions are dependent on the unique characteristics of the cerebral cortex.

The metabolic activity patterns (MAPs) exhibited by the brain reflect neuronal work. Increased brain activity induces the flow of electrical impulses to various regions of the brain, which increases the amount of glucose used in the local synapses in these regions. Glucose is the energy substrate used to create high—energy phosphate bonds necessary for ionic gradient maintenance across neural membranes. Specific tasks increase the electrical activity to those brain regions directly related to the task, resulting in higher glucose metabolism. Other regions, however, not specifically associated with that task, can also exhibit significant glucose metabolism.

Based on its primary role of receiving sensory information, assessing that information and acting on those data in terms of motor activity, the cortex plays a primary role in "determining" behavior. To the extent that a pathological process has modified a behavior and a brain's structure, it will also change cortical metabolic activity pattern (CMAP). Surface projections of the brain's (CMAP) provide an ideal template for analyzing and modeling brain activity dynamics.

Electrophysiology, neuropsychology, primate anatomical connection tracing and postmortem autopsy studies have shown that the cortical surface's gyri and sulci divide the cortex into functional regions. By superimposing functional (PET, SPECT) images onto high—resolution anatomical images (MRI, CT, etc.), clinicians and research investigators can assign functional importance to a particular gyrus or sulcal landmark. Using surface landmarks one can make maximum use of ancillary anatomical information.

Various researchers have experimentally supported the existence and stability of metabolic pattern maps. Moeller (1987) has reported on experiments which have indicated that the effects of neurologic diseases are exhibited as consistent patterns in metabolic profiles measured by PET. In addition, Haxby (1985), Foster (1986) and Young (1986) have shown that simple rCMRglu (regional cerebral metabolic glucose rates) are significantly correlated with neuropsychological test scores. Significant regional metabolic correlation and covariance patterns have also been observed by Clark (1984), Moeller (1987), Horowitz (1984), Metter (1985) and Volkow (1986).

The change in a subject's cortical surface metabolic activity pattern (CMAP) when transitioning from one level of difficulty to another in a single task or from one type of task to another will be similar to other subjects with normal physiobehavioral dynamics. Pathological processes, however, could possibly severely modify these dynamic relationships. The cortical surface MAP of the brain can be useful in evaluating altered or defective brain conditions or functions.

The cortical surface anatomy provides a topographic map of the metabolic activity related to specific tasks. Stable, carefully designed behavioral tasks can be designed to isolate various neurologic circuits in the brain based on metabolic response. The cortical surface metabolic map indicates the metabolic state of the brain. The state of the brain under different behavioral paradigms is indicative of the condition of the brain.

A common problem related to functional image analysis is the absence of anatomical landmarks. This makes the estimation of volumetric functional data difficult and subject to significant inter—observer error and interpretational differences. The problem is also linked to the inherent limitations associated with the generalization of three—dimensional shapes from conventional two—dimensional images–it is difficult to know what data out of the plane but proximal to that being viewed "looks" like. This prevents the researcher from identifying the anatomical substrate of a local measurement and from knowing what is happening in regions close to the area in a given two-dimensional plane. The cortex is a thin (approximately 2 mm) sheet which follows a highly convoluted surface, making it difficult to define on relatively poor resolution planar PET images. To overcome this deficiency, a technique has been developed that uses the high resolution MR images, which can easily be acquired as contiguous slices, to act as an accurate structural template.

Cortical surface displays overlaying anatomical and metabolic images, circumvent these problems. Magnetic resonance images (MRI) provide high—contrast, highly—resolved separation of gray, white, and spinal fluid compartments and are ideal substrates for anatomical localization. However, two—dimensional MR images alone do not permit easy, unambiguous identification of cortical surface landmarks. Positron emission tomography (PET) data are of low resolution, and are noisy and sensitive to distant (and local) functional perturbations.

Registered single axis serial image data sets, PET and MRI, are used for cortical surface projection. Each primary data set spans the total brain in at least one axis. The data set used to provide structural anatomy is MRI. The primary functional (metabolic) information data comes from PET. Cortical surfaces are constructed for various surface orientations–right and left sagittal, frontal, transaxial and occipital. The various views are required to isolate specific gyral patterns that relate to specific behavioral tests (e.g. the left sagittal view shows the auditory areas).

A specific cortical surface view is constructed from an appropriate orthogonal data set. The process creates a nested set of annular regions, each of which contains the appropriate ring of cortical grey matter. Successive rings are added together (logical AND) producing a cortical surface image for each image modality. These images are self-registered.

