M. Corboz
Recent MRI studies comparing schizophrenic patients with healthy controls have provided evidence of brain-morphologic changes in the patients with schizophrenia. However, quantitative data derived from these studies indicated that the differences between patients and controls are generally slight and that there is a substantial normal variation in any of the morphometric measures. In addition to the brain-morphologic findings, various studies in the literature have reported that certain EEG characteristics discriminate between schizophrenic patients and controls. Amongst these characteristics, EEG asymmetries play a prominent role. In this context the question has been arised of whether such asymmetries are artifactual, since the comparison of EEG potentials from different hemispheres implicitly assume that (1) the potentials derived from the scalps reflect the activity of the brain spot immediately underneath, (2) the anatomy of the two hemispheres is genuinely identical, and (3) the system of electrode placements in homotopic sites is unimpeachable. None of these conditions seems to be fulfilled with the International 10-20 system, so that inferences made about the state of the 2 bilateral compared areas appear problematic.
The principal goal of this project is to solve the MRI-EEG matching problem and to control for the adequacy as well as for the accuracy of electrode placements by means of structured brain-imaging techniques. Upon the successful solution of the above problems we will be able to estimate in each individual case the extent to which abnormal EEG lateralization patterns are explained by (1) the actual deviations from the « ideal » electrode placements, and (2) the actual deviations from the « ideal » hemispheric anatomy. To this purpose we will develop a package of computer programs that meet the requirements of routine applications and interface with our MRI-EEG database and with the standard AVS platform.
Recent neuroradiological and neuropathological studies using advanced Magnetic Resonance Imaging (MRI) techniques have yielded evidence that some subgroups of schizophrenic patients have slightly enlarged lateral ventricles, reduced temporal lobes, reduced corpus callosum size, subtle abnormalities in the region of the anterior hippocampus and affected thalamus (Gur et al. 1991, Youg et al. 1991, Günther et al. 1991, Jernigan et al. 1991, Delisi et al. 1992, Di Michele et al. 1992, Liebermann et al. 1992, Rossi et al. 1992, Shenton et al. 1992, Weinberger et al. 1992, Jurjus et al. 1993, Woodruf et al. 1993, Torrey et al. 1993/1994, Andreason et al. 1994, David 1994). However, quantitative data derived from these studies indicated that the differences between patients and controls in any of the morphometric measures were slight, and that there were substantial normal variations making it difficult to distinguish between "natural" fluctuations and significant changes. In order to reduce variations between patients and controls, and to enhance the power of morphometric measures to discriminate subtle neuropathological deviations, an investigation of the anatomical abnormalities in the brains of 40 monozygotic twin pairs concordant (N=13) or discordant (N=27) for schizophrenia has been carried out at the Neuroscience Center at St. Elizabeth Hospital, Washington DC (Torrey et al. 1994), using MRI techniques. Since monozygotic twins share a common genome and, if reared together, similar socio-economic, developmental and psychological backgrounds, the degree to which such factors might contribute to brain morphological variance is greatly reduced (Gottesman 1991).
The principal MRI findings showed that evidence of anatomical changes in the brain was present in almost every twin with schizophrenia, and that these changes appeared, at least in part, to be non-genetic. Significant quantitative differences between affected and unaffected twins were found in the lateral ventricles , third ventricle, and temporal lobe, including both side of the anterior hippocampus, and in the total volume of the gray matter in the left temporal lobe (Suddath et al. 1990, Casanova et al. 1990, Zigun et al. 1992, Weinberger et al. 1992, Torrey et al. 1993).
As a direct consequence of such anatomical and neurological abnormalities, changes in brain waves pattern (EEG) can be expected, at least in the respective subgroup of schizophrenic patients. Quantitative analyses of EEG recordings of schizophrenic patients have demonstrated reduced alpha band activity, increased alpha band width and an alpha power shift toward faster frequencies (e.g. Abenson 1970; Fenton et al. 1980; Merrin et Floyd 1992). However, these findings are not consistent across studies and there is some evidence that subgroups of schizophrenics may differ in overall or lateralized levels of alpha activity. An increase in alpha activity in the frontal regions in schizophrenics has been recently reported by Kahn et al. (1993), who pointed out that this finding is consistent with data from neuropsychological tests, metabolic imaging studies and evoked potential studies, thus suggesting impaired activation of frontal brain areas in patients with schizophrenia.