Functionally and anatomically, the cortex is demarked by highly convoluted surfaces and organized into hemispheres, lobes, gyri and sulci. These regions interact in a complex manner to initiate and maintain complex behaviors, giving rise to unique metabolic patterns which are characteristic of a particular subject class and behavioral task combination. The cortex additionally exists as a thin (approximately 2 mm) sheet which follows the convoluted surface, making it difficult to define on relatively poor resolution planar PET images. To overcome this deficiency, a technique has been developed that uses the high resolution MRI images, which can easily be acquired as contiguous slices, to act as an accurate region of interest (ROI) template.

The metabolic pattern associated with the performance of specific tasks can be measured by PET techniques using deoxyglucose and blood flow a metabolic marker. This technique is also applicable to other functional data such as blood flow. Abnormal conditions can be assessed by determining the characteristic metabolic patterns corresponding to behaviorally-induced metabolic brain states. A key element in this assessment is the design of significant behavioral tasks which are constructed to isolate the cortical projections of condition-sensitive circuitry. For example, it is possible to differentiate between normal and schizophrenic patients, using behavioral tasks which are attention and accuracy dependent since these defects have been associated with schizophrenia.

Anatomical correlation is used to isolate the metabolic pattern of the cortical surface. The spatial distribution of metabolic information is interpreted relative to the cortical gyri determined from registered MRI images. Cortical surface reconstruction of the PET and MRI planar data sets facilitates the development of pattern vectors. The cortical surface is divided by gyri and sub—gyri based on a standard atlas. The continuous metabolic data is analyzed in absolute units converted to binary values based on the average and standard deviation of the metabolic values developed for the cortical grey matter. The cortical grey matter is isolated for the entire brain of each subject using thresholding or eigenimage reduction of the MRI images.

The metabolic map can be transformed into a metabolic pattern vector related to the metabolic profile of the brain based on the functional areas isolated on the cortex. Elements of the pattern vector refer to specific anatomical—functional areas of the cortex. Both absolute and normalized values of metabolic activity can be used. Normalized pattern vectors are based on z—scores derived from the average values of the total gray matter to normalize the results. Cortical surface z—score levels are based on the average and standard deviation of the metabolic values for the total cortical grey matter of each subject derived from analysis of the individual slices. The anatomical identification of gyri can be accomplished using a standard atlas.

The z—score pattern vector is formed by subtracting the mean metabolic value from individual values and dividing by the standard deviation of the metabolic activity. The z—scores are transformed into binary values with the following transformation. Regions with average z—scores less than -2s are encoded as binary -1. Regions with z—scores greater than -2s and less than +2s are encoded as 0. Regions with values greater than +2sare encoded as +1. Binary encoding of the metabolic pattern vectors allows different subjects to be directly compared. This technique also allows the development of group data sets since the pattern vectors are independent of metabolic variation between subjects and are also independent of the variation in the shape, size and location of individual gyri. Differences between pattern vectors for individual subjects or conditions can be measured and easily assessed from the "Hamming" distance for the binary case.

Metabolic pattern vector allows two important capabilities. The first is that individual data can be combined into group data sets. This is possible since the state vectors are independent of the size and location of specific gyri. The state vector is derived from vector element averages. Membership in the group is assigned based upon the statistics of the vector set. The second capability is related to the potential use of a neural network to classify the vectors. The MAP can be reduced to a vector whose elements are the binarized z—score values for each cortical subregion. These vectors can be processed through a multilayer neural network whose inputs are the vector elements and whose outputs are the disease class assignment. Neural networks can be trained on large data sets consisting of single behavioral paradigms and single patient population classes such as normals or schizophrenics with depressive disorders, on and off drugs.

Because of its massively parallel structure, the cortex can exhibit a wide range of MAPs in association with subtle behaviors, mild illnesses or transient behavioral states. Only when a condition is persistent, repetitive, simple and actively pursued, is it likely to create a similar brain MAP in many individuals. Physiologically appropriate conditions can magnify and clarify neuropathology. For example, passive displacement of a limb will amplify the abnormal metabolic activity pattern (MAP) of Parkinsonism. When a subject performs the same test or a variation on that test (same angle judgments versus same length or same location of objects), it is possible to determine whether the appropriate regions exhibit expected correlations. When the patient is judging whether a pair of angles are similar, the resulting brain MAP should exhibit a consistent pattern of interregional correlations (parietal, occipital, and temporal). Some regions should also have positive or negative correlations with task performance. To the extent that these relationships are preserved, one may make judicious decisions regarding the normality or health of the subject's nervous system.