On the other hand, schizophrenic patients show, in comparison to healthy controls, a generalized pattern of increased delta and theta and of decreased alpha activity (Sponheim et al. 1994, Clementz et al. 1994). There were no significant differences between first episode and chronic schizophrenics. Manic, schizophrenic and depressed patients have been compared with healthy controls (Koles et al. 1994) by means of a multivariate spatial pattern method specifically designed to reduce the error inherent to multiple univariate testing. The authors reported significant differences in the EEGs from all the 4 groups under investigation: a right-sided temporal hyperactivity in the depressive group, a left-sided temporal hyperactivity in the schizophrenic group and bilateral temporal hyperactivity in the manic group. The frontal sites were not involved.
Moreover, asymmetric amplitude of auditory P300 event-related potential (ERP) at lateral electrode sites are amongst the most robust findings in schizophrenia. Since a considerable number of schizophrenics also show volume reduction in the left temporal gyrus, left hypocampus and left parahippocampal gyrus, it has been hypothesized that psychophysiological abnormalities, such as ERP asymmetries, are associated with altered brain structure (Blackwood et al. 1991, McCarley et al. 1993).
EEG asymmetries seen in the resting state and during cognitive effort are commonly regarded as a sign of cerebral lateralization or cerebral dominance. Based on this assumption, comparison of EEG potentials from different areas are a standard tool of psychiatry research, that deals with normal and aberrant brain laterality. Such an approach would be appropriate if potentials derived from the scalp reflected activity of the brain spot immediately underneath, and if the anatomy of the 2 hemispheres is genuinely identical and, at least, if the system of electrode placements in homotopic sites were unimpeachable. However structural brain imaging which allows to examine the adequacy and accuracy of electrode placements revealed that none of the outlined conditions is fulfilled with the international 10-20 system, so that inferences made about the state of the 2 bilateral compared areas are problematic (Myslobodsky et al. 1991).
These principal difficulties in measuring EEG laterality need further investigation. Specifically, upon successful solution of the MRI-EEG matching problem, EEG laterality can be more reliably determined in each individual case by compensating for genuine asymmetries of brain morphology. Moreover, the MRI-EEG matching can also be expected to be a powerful tool for studies that investigate the extent to which altered brain structures are associated with neurophysiological abnormalities.
The principal goal of this project is to solve the MRI-EEG matching problem, so that one can distinguish between "true" EEG lateralization effects and those due to asymmetries of the brain and related aberrations of electrode placements according to the 10-20 system. Specifically, we aim:
(1) to develop a brain morphological database so that anatomically relevant parameters can be revealed for a quantitative assessment of the brain asymmetries. These parameters should enable us to compare the intra- and inter-individual fluctuations of morphometric measurements with the standard deviations of the MRI assessments. (2) to estimate in each individual case the extent to which abnormal EEG lateralization patterns can be explained by deviations from the « ideal » electrode placements or by deviations from the « ideal » hemispheric anatomy. (3) to explore potential correlations between EEG and brain morphology.
The reproducibility of morphometric measures is a crucial point in prospective studies evaluating the progression of brain value loss (Hell et al. 1995). In view of the almost complete lack of such studies on even basic reference data, a normative MRI/EEG study has been started in 1995 under the auspices of Prof. D. Hell, PD Dr. H.H. Stassen and Dr. M. Kirsten-Krueger from the Psychiatric University Hospital Zurich (PUK) and of the PD Dr. W. Wichmann from the Department of Neuroradiology of the University Hospital Zurich.
Today, a database comprising a total of 54 healthy subjects (27 males and 27 females ranging in age between 20 and 35 years) and 13 monozygotic healthy twin pairs is available. For each subject we have 2 repeated assessments at 14-days intervals including EEG recordings (10-16 channels) and MRI images (SPGR technique 1.5 mm sagittal slices and dual-echo technique 3 mm axial slices).This normative study has not only provided data on the inter-individual scattering of morphometric and EEG measures in the general population, it has also enabled us to distinguish between natural fluctuation and significant changes in prospective studies with repeated assessments of the same individual.
MRI assessments are done at the University Hospital Zurich with a 1.5 Tesla scanner [General Electric Medical Systems, Milwaukee, WI]. We have both 3D-SPGR images of 124 sagittal slices with TR 34 / TE 3 (vox. 0.9375/0.9375/1.5 [mm]) and 3D-dual-echo images of 60 axial slices with TR 2500 / Dual-TE 30/70 with (vox. 0.9375/0.9375/3.0 [mm]). Our standard image processing software is the Application Visualization System [AVS, Advanced Visual System Inc., Waltham, MA].