There are three principal rules to be followed in behaviorally linked metabolic mapping: (1) the subject must be engaged in a task that requires cognitive and motor responses, (2) tasks should be behaviorally stable for the subject and (3) tasks should be used that effectively utilize brain regions of interest for the subject's illness. This last rule is both important and complex. Most of the brain is activated by consciousness alone. Changes in posture or anxiety level have large global effects on the brain MAPs of many subjects. By using repetitive, simple tasks the investigator can "direct" or "track" the range of possible brain MAPs associated with an individual's pathology.

The time intervals separating the individual studies should be minimized when the investigator is attempting to use a series of MAPs to create a comprehensive superMAP (a series of sets). When using week to month intervals between testing sets as a baseline, a control study should be included. When it is possible, repeat studies are included to assess intra—individual variability. This is particularly important when an illness is progressing or robust interventions (surgery) have occurred between study sets.

Neuroleptics are postulated to control antipsychotic action through dopamine receptor blockade. Neuroleptics exert both acute and chronic effects on the dopamine system and on other neural systems. Even after significant research, questions still exist on the exact mechanisms and areas of action of neuroleptics. The effects of neuroleptics on the regional cerebral metabolism of glucose (rCMRglu) was measured from images derived from positron emission tomography (PET) experiments with fluorodeoxyglucose (FDG). Neuroleptics increase glucose utilization in almost all brain regions. These increases occur in both dopamine rich and dopamine poor regions.

Cognitive deficits occur in schizophrenics and improve with neuroleptic treatment. Schizophrenics exhibit cognitive impairment when performing information processing tasks. Neuroleptics appear to counter these defects by enhancing the patient's concentration and attentiveness and increasing the accuracy of perceptual judgment.

It has been shown that schizophrenics tend to perform poorly on continuous vigilance tasks. This performance deficit has been localized to the middle prefrontal cortical pole in PET/FDG studies by Cohen et al. Additionally, it has been noted that neuroleptic treatment tends to mitigate this performance deficit. Both normal controls and medicated schizophrenic patients exhibit significant correlation between metabolic uptake and performance. These studies indicate hyperactivity in the medial, right anterior, right posterior and left anterior frontal cortex in a continuous performance auditory tone discrimination task. Metabolic hypoactivity was exhibited in the cingulate gyrus and superior parietal cortex. Nontreated schizophrenics showed hypoactivity in the frontal areas in which the normals and treated schizophrenics showed elevated metabolism.

Mesulam, using electrophysiologic probes, has identified the posterior parietal cortex (sensory component), the middle frontal cortex (motor/output) and the cingulate cortex (motivational/attention) as significantly contributing to basic vigilance and arousal status organized at the brain stem level. These observations have been supported by cerebral blood flow measurements by Roland. Pure auditory tasks stimulate the auditory cortex and the auditory association areas. Pure visual discrimination tasks stimulate the posterior superior parietal cortex and the visual association areas. Certain cortical areas are commonly stimulated by mixed modality (auditory and visual) discrimination tasks: the parieto-temporal cortex and the anterior and mid-frontal cortex. These results led to the hypothesis (Tamminga and Holcomb) that in the human brain, attention is related to a neural circuit with the following principal components: cortical areas (middle frontal, inferior posterior parietal and cingulate), and subcortical/grey matter areas in the basal ganglia (caudate, globus pallidus, substantia nigra) and the thalamus.

PET Imaging Protocol: A series of PET/FDG experiments on normal and schizophrenics were performed with repeated measures for each subject. The subject is seated in a comfortable chair in a quiet room in close proximity to the PET imaging facility. Venous blood sampling is performed on a specified time schedule. Thirty minutes prior to tracer infusion a 3 milliliter blood sample is withdrawn for measurement of endogenous glucose. This sample is repeated at specified intervals following tracer injection. Task performance is initiated 1-2 minutes before tracer injection.

Rapid sampling for 18F-2DG follows a 30 second tracer injection. The "critical" uptake phase of the procedure is concluded 30 minutes after tracer injection. Sampling for FDG tracer continues throughout the duration of the scan. Figure 1 shows the images collected for an individual experiment for a normal subject performing the auditory CPT. The PET machine delivers a set of 12 transaxial slices parallel to the AC-PC plane. The resolution of the PET scanner is 2.774 mm FWHM (Full Width Half Maximum) in-plane and 8 mm FWHM in the z axis.