The collaboration with the MRI Department of the Brigham & Women’s Hospital Boston gives us the opportunity to use the most recent upgrade version of their image analysis software MRX. In the MRX package a new "adaptive segmentation" method for MRI data is implemented. This segmentation is based on the expectation-maximization (EM) algorithm that uses simultaneous convergences to estimate tissue class and correcting gain field (Wells et al. 1994/1995). The adaptive segmentation method should lead to the correction of inter-scan intensity inhomogeneities due to the operating conditions so that a better reproducibility of segmentation of tissue type through intra- and inter-individual assessments can be expected.
All EEG recordings are carried out using 16 parallel EEG channels (some earlier recordings used 10 channels) of the International 10-20 system (FP1, FP2 , ...., T5, T6). The recording procedure encompasses 3 minutes of EEG time series. The experimental condition is "eyes closed" and "quiet wakefulness". Each time series is subdivided into epochs of 20 seconds length and spectral-analyzed at a resolution of 0.25 Hz over the frequency range of 0-32 Hz.
EEG parameters (absolute power, relative power, centroid, symmetry, peak amplitude, peak frequency) are extracted by means of the program package MASTER.EEG. This program and its database provide normative data on the long-term stability of brain wave patterns. Thus, it is possible to distinguish between « natural » fluctuations and « significant » EEG changes.
To determine the distribution of electrical activity over the scalp electrodes must be placed in a standardized, reproducible scheme. To this purpose the International "10-20 system" for electrode placements has been proposed in 1958 by a committee of the International Federation of Societies for EEG (Binnie et al. 1982). Today the International 10-20 system is generally accepted and has become a de facto standard.
Symmetry about the sagittal plane is one of the crucial assumptions of the 10-20 system. Additional problems are (1) the position measurement relative to standard landmarks on the scalp and with respect to skull size and shape, (2) an adequate coverage of all parts of the head and (3) equally spaced electrodes along antero-posterior and transversal axes in order to ensure equal inter-electrode distances in bipolar chains.
It is quite easy to reconstruct with good precision the scalp of each patient on the basis of 3D-SPGR assessments. To this purpose, sagittal slices are better suited than axial ones, since the 10-20 system is based on symmetry about the sagittal plane and the inter-slice distance is only 1.5[mm]. We will develop a semi-automatic procedure to set the standard landmarks on the scalp (nasion, inion and the right and left pre-auricular points) and to derive directly from it the standard positions of all electrodes.
We are interested in the graphical reconstruction of the « ideal » 10-20 system independently of real positions of the electrodes during the assessments. However we will assume that the variations of the electrode positions are small when compared with asymmetries related to the real brain positions. With respect to left-right EEG differences, we will correlate brain symmetries or asymmetries with the positions and the signal differences of each pair of symmetrical electrodes (e.g. O1-O2). This will be done independently for each individual, each assessment and each electrode pair.
In view of future assessments we plan to adapt the Polehmus tracer system [Isotrak], that will allow us to measure the exact electrode positions within each EEG recording so that a direct comparison between « ideal » and « real » electrodes positions becomes possible. In addition this method enables the application of techniques such as brain mapping that can be used for cross-validation of results. In this question, we are in contact with the Functional Brain Mapping Laboratory of the Neurology Department of the University Hospital Geneva where a powerful tomographic EEG method called LORETA (Pasqual-Marqui et al. 1994) has been developed.
Sagittal, axial and coronal slices are required to quantify mesial brain structures (Coppola et al. 1995). We will use SPGR and dual-echo assessments simultaneously to improve the quality of 3D brain reconstruction and subsequent reslicing. The corpus callosum extremities and the center of the pons will serve us as intracranial landmarks to establish a standard orthogonal reference system and to generate a resliced 3D-volume in a variety of ways. In our perspective study designed to assess the time development of subtle brain morphologic abnormalities associated with mental illness, a semi-automatic procedure will be used for the structural recognition of various areas of the brain under standardized conditions. The anatomic structures that are of principal interest with respect to psychiatric disorders are: (1) lateral ventricles, (2) third ventricles, (3) hippocampus, (4) amygdala, (5) thalamus (6) corpus callosum and (7) temporal lobes.
A principal problem is the definition of measures that reliably describe brain asymmetries. Some are purely geometric as for example the plagiocephaly indexes (Myslobodsky et al. 1989) or the radius of gyration (Bullmore et al. 1995).