Fig. 1

Image Registration: Registered MRI and PET images are obtained for each subject session using the registration technique of Holcomb and Loats. The registered MRI images are used to develop an anatomical region of interest sampling atlas for the definition of cortical surface anatomical regions. Registration between the MRI and PET image modalities is accomplished using a thermolabile custom—fitted mask with fitted fiducial markers which appear in each image modality. The images are acquired in planes parallel to the AC-PC plane and therefore can be referenced to standard stereotaxic coordinates. The fiducials for the MRI are evident as small dots on some of the MR images.

Image Reconstruction: PET images are reconstructed using filtered back-projection and a high resolution Shepp-Logan filter. An ellipse is fitted around the outer edge of the scalp and a uniform 0.88/cm attenuation coefficient is assumed within the ellipse. The projection data is corrected for attenuation and the images are reconstructed a second time. The Phelps-Huang modification of the Sokoloff equation is used to convert PET numbers to local cerebral metabolic glucose rate (LCMRglu). The collection of complete data sets in each of the three principal axes is possible with MRI. However, the collection of contiguous data sets in all three principal axes is both costly and time—consuming. A single series of MR transaxial images which span the brain are collected.

Reconstruction of the PET and MRI data sets is accomplished by computer digital resampling. The PET digital image is presented as a 100 x 100 matrix of 16 bit numbers having a resolution of 2.77 x 2.77 mm / pixel in-plane. It is resampled to match the 256 x 256 display of the MRI. Contiguous slices with either 3 or 5 mm spacing are collected in the transaxial direction parallel to the AC-PC line to match the PET image acquisition. Figures 2 and 3 show the sagittal and coronal data sets reconstructed from the original 12 transaxial images.

Fig. 2

Fig. 3

Registered Cortical Surface Reconstruction: The fiducially controlled MRI and PET image data sets are resampled and produced as uniformly sampled volume data sets, from which sagittal surface images are reconstructed. A combination of linear and splined interpolation is used for resampling to adjust both image data sets to a common size, and at the same time, to make a final correction for residual roll, pitch, yaw and translation offsets. Appropriate cortical surfaces are constructed by the techniques described previously based on these resampled volume data sets. The image sets are registered and resized to produce a comparable set of uniformly sampled data as shown in Figures 4, 5 and 6.

Fig. 4

Fig. 5

Fig. 6

Cortical Surface Images: Cortical surface images for both PET and MRI are constructed by superimposing serial images of the appropriate orientation. The slices of the reconstructed data are arranged in successive order and reduced to an annular volume equivalent to depth of the cortical surface at the appropriate slice level. The successive rings are composited into a surface projection. Figure 7 shows the left and right surfaces created for the MRI data set illustrated above. A process of local contrast is performed on the MR images to enhance the ability of the clinician to identify landmarks. The equivalent surface images for the PET data set are shown in Figure 8.

Fig. 7

Fig. 8

The results of a test and retest paradigm for a schizophrenic subject on neuroleptics, performing a controlled continuous attention visual are shown in Figures 9 and 10.

Fig. 9

Fig. 10

Contrast this with the equivalent pairs of images shown in Figures 11 and 12 for schizophrenic subjects performing the same continuous attention visual task with a period of no neuroleptic treatment between the first and second experiment. For these experiments, the subjects were taken off neuroleptics for a period of six weeks prior to the performance of the second test. A significant change in metabolic state for both subjects is clearly evident as a significant overall reduction of metabolic activation for the off neuroleptic condition. When the subject is his own control, the pattern stability is evident and is preserved over group data.

Fig. 11

Fig. 12

Auditory Discrimination Task: A continuous tone discrimination task employing two distinct levels of 750 and 1500 Hz was used. Single, separated 1.5 second tones are presented to a blindfolded subject through binaural earphones. Testing begins 5 minutes prior to FDG injection and continues for 30 minutes. Tones are generated in a random sequence and the subject's response time and accuracy is automatically monitored. The subject's performance is controlled by changing the level and volume of white noise background. The order of "hard" and "easy" is randomized over the subject tests. In a continuous-attention auditory discrimination task, the subject's response is continuously monitored on-line during the test. The continuous monitoring allows the investigator to control the "noise" in the system during the task. Reduction in the discriminability increases the subject's uncertainty and error rate. Specified and controlled error rates permit the comparison of task difficulty and permit the group description of subjects.

The surface technique provides increased information when contrasted to slice data only. This is illustrated by the two cases of paired PET studies illustrated in Figure 13 which shows an auditory continuous performance task and contrasts the high and low error rate conditions. The subject is a 30-year old male who performed the auditory CPT on two occasions. In each instance, he pressed the right button when he heard a low frequency (750 Hz) tone and alternatively pressed a right-hand button when a higher frequency (1500 Hz) tone was presented. Performance accuracy was degraded by increasing the background white noise and decreasing target tone amplitude. In the figure, Scan 1 is a low error rate task condition and Scan 2 is the high error rate condition. Ratio images were formed by dividing Scan 2 by Scan 1. The ratio image in Figure 14 reveals a focus of significantly higher activity in the left auditory region of the temporal lobe. This represents a higher focus of activity during the more difficult CPT task.

Fig. 13

Fig. 14

Visual Discrimination CPT: An equivalent analysis was done for a visual CPT for a schizophrenic patient on and off neuroleptics. In this instance, a 24—year old male diagnosed as schizophrenic four years prior to this test performed an easy (3—5% false responses) continuous performance visual task requiring the subject to press a right—hand button when presented with a letter with an ascending stem. Alternatively, the subject is required to press the left—hand button when presented with a letter possessing a descending stem. Although able to perform the task equally well on and off neuroleptics, the subject off neuroleptics exhibited significantly reduced metabolic activity over much of the cortex. The ratio image for this task showed large focal changes in the frontal and parieto—occipital regions, consistent with multiple studies showing higher metabolic activity in the cortex of patients on neuroleptics.

The analyzed PET images can also be directly overlayed on the structural MRI's to permit the simultaneous observation of the functional activity projected onto the high-resolution MRI images. Figure 15 depicts the results for a schizophrenic subject performing the continuous attention visual task on neuroleptics for scan 1 and off neuroleptics for scan 2. These pictures illustrate the use of cortical surface image superposition to identify significant characteristics of the cortical surface metabolic patterns. Increasing z-score levels are illustrated by the color palette, ranging from a z-score level of 0 (blue) to a z-score level of greater than +1.5s (white). The z-score levels reveal a maximal activity pattern in regions which are associated with repetitive visual decisions with associated hand motor activity. From the figure it can be seen that the general pattern of activation is maintained between the two scans. At the higher z—score levels (e.g. maximal activation) the activity is centered in the middle temporal gyrus, the inferior parietal gyri and the inferior and middle frontal gyri. Scan 2 off neuroleptics shows higher activity in the parietal lobe.

Fig. 15

Discussion: The cortical surface mapping techniques developed above were designed to address an important problem in analyzing the cortex by the identification of distinct gyral regions. This definition is very difficult and nearly impossible from transaxial slices, even using the capability of redirecting the ROI from the high—resolution spatially—registered MRI images. To overcome this difficulty, the cortical surface reconstruction technique was developed. This allowed the easy identification of gyral regions from MRI surface images. The surface images are usually constructed for the primary orthogonal views (left and right sagittal, posterior and anterior). Alternatively, surface views from any angle can also be constructed to aid in sulcal or gyral landmark identification. The development of subcortical anatomic ROI's can also be performed by conventional edge—finding techniques. The techniques and stability of cortical and subcortical pattern vectors can be used for behavioral brain—state classification and analysis.

The experimental paradigms employed use the subject as his own control in a series of related behavioral tasks to elucidate the state and condition of the cortical circuitry. The technique produces a metabolic "pattern vector" including both cortical and subcortical ROI's. The pattern is stable for the same subject performing the same task and indicates significant pattern differences for that patient either performing a different task or for the patient performing the same task under a different physiologic state such as is produced in schizophrenic subjects on and off neuroleptics.

Metabolic pattern vectors are stable for the same subject performing the same task for each subject/behavioral task replication, and there are significant metabolism changes in appropriate areas related to changes in behavioral state. The extension of the technique to group data formation can be accomplished since the elements of the pattern vector are independent of the spatially variable anatomic feature location of each patient. Using the group data we are able to compute the mean and variance of the metabolic activity of individual pattern vector elements. From this, we can derive statistical measures of pattern vector element difference and also develop clusters and multivariate discriminant measures.

Metabolic patterns expressed on the cortical surface discriminate between various brain states and can be indicative of various conditions of the brain. The use of specific neuronal tracers permit the visualization of specific neuronal populations and thus can be indicative of in vivo neuronal pathways and projections. The use of PET, coupled with anatomically—based functional pattern recognition can provide a new tool for understanding the natural course of disease, the efficacy of potential treatments, and to aid in the selection of viable therapies.