The evidence for a close relationship between EEG scalp potentials and brain morphology is based on the general assumption that there exists a relationship between the "power" of sources and the "intensity" of received signals. Brain mapping techniques (e.g. Kavanagh 1972, Rao et al. 1973, Remond 1977, Nunez 1981, Titterington et al. 1985, Lehmann et al. 1987, Pasqual-Marqui et al. 1988, Torello 1989, Ioannides et al. 1990, Lehmann et al. 1990, Maurer et al. 1991, Greenblatt 1993, Pasqual-Marqui et al. 1994, Skrandies 1994/1995, Montgomery et al. 1995, Wilson et al. 1995) offer us many solutions to model charge distributions between selected brain voxels and electrodes. In such models each voxel has a specific charge depending on its tissue type and its brain functionality. The selection of the voxels is parametrized by the definition of the sensitive brain area of each electrode (e.g. whole brain, respective hemisphere, nearest cortical structures or simple perpendicular slice cut relative to the electrode pair).
On the other hand some areas hypothesized to be associated with mental illness are deep central brain structures that will not be revealed by our asymmetry approach(Günther et al. 1991, Jurjus et al. 1993, Woodruff et al. 1993, Torrey et al. 1994, Andreasen et al. 1994).
Although it is relatively easy to reconstruct the scalp surface from MRI images, the automatic reconstruction of the electrode placements is much more complicated. In fact, electrode positions are determined by measuring four standard landmarks on the scalp: (1) nasion, (2) inion and the (3) right and (4) left pre-auricular points. Consequently, a major pitfall of any technique of EEG electrode placement is the presence of inconspicuous cranial deformities that cannot be derived from the currently used cranial landmarks and measurements (Binnie et al. 1982, Myslobodsky et al. 1989). Thus a fully automated computerized method may not be realistic.
Ideally, unbiased procedures are required to analyze MRI data, especially with respect to data collected under standardized conditions. The first advantage of an unsupervised procedure is the ability to run as a batch job without any tedious manual interaction. Second and most important, the automated parameter learning is fully reproducible because variations created by human interaction during the training of the tissues are avoided. Adaptive segmentation has proven to be an effective fully-automatic means of segmenting brain tissue in a study including more than 1000 brain scans (Wells et al. 1994). Correcting intra-scan and inter-scan intensity inhomogeneities increases the robustness and automation level available for the segmentation, and also facilitates further processing of MRI images.
The basic idea of our resclicing algorithms is the development of an anatomical standard reference system based on mesial structures like corpus callosum and pons. As shown in some references (Coppola et al. 1995) the automated recognition of these intracranial landmarks will require a lot of prospective work, and it is planned as a second step of the project. On the other side, the slice convolution adjustment seems to be relatively easy to be achieved in an automated procedure. The development of such a standard observation system will allow us to make cross-comparisons and to test the reproducibility of MRI assessments.
A principal goal of the Zurich normative is the development of methods that enable an intra-individually reproducible assessment of morphologic structures. In particular, we aim to distinguish between natural fluctuations and significant pathological changes. Our anatomical database will provide means for detection of time-invariant, presumably genetically determined, features on the base of repeated assessments at 14 day intervals and healthy monozygotic twins pairs, similar to what has been achieved for EEG recordings with the MASTER.EEG system.
The result of this normative study will provide data on the inter-individual scattering of morphometric measures in the general population (Hell et al. 1995). We will apply cluster analysis in order to derive a classification of brain morphology in terms of intrinsic groupings. However, the inter-individual classification will not necessarily lead to the same subtyping for MRI assessments as for the EEG records (Stassen 1992).
Under the assumption that abnormalities at lateral electrodes are the most robust findings in schizophrenia, we will carry out some multi-correlation processing of the EEG signals through the anatomic information over each opposite symmetrical electrode pair, over each assessment and over each individual control of our database. The main goal of our project is to develop a model that explains, at least to some extent, the observed EEG asymmetries on the basis of morphologic quantities.
We will define the most optimal parameters to ensure the most effective correction of EEG left-right differences due to morphological asymmetries of the brain and electrode misplacements. Thus the power-intensity approach will provide a mathematical weight to all electrodes that would be normalized relatively to each electrode pair or relatively to the 10-20 system as a whole. A potential next step could be to project our results on a ideal brain and on a electrode reference model. Such probes have proven (Andreason et al. 1994) to detect brain abnormalities with respect to an "average" normal brain.
If you have questions or comments concerning the research projects listed above send e-mail to one of the following addresses: