Clinical Neurophysiology: Computer Analysis of EEG

1TomoyukiAkiyama,2HiroshiOtsubo,2AyakoOchi,2RajeshRamachandranNair,2IreneElliot,2ElizabethDonner,2Shelly K.Weiss,2James T.Rutka, and2O. CarterSnead III(1Department of Child Neurology, Okayama University Hospital, Okayama, Okayama, Japan; and2 Division of Neurology, The Hospital for Sick Children, Toronto, ON, Canada ) Rationale: Multiple band frequency analysis (MBFA), a new frequency analysis method, provides better frequency and temporal resolutions than fast Fourier or wavelet transform. We evaluated dynamic changes of high frequency oscillations (HFOs) on brain surface maps using MBFA to localize epileptogenic zones in neocortical epilepsy. Methods: We studied 2 children with intractable neocortical epilepsy by subdural EEG (SDEEG) at 1 KHz sampling rate. We performed MBFA for 1–5 second ictal EEG using the software Short Spectrum Eye (Gram, Saitama, Japan). We identified HFOs, determined the frequency range, and calculated averaged powers at every 10 ms for all electrodes. We imported the averaged powers into a topographic mapping program Prism and Insight (Persyst, Prescott, AZ) to visualize HFO powers on the brain surface picture. To analyze temporal interrelationships of regions with HFOs, we arranged HFO power maps at every 20 ms to project movies of ictal HFOs on the brain surface. Case 1: A 17 year old right handed girl presented with simple partial seizures consisting of numbness and tingling sensations in the right arm followed by clonic movements and secondary generalization. Case 2: A 14 year old left handed boy with asymmetric epileptic spasms in clusters. Results: Case 1; During partial seizures, MBFA of 25‐channel SDEEG showed 80–130 Hz HFOs over the middle portion of the left postcentral gyrus and upper portion of the left post‐ and pre‐central gyri. Sequential power maps of 80–130 Hz HFOs revealed high powers over these 2 regions, and reverberating power changes between them with a maximum at the middle portion of postcentral gyrus. We resected the upper and middle portions of left postcentral gyrus. She has rare and brief partial motor seizures without sensory aura and secondary generalization 9 months after surgery. Case 2; During spasms, MBFA of 106‐channel SDEEG showed extensive but noncontiguous 60–150 Hz HFOs over the right superior and inferior frontal, and middle temporal gyri. Sequential power maps of 80–120 Hz HFOs revealed initial high power over the right superior frontal gyrus and subsequent activations over other 2 regions. We performed partial right frontal lobectomy, cortical excisions of the superior and inferior frontal gyri, and middle temporal gyrus where predominant HFOs existed. He is seizure free 6 months after surgery. Conclusions: The combination of MBFA and sequential brain surface HFO power mapping enabled us to recognize the localization and dynamic changes of ictal HFOs. This spatial and temporal analysis of HFO cortical mapping may disclose the behavior of ictal networks in the cortical epileptogenic zones. 1EishiAsano,1CsabaJuhasz,1AashitShah,1OttoMuzik,1Diane C.Chugani,1SandeepSood, and1Harry T.Chugani(1 Pediatrics, Neurology, Radiology, Neurosurgery, Children's Hospital of Michigan, Wayne State University, Detroit, MI ) Rationale: Cortical tubers are the hallmark of tuberous sclerosis complex (TSC) and are usually associated with overlying and surrounding gray matter on MRI. In the present study, we have quantitatively analyzed ictal electrocorticography (ECoG) data using subdural and depth electrodes to determine whether epileptic seizures originate more frequently from glucose hypometabolic cortex surrounding a tuber than normometabolic cortex. Secondly, we asked whether epileptic seizures are generated by a cortical tuber itself or the cortex surrounding a tuber in children with TSC. Methods: We studied a series of 16 children with TSC and uncontrolled seizures (age: 5 months to 16 years; 11 boys), who underwent MRI, glucose metabolism PET and prolonged intracranial ECoG recording for subsequent cortical resection. The location of electrodes was derived from planar X‐rays and displayed on the 3D brain surface reconstructed from MRIs showing the location of tubers and also hypometabolic areas from co‐registered PET scans (Muzik et al, Neurology 2000). For each seizure event, the center of ictal discharges was objectively defined as the electrode showing the maximal increase of ECoG magnitude specific to an ictal onset in a seizure consisting of continuous rhythmic activity or the electrode showing the maximal spike magnitude in a seizure consisting of periodic spike activity or a single spike followed by fast wave bursts (Asano et al, Epilepsia 2005 [in press]). We performed a one‐sample t‐test to determine whether the center of ictal discharges was located more frequently in glucose hypometabolic regions than normometabolic regions, and whether the center of ictal discharges was located more frequently in the cortex surrounding a tuber than the cortical tuber itself. Results: A total of 70 seizures were objectively analyzed in the subjects where subdural electrodes but not depth electrodes were placed on the presumed epileptic hemisphere. In 57 out of the 70 seizures, the center of ictal discharges was located in the glucose hypometabolic cortices. Ictal discharges arose more frequently from the glucose hypometabolic cortex than from normometabolic cortex (p < 0.001). In a single subject where a depth electrode was inserted into a tuber, no spontaneous seizures were captured and we were not able to determine the location of ictal onset zones. Conclusions: These results suggest that most of the epileptic seizures in children with TSC may be generated by glucose hypometabolic regions associated with either the dysplastic cortex surrounding a tuber or the tuber itself. Further studies of patients with depth electrodes are required to determine whether epileptic seizures are generated by a cortical tuber itself or the cortex surrounding a tuber in children with TSC. (Supported by NIH: K23NS047550.) 1ChadCarlson,1CatherineSchevon,3WernerDoyle,3HowardWeiner,2JoshuaCappell,2RonaldEmerson,2LawrenceHirsch,2RobertGoodman,1OrrinDevinsky,1StevenPacia, and1RubenKuzniecky(1Department of Neurology, New York University School of Medicine, New York, NY;2Neurological Institute, Columbia College of Physicians and Surgeons, New York, NY; and3 Department of Neurosurgery, New York University School of Medicine, New York, NY ) Rationale: Traditional review of scalp and intracranial subdural video‐EEG studies rely primarily upon the visual analysis of the interictal and ictal patterns. In many patients, particularly those with multifocal or neocortical epilepsy, seizures are difficult to localize due to rapid spread of ictal activity and high frequency rhythms that may elude detection, potentially affecting the surgical outcome. Time‐frequency analysis is a conceptually simple technique that, when applied to intracranial EEG recordings, may reveal information that is not apparent to visual inspection alone. To facilitate its use, an easily interpreted visual presentation was created that provides an overview of the activity of the complete EEG montage. This case series examines the role of spectral time‐frequency analysis in five patients with medically refractory epilepsy who underwent invasive subdural electrode video‐EEG studies. Methods: Interictal and ictal recordings were studied from five patients who underwent intracranial subdural electrode studies for seizure localization as part of a presurgical evaluation. Time‐frequency spectral analysis was performed on all channels over sequential one‐second time windows. The spectrogram over the 1–200 Hz frequency range was displayed using a colormap with intensity corresponding to the logarithm of the FFT of the input signal over a one‐second time window; the colormap is fixed for a given sequence to allow comparisons over time and between channels. EEG was recorded on a Nicolet Bravo system with a sampling rate of 400 Hz/channel and bandpass filtered (4 th order Butterworth) at 0.5–125 Hz. All computations were implemented in Matlab. Results: Spectral analysis of seizure onsets are shown for the five patients and compared to the clinical interpretation. In four cases, the additional information provided confirmatory support for the clinical localization, while in the fourth case the analysis supported an alternative candidate seizure focus, an observation consistent with the surgical outcome. Conclusions: This case series illustrates the complementary role of spectral time‐frequency analysis in the setting of intracranial EEG recordings, as well as the importance of a unified presentation that permits direct comparisons between channels. The role of this technique in providing information not apparent on visual EEG interpretation and its contribution to the evaluation of complex neocortical epilepsies, as well as its potential effect on surgical outcome, warrants further assessment. (Supported by FACES.) 1Wanpracha A.Chaovalitwongse,2Rajesh C.Sachdeo,3Panos M.Pardalos,4Leonidas D.Iasemidis, and5J. C.Sackellares(1Industrial and Systems Engineering, Rutgers University, Piscataway, NJ;2Comprehensive Epilepsy Center, St. Peter's University Hospital, New Brunswick, NJ;3Industrial and Systems Engineering, University of Florida, Gainesville, FL;4Bioengineering, Arizona State University, Tempe, AZ; and5 Biomedical Engineering, University of Florida, Gainesville, FL ) Rationale: At least 2 million people in the U.S. (1% of population) currently suffer from epilepsy. The most disabling aspect of epilepsy is recurrent seizures, in which the majority of epileptic patients (at least 1 million) feel that they get inadequate treatments or seizures cannot be controlled by available treatments. It is very clear that improved treatments are desperately needed. In this study, we herein propose novel quantitative approaches that can detect abnormalities in the brain's electrical activity. The results of this study will pave our way to the development of an Automated Brain Activity Classifier, which is a prerequisite of seizure prediction process. Methods: Continuous 26‐channel long‐term intracranial EEG recordings previously obtained in 3 patients with medically intractable partial seizures were used to test the automated brain activity classifier. Patient 1 had 15 seizures in a 10‐day recording; patient 2 had 8 seizures in 6 days; patient 3 had 7 seizures in 12 days. The automated brain activity classifier involved the following steps: (1) quantify the chaoticity properties (i.e., Lyapunov exponents, angular frequency, entropy) of the brain dynamics, (2a) use the statistical cross validation technique to estimate statistical distances between an EEG epoch and the brain activity from different physiological states (normal, pre‐seizure, post‐seizure), (2b) use optimization techniques to find support vector machines to separate different brain physiological states, (3) classify the EEG epoch to the physiological state of the most brain activity (smallest statistical distance). Results: The sensitivities of the statistical cross validation approach in classifying pre‐seizure EEG's in patients 1, 2, and 3 were 89.39%, 85.71%, and 84.44%, respectively and the sensitivities in classifying normal EEG's in patients 1, 2, and 3 were 93.50%, 78.00%, and 75.00%, respectively. The sensitivities of the support vector machines approach in classifying pre‐seizure EEG's in patients 1, 2, and 3 were 81.21%, 71.18%, and 74.13%, respectively and the sensitivities in classifying normal EEG's in patients 1, 2, and 3 were 87.46%, 76.85%, and 70.60%, respectively. Conclusions: Based on the proposed statistical classification approaches, this automated brain activity classifier can classify EEG epochs into the accurate brain physiological state with performance characteristics that could have practical clinical utility. The classifier could be incorporated into clinical EEG monitoring system or, incorporated into a seizure warning system used to activate timed physiological or pharmacological interventions. 1LeiDing,2Gregory A.Worrell,2Terrence D.Lagerlund, and1BinHe(1Department of Biomedical Engineering, University of Minnesota; and2 Department of Neurology, Mayo Clinic ) Rationale: Substantial interest exists in the development of noninvasive localization of epileptogenic foci, since the current standard utilization of subdural electrocorticography and depth electrodes requires surgical implantation with some additional risks to the patient. The aim of the present study is to evaluate a novel noninvasive source localization approach for localizing epileptogenic foci from scalp EEG. Methods: The traditional dipole source localization approach faces problems for multiple dipole source localization as the number of sources increases. We recently developed a new subspace source localization approach, i.e. three‐dimensional first principle vectors (3D‐FINE), for EEG‐based epileptic focus localization (Xu et al, Phys Med & Bio, 49(2):327–343, 2004). The approach uses a three‐dimensional search instead of solving the extremely complicated nonlinear optimization problem. It employs the projection onto a particular set of vectors in noise‐only subspace instead of entire noise‐only subspace to achieve high accuracy and high spatial resolvability. Computer simulation studies were conducted to test the feasibility and performance of 3D‐FINE.Three patients, from the Mayo Clinic, with medically intractable partial epilepsy and symptomatic lesion demonstrated by the magnetic resonance image (MRI) were analyzed by 3D‐FINE using interictal data. The patient's EEGs were recorded using 31 electrodes in the modified 10/20 system and each patient had a standardized seizure protocol MRI. The interictal data was identified and selected by the experienced neurologists. The results from the 3D‐FINE approach were evaluated and validated by the brain lesions demonstrated on the MRI. Results: Both computer simulations and interictal data analysis indicate the higher spatial resolution and higher source localization accuracy of 3D‐FINE in multiple source localization in comparison with the well established MUSIC approach, which suggests that 3D‐FINE has the greater ability to distinguish closely‐spaced sources. The identified sources from both MUSIC and 3D‐FINE are well correlated with the MRI lesions. They appear on the border or in the vicinity of the MRI lesion. The detailed electrical activity distributions during the interictal spikes are revealed by 3D‐FINE suggesting the excellent performance of 3D‐FINE in localizing epileptiform activities from scalp EEG. Conclusions: We tested the feasibility of imaging epileptic foci using a novel multiple dipole source localization approach, i.e. 3D‐FINE. The patients studied are with MRI lesions because it provides a unique way to cross validate the epileptic foci localization results using different imaging modalities. However, the usefulness of such technique is to image epileptic foci in epilepsy patient without explicit MRI lesions, where MRI technique may fail to identify the epileptic foci. (Supported by NIH EB 00178 and NSF BES‐0411898.) 1Joshua B.Ewen,2Nathan E.Crone,3Eric H.Kossoff,3Laura A.Hatfield,3Thomas M.Kelley, and3Anne M.Comi(1Neurology and Developmental Medicine, Kennedy Krieger Institute, Baltimore, MD;2Neurology, Johns Hopkins University School of Medicine, Baltimore, MD; and3 Neurology and Pediatrics, Johns Hopkins University School of Medicine, Baltimore, MD ) Rationale: Ten to twenty percent of children with facial port‐wine stain (PWS) develop neurologic features characteristic of Sturge‐Weber syndrome (SWS), which may include seizures, visual field defects, hemiparesis, and cognitive impairment. Screening often involves serial neuroimaging, which carries a sedation risk and is often falsely negative in infants. We have reported that decreased power on the affected side correlated with neurologic impairments in subjects with SWS, and efforts are ongoing to develop quantitative EEG (qEEG) as a tool for early diagnosis in SWS. Methods: Three infants with unilateral facial port‐wine stains had standard 16‐channel EEG recordings at approximately 4 month intervals. Thirty artifact‐free, two‐second epochs were selected from each record. Bipolar data from each epoch was submitted to a fast Fourier transform with a frequency resolution of 0.5 Hz. Power spectra from each side of the head were compared using a laterality score (LS = (Left ‐ Right)/(Left + Right). The LS were calculated for delta (2.0–3.5 Hz), theta (4.0–7.5 Hz), alpha (8.0–12.5 Hz), beta (13.0–32.0 Hz), and total power (2–32 Hz). All EEGs were independently read by an experienced electroencephalographer. Results: Subject #1 had a left‐sided PWS and initially left‐sided brain involvement with his first focal seizure at 8 months of age. EEGs at 11 and 15 months were read as symmetric, but quantitative EEG showed significantly lower power in the left hemisphere. MRI at this time showed left‐sided leptomeningeal angiomatosis. MRI at 18 months showed bilateral occipital lesions. Further EEGs at 19 and 23 months yielded no statistically significant asymmetries in total power, possibly reflecting either his bilateral involvement or normal development and fully controlled seizures. Subject #2 had EEGs at 2, 6, and 10 months of age that were read as normal; however, qEEG showed significantly lower ipsilateral power at 2 and 10 months. MRI at 6 months showed unilateral leptomeningeal enhancement. This child has never seized and is developmentally normal. Subject #3 had a single EEG at 4 months that was read as normal; qEEG showed no significant asymmetry. MRI at six months showed no stigmata of SWS. This infant has had neither seizures nor any neurodevelopmental delay. Subjects #2 and #3 will continue to have EEGs every 3 months for a total of two years. Conclusions: In infants with a unilateral port‐wine stain, serial qEEG shows decreases in power over the hemisphere that is affected or at‐risk for neurological impairment. Further studies are needed to determine if qEEG is a reliable and sensitive method for identifying children with unilateral PWS who need neuroimaging and determining the optimal timing of imaging. 1Michael D.Furman,1Jennifer D.Simonotto,2Paul R.Carney, and1William L.Ditto(1Biomedical Engineering, University of Florida, Gainesville, FL; and2 Department of Pediatric Neurology, University of Florida, Gainesville, FL ) Rationale: Currently, neural activity and focus localization in epileptic patients is essential for pre‐surgical evaluation. Several methods have been devised to locate activity by applying combinations of EEG, MEG, PET, SPECT, and fMRI. Phase‐Dynamic Quantification combines Recurrence Quantification Analysis, phase correlation, and three‐dimensional temporal and spatial alignment to track seizure activity along with pre‐ and postictal responses. Utilizing high‐frequency (12–25kHz) multi‐channel brain activity, we introduce a method for correlating dynamic temporal phase which permits tracking of the seizure focus. Methods: Multichannel neuronal potentials were collected synchronously over the course of 2 mos from a chronically seizing animal model of limbic epilepsy (n = 1). Arrays of tungsten microwire electrodes were stereotaxically implanted in the dentate and CA1‐2 of the hippocampus. Neuronal activity was recorded at 24414.1Hz. Electrode placement and CA1‐2/Dentate targeting were verified post‐mortem using a high‐field (17.6 Tesla) MRI brain imaging instrument. Hippocampus depth and grid recording were also obtained in humans undergoing presurgical epilepsy evaluations (n = 2). Cortical activity was recorded continuously at 400Hz. All data sets were analyzed for phase correlation. Results: In all time series analyzed, the results demonstrated a well‐defined seizure path within the epileptogenic focus. In each human data set, phase correlation coincided with the seizure focus that was determined by visual analysis of temporal lobe depth and grid recordings. Figure shows dynamic moving focus phase correlation using three 400 Hz signals. Conclusions: Phase‐Dynamic Quantification reveals epilepsy as a highly organized contiguous neural pathway that can be localized and tracked. This method may of clinical utility assist with seizure localization. (Supported by Office of Naval Research Funding, National Institutes of Health, National Science Foundation, Children's Miracle Network.) 1Georges A.Ghacibeh,2Deng‐ShanShiau,2LindaDance,1StephanEisenschenk, and1,2,3J. ChrisSackellares(1Department of Neurology, University of Florida, Gainesville, FL;2Department of Neuroscience, University of Florida, Gainesville, FL; and3 Department of Biomedical Engeneering, University of Florida, Gainesville, FL ) Rationale: Short‐Term Maximum Lyapunov Exponent (STLmax) is a measure of the order of a dynamical system. Smaller STLmax values indicate that the signal is more ordered. In temporal lobe epilepsy, the dynamical properties of the preictal, ictal and postictal states are distinctly different and can be defined quantitatively. STLmax has been observed to be the lowest during a seizure, intermediate in the preictal period and highest in the postictal state. Antiepileptic medications are known to reduce seizure frequency in most epilepsy patients. We postulate that high levels of antiepileptic medications are associated with higher STLmax. The objectives of this study are to compare STLmax values from intracranial EEG recordings when patients have high, medium and low levels of antiepileptic medications. Methods: Two patients with intracranial subdural grids were recruited: patient 1 had right fronto‐temporal and patient 2 had left fronto‐central grids. Patients were tapered off their antiepileptic medications to record seizures. Patient 1 was on phenytoin, levetiracetam, carbamazepine and zonisamide and Patient 2 was levetiracetam and topiramate. Medication dosages were collected daily for all antiepileptic drugs during the entire monitoring period and medication levels were obtained for the older drugs. For each patient, three distinct time segments were determined based on their medication dosages and levels. Day 1: high medication levels, Day 2: medium levels and Day 3: low levels. All three segments were chosen to be at least 6 hours prior or after a seizure. For each patient, a total of 8 channels were selected for analysis. Four channels were within the epileptic focus and four outside the epileptic focus. For each analyzed electrode, STLmax was calculated for each non‐overlapping 10.24 second segment (4096 EEG points in 400 Hz recording). A total of 100 mean STLmax values were randomly sampled from time intervals in each of the three medication levels. One‐way ANOVA test was applied to test the significance of the effect among medication levels. If the effect was found significant, multiple comparison tests were utilized to test the significance of pairwise differences. Results: In both patients, within both epileptic and non‐epileptic areas, EEG signals recorded during high medication level (day 1) exhibited significantly higher (p < 0.01) STLmax values (less ordered) than during medium (day 2) and low (day 3) medication levels. Conclusions: Antiepileptic medications have an influence on STLmax. High levels are associated with increased STLmax value. This effect may be related to the anticonvulsant properties of these medications. 1Julie D.Golomb,2MichelShamy,2April R.Levin,2KathrynDavis,2Kelly A.McNally,3KennethVives,3Dennis D.Spencer,2,3Susan S.Spencer,2HittenZaveri, and1,2,3,4HalBlumenfeld(1Interdepartmental Neuroscience Program, Yale University, New Haven, CT;2Department of Neurology, Yale University School of Medicine, New Haven, CT;3Department of Neurosurgery, Yale University School of Medicine, New Haven, CT; and4 Department of Neurobiology, Yale University School of Medicine, New Haven, CT ) Rationale: Prior studies suggest that temporal lobe seizures cause impaired function in widespread neocortical areas in addition to mesial temporal lobe regions. A recent study by Blumenfeld et al. (NEUROLOGY, 2004) found increased low voltage fast and polyspike ictal activity in ipsilateral mesial and lateral temporal regions, with simultaneous large amplitude ictal slowing in frontoparietal regions. However, these observations were based on subjective ratings of the intracranial EEG recordings and were not correlated with behavior. In the present study we employ quantitative analyses to more accurately describe ictal and post‐ictal changes associated with various brain regions during complex and simple partial seizures. Methods: 12 Patients with surgically confirmed mesial temporal lobe epilepsy who underwent intracranial EEG monitoring and had no seizures during a follow‐up period of at least 1 year after temporal lobe resection were selected for analysis. Seizures that secondarily generalized were excluded. Electrode contacts were assigned to one of nine cortical regions based on MRI surface reconstructions. Power measurements for classically defined spectral bands were calculated with the Fourier transform for the ictal and peri‐ictal time periods for each EEG signal. Results: A small, brief increase in beta (13–25 Hz) and gamma (25–50 Hz) frequency power was seen in ipsilateral mesial temporal regions immediately following seizure onset. This was followed by a much larger increase in alpha (8–13 Hz), beta, gamma, and particularly theta (4–8 Hz) power in ipsilateral mesial and lateral temporal regions. Simultaneously, a large increase in delta (0–4 Hz) power was seen in lateral, medial, and orbital frontal regions, which sometimes persisted throughout the post‐ictal period. Subclinical seizures involved transient changes confined to the ipsilateral mesial temporal contacts. Although considerable between‐patient variability was found, individual patients showed remarkable consistency in EEG patterns across their seizures. Conclusions: Quantitative power analysis showed increased high frequency activity in temporal regions and low frequency activity in frontal regions during partial seizures of temporal lobe origin. These results support and expand upon previous studies of EEG activity positing that neocortical slowing represents a form of “surround inhibition” during such seizures. Further analyses may reveal specific differences between EEG signals in complex and simple partial seizures, and contribute to an understanding of behavioral manifestations of temporal lobe seizures. 1ElenaUrrestarazu,1PierreLeVan, and1JeanGotman(1 EEG, Montreal Neurological Institute, McGill University, Montreal, QC, Canada ) Rationale: The aim of this study was to investigate whether Independent Component Analysis (ICA) can be applied in intracranial recordings to analyze interactions between the temporal lobes during temporal lobe seizures. ICA has already been used in scalp EEGs to isolate spikes from background and to help lateralize temporal lobe seizures Methods: Intracranial recordings of temporal lobe seizures from 8 patients with bitemporal implantation were analyzed. Scalp EEG recordings showed that the patients had seizures originating from both temporal lobes. The seizures of each patient were classified in four categories: unilateral right or left (UR or UL), and bilateral with right or left onset (BR or BL). ICA decomposes the EEG into as many statistically independent components as there are channels. ICA was applied to the seizures, and only components with ictal activity in the first 10 seconds of the seizure were studied. Correlation coefficients were calculated between these components and every channel of the original EEG, in the time interval preceding the appearance of bilateral activity. Components were classified as unilateral if their correlation was greater than 0.2 exclusively with channels in one hemisphere; if the correlation was greater than 0.2 with channels in both hemispheres, the component was labeled as bilateral. Nonparametric statistics were applied. Results: 46 seizures were analyzed (11 UR, 13 UL, 12 BR and 10 BL). There were five patients with seizures beginning independently in both temporal lobes, two patients with right temporal seizures only, and one patient with left temporal seizures only. Even though the correlation was calculated when seizure activity was unilateral, the proportion of components that were bilateral was significantly lower (p = 0.001) in seizures that stayed unilateral (22% for UR and 26% for UL) than in seizures that became bilateral (46% for BR and 48% for BL). There were no significant differences between right and left seizures. Bilateral seizures were significantly longer than unilateral seizures (p = 0.000), but the duration of the seizure was not significantly correlated (p > 0.05) with the proportion of bilateral components. Conclusions: In patients with bitemporal epilepsy, more than 20% of the components extracted using ICA have a bitemporal distribution even at the time when the seizures are apparently unilateral. The proportion of bilateral components during the unilateral phase is significantly higher in seizures with subsequent contralateral spread. It therefore appears that minimal contralteral seizure activity is present even when the discharge appears unilateral, and that the more such contralateral activity is present, the more the seizure is likely to spread. (Supported by CIHR Grant # MOP 38079 Postdoctoral research grant from the Department of Education of the Basque Government.) 1C.Grova,2J.Daunizeau,1A.Bagshaw,1E.Kobayashi,2J.M.Lina,1F.Dubeau, and1J.Gotman(1Montreal Neurological Institute, McGill U., Montreal, QC, Canada; and2 Centre de Recherche en Mathématiques, Montreal, Canada ) Rationale: The first objective is to compare two imaging modalities to better identify brain areas where spikes are generated:– EEG source localization , which allows estimating a current density along the cortical surface at each time sample of the spike – Simultaneous EEG/fMRI , which allows measuring the hemodynamic correlates of EEG activityThose modalities highlight several aspects of spike generation, because signals have different physiological origins and complementary spatial and temporal resolutions. Moreover, EEG source localization has no unique solution and adding prior information is necessary. The second objective is to evaluate whether it is relevant to include fMRI information in EEG source localization. Methods: 9 patients with focal epilepsy underwent EEG/fMRI examination and EEG recording outside the scanner. The EEG/fMRI protocol consists of recording 19 EEG channels inside the MR scanner. After manual detection of interictal spikes, data were analysed using a linear model combining several hemodynamic responses after each spike. T maps showing brain areas with significant fMRI responses were interpolated on the cortical surface, to compare with EEG source localization. Source localization was applied to averaged spikes recorded outside the scanner (43 channels). We applied the Maximum Entropy on the Mean (MEM) approach 1 to estimate the current density on the cortical surface at each time sample of the spike. fMRI results were compared with MEM results obtained at the peaks of the spike. The level of concordance was assessed by measuring the minimal geodesic distance D between local extrema of the fMRI and MEM maps. We also measured an index A to assess whether fMRI could be included as a prior for EEG source localization, using a hierarchical Bayesian model 2 . Large positive values of A mean that fMRI data were highly relevant information regarding EEG data. Results: 3 patients did not show any fMRI response. Among the 6 other patients, MEM and fMRI results showed good concordance in 4 (D < 2.5 cm). However, qualitative analysis revealed that most exhibited concordant areas and areas unique to each modality. Our index A was in good agreement with the qualitative comparison of MEM and fMRI results. Conclusions: Our study highlights that EEG source localization and EEG/fMRI clearly explored different phenomena linked to spike generation. Brain areas involved could be concordant or complementary regarding both modalities. The index of relevance A seems promising to decide whether fMRI or a part of the fMRI map should be used as prior for source localization. By analysing more data, those approaches may provide a better understanding of the mechanisms involved during spike generation and propagation. REFERENCES 1.  Amblard C et al2004 IEEE TBME , 51 ( 3 ). 2.  Daunizeau J et al2005 IEEE TSPS , in press (Supported by Canadian Institute of Health Research, Jeanne Timmins Costello Fellowship.)1Kurt E.Hecox,1FengmeiLui,1SeaonMarler,1JenniferDwyer,1MichaelKohrman,1ArnettaMcGhee, and1JoelFontanarosa(1 Pediatrics, University of Chicago, Chicago, IL ) Rationale: Dynamic Systems analysis has been applied to the study of interictal and ictal EEG data for more than a decade. Abnormalities in these measures have been reported in the detection, prediction and localization of seizures. The purpose of this study is to extend these observations to the pediatric population, to compare ictal data to three forms of baseline data and to describe some of the features of the spatial distribution of the systems changes, for multiple metrics. Methods: Intracranial and extracranial recordings were obtained from a series of pediatric aged patients undergoing evaluation for surgical treatment. Ages varied from 5 to 19 years of age. Three examples of seizures, interictal sleep, interictal awake and preictal (within one minute of seizure onset) were selected and analyzed for each patient. Metrics were calculated for correlation dimension (least squares), correlation dimension (maximum likelihood), Kolmogorov entropy, eigenvalues and Z (a global measure of nonlinearity) in each case. These metrics were calculated for at least eight surface electrodes (four over each hemisphere) and for at least 16 intracranial electrodes. The intracranial electrodes were further subdivided into “involved” versus “uninvolved” in the ictal event according to visual analysis. These data were subjected to analysis of variance and chi square analyses. Results: Eighty percent of the ictal events were detectable. Differences were seen between sleep and wakefulness baselines for 70% of the the group. Preictal changes were seen for 40% of the group, compared to baseline. Seemingly uninvolved electrodes show changes in these metrics both ictally and interictally. The magnitude of the change and the metris showing the changes varied across patients but eigenvalue was most commonly changed amongst the measures. Eigenvalue and correlation dimension (least squares) were the least specific spatially, while Kolmogorov entropy and Z were the most spatially specific. Conclusions: We continue to observe ictal changes in dynamic systems measures in pediatric aged patients with seizures. The measures are also abnormal in a number of patients interictally. No single metric is complete enough to stand alone, from a detection or anticipation perspective. Anticipation was possible in nearly half of the cases. The most surprising findings of this study were that electrodes felt to be “control” or uninvolved in the seizures can show significant changes in these metrics and that the metrics vary considerably in their spatial specificity. Localizing a focus using these metrics must be performed with considerable caution. Further, the differences in spatial distribution across metrics suggest that these metrics do not all reflect the operation of a single process. (Supported by Falk Medical Trust Foundation.) 1Christophe C.Jouny,1Piotr J.Franaszczuk, and1Gregory K.Bergey(1 Department of Neurology ‐ Epilepsy Center, Johns Hopkins University School of Medicine, Baltimore, MD ) Rationale: Partial seizures originate from focal regions of epileptogenesis. The dynamic of seizure onset and pattern of ictal discharges may reflect the cerebral region of seizure onset. To date, however, these observations have been largely derived from visual analysis or simple frequency characterization. We here report detailed comparative analysis of seizure dynamics at seizure onset. Methods: We analyzed data from multiple partial seizures from patients with mesial temporal onset epilepsy (MTLE; n = 3) or with neocortical lesional epilepsy (NLE; n = 6) monitored in the epilepsy monitoring unit for presurgical evaluation with intracranial subdural grid arrays. The time‐frequency decomposition was obtained with the matching pursuit (MP) method. The Gabor atom density (GAD) (Jouny et al. 2003), derived from MP, provides a measure of signal complexity. Different features of the intracranial EEG (ICEEG) or of GAD can be used to synchronize seizures. Onset of complexity increase was used to synchronize the events. Time‐frequency maps reconstructed from the MP decomposition were then re‐aligned and averaged. Results: GAD plots reveal that all mesial temporal onset seizures and neocortical onset seizures have an increase in complexity at seizure onset and have a reproducible pattern of onset. These consistent features can be used to synchronize the events. Because of the slight variability in timing between different phases of the seizure, one can enhance the visibility of any of these phases by choosing their feature as the synchronizing event for the re‐alignment. Onset features which are not time‐locked will not be enhanced by averaging. In this study, averaged seizure onsets exhibit distinguishable predominant frequency components in 5 out of 9 patients (either ∼7Hz, ∼15Hz, ∼20Hz or ∼40Hz) or activities with a broader spectral signature in 3 out of 9 patients. Conclusions: The assumption that multiple seizure onsets in a given patient have stereotypical features is based predominantly on clinical experience and visual analysis. We studied here the detailed time‐frequency components of partial seizures and, by averaging their reconstructed maps, isolated the components that are common to all seizures. One of the advantage of the method was to reveal an onset pattern including high‐frequency components (>40Hz) with a decreasing main frequency which was not evident on ICEEG. High‐frequency components during onset might go unnoticed ‐ especially when embedded in electrodecrement pattern of onset. High‐frequency recordings and detailed time‐frequency analysis can reveal the dynamic evolution of onsets and provide important information regarding timing of onset pattern and localization. (Supported by NIH grant NS 33732.) 1AnnaKorzeniewska,2RafalKus,1PiotrFranaszczuk,3CiprianCrainiceanu, and1Nathan E.Crone(1Department of Neurology, The Johns Hopkins University School of Medicine, Baltimore, MD;2Laboratory of Medical Physics, Warsaw University, Warsaw, Poland; and3 Department of Biostatistics, The Johns Hopkins University, Baltimore, MD ) Rationale: The cortical regions responsible for word production have been mapped with a variety of techniques, but the dynamic interactions between these cortical regions in real time have yet to be explored in depth. To investigate these interactions we recorded electrocorticographic (ECoG) signals while subjects spoke words in response to either visual or auditory stimuli, and we studied event‐related gamma activity flows among different cortical regions. Methods: The short‐time direct directed transfer function (SdDTF) is a new method for estimating causal interactions between different brain regions. SdDTF evaluates the direction and intensity of EEG activity flows during consecutive short‐time epochs, for selected frequency bands. This multichannel approach enables comparisons of the strengths of relations between sites. ECoG signals recorded during two tasks, word repetition (auditory stimulus) and picture naming (visual stimulus), were analyzed. Event‐related changes in cortical network interactions were studied with emphasis on 60–140 Hz (high gamma). Results: Auditory word stimuli were associated with activity flows mainly between sites within superior temporal gyrus (STG) and from STG to inferior frontal gyrus (IFG) and to ventral sensorimotor cortex (vSMC); interactions between STG and basal temporal language area (BTLA) were also observed. In the interval between offset of the auditory word stimulus and onset of the verbal response, flows were observed predominantly from STG to IFG and BTLA. Flows between IFG, STG, and vSMC were also present. During picture naming the interval between onset of the stimulus and the verbal response was associated with many bidirectional interactions between IFG and STG, and between IFG, BTLA and vSMC. Spoken responses to both visual and auditory stimuli were associated with relatively fewer interactions, albeit involving all aforementioned sites. Conclusions: Cortical network interactions during word production predictably depend on stimulus modality. These interactions are reflected by changes in high‐gamma activity flows between and within cortical regions commonly involved in language processing. (Supported by NINDS R01‐NS41598.) 1LorantKovacs,2NandorLudvig,2OrrinDevinsky, and2Ruben I.Kuzniecky(1Software Division, ESCO, Garden Grove, CA; and2 Neurology, Comprehensive Epilepsy Center, NYU School of Medicine, New York, NY ) Rationale: The objective was to develop a program (“SeizureGuard”) that is ideal for recognizing/predicting EEG seizures, real‐time, within the limited power resources of future, fully implanted therapeutic devices, like the hybrid neuroprosthesis. Methods: Instead of handling the EEG signals as a succession of waves, the SeizureGuard program decomposed these signals into vectors, describing each vector with its angle and magnitude. From the derived vector‐stream, the program computed various neurobiologically relevant parameters, including inter‐vector interval and vector‐periodicity. These parameters were indexed according to their relevance to the characteristics of electrographic seizure‐onsets. No complex numbers were used in the calculations. To validate the algorithm, a data miner utility program was utilized to extract continuous EEG signals from files generated with a Nicolet BMSI 6000 recording system. All recordings were obtained from temporal lobe epilepsy patients presurgically implanted with subdural strip‐ and grid‐electrodes. Results: The program accurately recognized subclinical EEG seizures within 1–2 sec from their onset. Artifacts and non‐epileptic rhythmic discharges and large‐amplitude waves were not falsely indicated. Interestingly, the occasional false‐positive seizure detections by the program in the interictal phase actually indicated bursts of sharp waves/spikes that occurred in the recording channel(s) of seizure activity. Conclusions: The SeizureGuard program, utilizing a computationally inexpensive algorithm that decomposes the EEG waves into vectors, is suitable for recognizing the onsets of focal subclinical seizures and has the potential to indicate epileptiform events prior to these seizures. As such, it seems to be ideal for use in implanted seizure‐controlling devices. (Supported by NYU/FACES.) 1YuanLai,2WimVan Drongelen,2David M.Frim,2Kurt E.Hecox, and1BinHe(1Biomedical Engineering, University of Minnesota; and2 Pediatrics, University of Chicago ) Rationale: In past decades, epilepsy surgery has been recognized increasingly as a viable treatment for patients with medically refractory seizures. However, the large gap between number of potential surgical candidates and number of patients being treated with surgery still exists due to the lengthy, complex, and expensive procedure of current practice of using invasive procedures to identify epileptic foci responsible for the seizures. If epileptic foci can be identified noninvasively, numerous epilepsy patients undergoing surgical treatment will benefit significantly from the noninvasive surgical planning. The aim of present study is to evaluate a novel noninvasive electrophysiological cortical imaging technique (He et al., IEEE‐TBME, 46:1264–1268, 1999) to image and localize ictal activities from scalp EEG. Methods: All patients are pediatric patients with intractable partial epilepsy in the Pediatric Epilepsy Center at the University of Chicago. For each patient, multiple habitual seizures were visually identified according to the IFSECN criteria. Seizure activities with sudden propagation to whole brain were excluded from further analysis. For ictal EEG recordings with sustained rhythmic focal morphology, time‐frequency analysis has been applied as preprocessing. Spatial‐temporal evolution of ictal discharges were examined using wavelet analysis to obtain a time‐varying energy distribution in each frequency band. The dominant frequency is determined with highest energy concentration, at which the phase encoding is used for multi‐channel EEG around seizure onset to generate a scalp potential map representing the major ictal activity. The cortical imaging algorithm (He et al., 1999) was then applied to reconstruct the cortical potential distribution from the phase encoded scalp potential map. Since pediatric patients usually have seizure originating from neocortex, the estimated cortical potential generated by different underlying sources are less likely to overlap so that it can be used to observe the source activity with much less deviation as compared to that from a deeper source. Results: Five pediatric epilepsy patients by various causes were studied, who are either seizure free or having substantial seizure reduction after neurosurgical resections. For all patients, the cortical imaging analysis has been able to successfully reveal underlying epileptogenic cortical zones at different lobes (temporal or extra‐temporal, one or two epileptogenic foci for each patient), which are consistent with clinical findings and confirmed by neurosurgical resections and outcomes. Conclusions: Cortical imaging can reliably localize cortical activity in regions displaying epileptiform activity; these regions were confirmed by successful surgical resection. Cortical imaging may become a useful alternative for noninvasive pre‐surgical and surgical planning in pediatric epilepsy patients. (Supported by NIH EB 00178 and NSF BES‐0411898.) 1PierreLeVan,1ElenaUrrestarazu, and1JeanGotman(1 EEG, Montreal Neurological Institute, McGill University, Montreal, QC, Canada ) Rationale: The objective of this study was to devise an automated system to remove artefacts from ictal scalp EEG, using independent component analysis (ICA). Current ICA methods rely on a tedious visual identification of artefactual sources among extracted components, hence the desire for a fully automated method. Methods: 69 ictal scalp recordings from 16 epileptic patients were analyzed. The seizures were contaminated by various artefacts such as eye movements, EMG, and patient movement. For each seizure, ICA was applied to a 30s interval beginning approximately 10s before the suspected seizure onset. The extracted independent components were divided into 2s epochs and each epoch was labelled as either EEG or non‐EEG. Data from half of the patients was used as a training set to induce a Bayesian network classifier, while the remaining data was reserved for use as a validation set. The classifier used the following features: component negentropy, epoch variance, spectral entropy between 5 and 30Hz, and relative power in several frequency bands. The spatial distribution of each component was then fitted with a dipole, the position of which was also used as a feature whenever the residual variance of the fit was less than 20%. For each 2s epoch, the output of the classifier was the probability that the epoch represented EEG activity. A component was considered to be artefactual if the sum of the probabilities for its 15 epochs was less than 4. To evaluate the performance of the system, an expert neurologist reviewed the original seizure recordings in the validation set and compared them with the records reconstructed after rejection of the artefactual components. A qualitative score was given with respect to the reduction of artefactual activity and the preservation of EEG activity. Results: The validation set contained 33 seizures from 8 patients. The system correctly classified 5047/5920 (85%) EEG epochs and 5628/6950 (81%) non‐EEG epochs. The reviewer noted that 23 seizures were contaminated by a significant amount of artefacts. After the automated rejection of artefactual components, the reviewer determined that the majority of artefacts were removed in 17 seizures, while 5 of them had minor improvements, and only 1 record had no reduction of artefactual activity. The system preserved all the EEG activity in 17 seizures, attenuated some minor EEG activity in 4 cases, and removed some significant ictal activity in the remaining 2 cases. The automated method was also applied to the 10 seizures that did not contain significant artefacts; in 9 cases, all the EEG activity was preserved, while the remaining case had only minor EEG attenuation. Conclusions: Temporal and spatial features of ICA components were used in a Bayesian framework to classify component epochs as either EEG or non‐EEG. This allowed the proposed system to automatically eliminate a large proportion of artefactual components in ictal scalp recordings, while minimally affecting the EEG activity. (Supported by NSERC CGS‐M, CIHR MOP‐10189.) 1John W.Miller,1WonsukKim,1MarkHolmes, and2SampsaVanhatalo(1Neurology, University of Washington, Seattle, WA; and2 Clinical Neurophysiology, University of Helsinki, Helsinki, Finland ) Rationale: We recently described techniques to perform longterm DC‐coupled EEG recordings at the bedside, and have used these methods to demonstrate that inspection of infraslow (<0.5 Hz) activity can correctly lateralize temporal lobe seizures (Vanhatalo et al, Neurology.60:1098, 2003). However, even high amplitude infraslow activity can sometimes be difficult to localize by simple visual inspection if there is considerable overlying faster EEG activity or obscuring slow artifact. Also, this earlier study did not investigate extratemporal seizures, which are often more difficult to localize. We address these issues with improvements in the DC‐coupled EEG recording and analysis technique, and by extending our observation to seizures arising from a variety of cerebral regions. Methods: Recordings were performed on patients receiving presurgical monitoring for medically intractable localization related epilepsy with 20 seizures in 11 patients included in analysis. Recordings used a commercial system for DC‐coupled recording, with sintered Ag/AgCl electrodes in a standard array defined by the 10:10 system, with additional anterior temporal electrodes, as well as subtemporal chains. Afterwards, ictal events were exported to an EEG analysis software package (BESA 5.0) to review the seizures with source montages. Recognition and localization of the infraslow signal was made easier (i) by compressing the time scale (30 to 90 seconds per screen), ii) by filtering the signal in time domain (i.e. regular high pass/low pass or band pass filtering), iii) by filtering the signal in space (e.g. with a source montage), and iv) by using a moving average to smooth the recording. The spatial filtering properties of the source montage also makes it easier to differentiate the ictal change from physiological artifacts, such as eye movements, that arise from other head regions. In addition, the time frequency characteristics of some seizures were analysed by FFT and by use of Morlet wavelets. Results: Infraslow signals were present with all focal seizures, sometimes with amplitude orders of magnitude higher than the ictal signals in the convention EEG frequency band (0.5 to 70 Hz). Analysis of these infraslow signals was a reliable localization tool. Five patients in this study received epilepsy surgery and had follow‐up documenting significant seizure reduction, and analysis of infraslow signals correctly localized epileptic foci in all five, while conventional noninvasive EEG recording with visual analysis of seizures correctly localized only two. Conclusions: DC‐coupled EEG recordings can give additional information that is clinically useful to noninvasively localize the epileptogenic focus. The value of this method is increased by source analysis tools that can reveal highly localized changes more clearly than direct visual inspection. (Supported by The UW Regional Epilepsy Center.) 1Victoria L.Morgan,2ElianeKobayashi,2AndrewBagshaw,2ColinHawco,3BasselAbou‐Khalil,2FrancoisDubeau, and2JeanGotman(1Radiology, Vanderbilt University, Nashville, TN;2Montreal Neurological Institute and Hospital, McGill University, Montreal, Canada; and3 Neurology, Vanderbilt Unversity, Nashville, TN ) Rationale: Simultaneous fMRI and EEG may localize the generators of interictal spikes recorded on scalp EEG. Temporal clustering analysis (TCA) is a method to analyze fMRI images independently of EEG information. TCA seeks to identify resting state activations related to the epileptogenic zone. The objective of this study was to compare the results of these two analyses in a series of patients with temporal lobe epilepsy (TLE). Methods: BOLD fMRI images were acquired in 22 TLE patients on a 1.5T MRI scanner with simultaneous continuous EEG in two‐hour sessions at the MNI (64 ×64 ×25, 5mm × 5mm × 5mm, TR = 3sec, 120 volumes per series, 8–10 series per patient). For EEG‐BOLD analysis, EEGs were processed offline and spikes were identified in the filtered EEG for event‐related analysis. For the TCA‐BOLD analysis, only the first three series were used. Data were analyzed by previously published TCA methods in which a histogram is created indicating the number of voxels that reach at least 80% of their maximum at each time point. Activated regions were determined by using a fixed‐effects analysis of the three series. The concordance of the temporal lobe activation identified by the two methods and the temporal lobe spikes were assessed. Patients were divided into those with unilateral and those with bilateral independent temporal lobe spikes on EEG. One patient's fMRI was not analyzed due to excess motion. Results: Fourteen patients had unilateral spikes on EEG: 12/14 had TCA‐BOLD temporal activation (6 were ipsilateral to the EEG spikes, and 2 had bilateral activation with predominance ipsilateral to the spikes) and 9/14 showed EEG‐BOLD activation (2 were ipsilateral to the EEG spikes, 5 had bilateral activation with predominance ipsilateral to the spikes, one had bilateral activation with no side predominance). Seven patients had bilateral independent temporal spikes on EEG: all patients showed TCA‐BOLD activation in the temporal lobes, but it was bilateral (concordant with the EEG) in 3; 4/7 patients showed EEG‐BOLD temporal lobe activation, bilateral in 3. Overall, we observed an agreement between both methods in 7/11 patients. Conclusions: Activations in the temporal lobes were detected in 90% of patients using TCA‐BOLD and in 68% using EEG‐BOLD. TCA, which has the advantage of not using the information from the EEG, was concordant with EEG‐BOLD analysis in 64% of those who showed activations, and disclosed responses not present in the former one in 5. TCA‐BOLD could lateralize the underlying structural MRI lesion in 52% of the patients, similar to that observed with EEG‐BOLD analysis (48%). More development is required for both methods. (Supported by Epilepsy Foundation and the Canadian Institutes of Health Research, Preston Robb MNI fellowship, and Savoy Foundation for Epilepsy.) 1AmitMukherjee,2PradeepModur, and1NicolaosKarayiannis(1Electrical & Computer Engineering, University of Houston, Houston, TX; and2 Neurology, University of Louisville, Louisville, KY ) Rationale: We describe a novel method for detection of seizure onset using a quantum neural network (QNN) that was progressively trained by quantitative features (in both the time and frequency domains) extracted from intracranial EEG (iEEG) signals. QNNs are particularly helpful in the classification of uncertain data without any restricting assumptions. [Purushothaman and Karayiannis, IEEE Trans Neural Networks 1997; 8:679–693]. Methods: Raw iEEG data (subdural and depth recordings) from 4 subjects, sampled at 1000 Hz, were analyzed. Bandpass was set from 1.6 to 300 Hz. Seizure onset was marked at the time of earliest change in EEG. The EEG segment, 10 sec before and after seizure onset, was divided into onset epochs (OE), each of 2 sec length, with 50% overlap between consecutive epochs. Background epoch (BE) was defined as the average of 30‐sec EEG during the interictal period, preceding the OE by at least 2 min. The signal was decomposed using a Daub‐2 wavelet and scales 2 to 6 (corresponding to frequency bins 250–125, 125–63, 63–32, 32–16, 16–8 and 8–4 Hz respectively) were used for feature analysis. Two parameters were computed for each scale: relative energy (ratio of the energy in a given scale in the OE to the energy of the same scale in the BE); average of the distribution tail (ratio of the average of the top 80th percentile squared coefficient to the average of the squared coefficient for a given scale). In addition to this average peak‐to‐peak amplitude and curve length (measure relating to the fractal dimension) of the OE are also computed and used as features. QNN was initially trained using seizure onset patterns from multiple subjects. With each subsequent seizure, the QNN was progressively re‐trained by populating the training with 50% of the epochs belonging to the last seizure, 25% belonging to the previous to last seizure and 25% from a bank containing a variety of patterns from multiple subjects. Decision‐making was based on the inherent ability of the QNN to identify and quantify uncertainty, represented by its output values between 0 (non‐seizure) and 1 (seizure), in conjuction with a spatio‐temporal rule. Thus, seizure onset was reported when the activity was observed in at least 2 adjacent channels and the activity was sustained for at least 6 sec (with QNN output value >0.65). If a channel had reported seizure in the past, its membership threshold was reduced to 0.4 and the spatio‐temporal rule was applied. The QNN classifier was tested on 8 seizures from 2 subjects (each with 4 seizures). Results: All the seizure onsets (100%) were detected using the above method. There were only a few false detections. Conclusions: Progressively trained QNNs appear to be promising in detection of intracranial seizure onset. Further investigation will focus on detection of a wider variety of seizure onset patterns in multiple subjects and different decision‐making strategies. 1AyakoOchi,1HiroshiOtsubo,1RyoichiIwata,1TakanoriFunaki,1TomoyukiAkiyama,1RohitSharma,1Shelly K.Weiss,1ElizabethDonner,1IreneElliott,2James T.Rutka, and1O. CarterSneadIII(1Neurology, The Hospital for Sick Children, Toronto, ON, Canada; and2 Neurosurgery, The Hospital for Sick Children, Toronto, ON, Canada ) Rationale: High frequency oscillations (HFOs) in the frequency range of 80 Hz and higher has been investigated from ictal subdural EEG data in patients with intractable localization‐related epilepsy. Multiple band frequency analysis (MBFA) provides high resolution of frequency changes and time courses in HFOs. We studied ictal HFOs from extraoperative video subdural EEG monitoring (SDEEG) using MBFA to characterize patterns of HFOs in children with intractable neocortical epilepsy. Methods: We retrospectively studied 8 children (3 girls and 5 boys; mean age, 14 years; range, 8 to 17 years) who presented with intractable neocortical epilepsy and underwent SDEEG consisting of 48 to 116 contacts (AD‐TECH Medical instrument corporation, Racine, WI, USA). We recorded SDEEG for 48 to 144 hours with 1,000 Hz sampling rate (HARMONY 5.4, Stellate, Montreal, PQ, Canada). We analyzed 4 to 11 seizures by selecting SDEEG data epochs of 10 to 50 seconds including 5 seconds before the ictal EEG onset. We used MBFA using the software program Short Spectrum Eye (Gram, Saitama, Japan) for frequency and distribution analyses of HFOs. The power spectrograms of frequency bands between 5 Hz to 300 Hz were calculated with a frequency resolution of 2 Hz and a temporal resolution of 25 ms. We arranged all power spectrograms in the same order as subdural grid electrodes and evaluated the distribution of HFOs. Results: Seven of 8 patients presented with regionally predominant ictal HFO changes (mean, 23 electrodes). We found two patterns of HFOs consisting of existence or absence of sustained narrow band of HFOs. Four of 7 patients had initial wide range of HFOs (up to 115 Hz) lasting 2 to 5 seconds before the clinical seizure onset, followed by a sustained narrow band of HFOs (mean, 101 Hz) during ictal symptom of partial seizures. The seizures in these 4 patients evolved into secondary generalization. The other three patients had wide range of HFOs (up to 214 Hz) before and after the clinical onset without following sustained narrow band of HFOs. They presented with brief partial seizures in cluster in one patient, epileptic spasms in 2, but they did not evolved into secondarily generalized seizures. The cortical resection areas in 7 patients included the region of predominant HFOs (mean maximum frequency, 138 Hz). They achieved good postsurgical outcomes. The remaining one patient had no regionally predominant ictal HFO changes. This patient has residual partial seizures after frontal cortical excision. Conclusions: MBFA characterized 2 patterns of ictal HFO changes between seizures with and without secondarily generalized in pediatric neocortical epilepsy. The regionally predominant ictal HFOs represented the ictal onset and ictal symptomatogenic zones. 1,2HannesOsterhage,1FlorianMormann,1Ralph G.Andrzejak,1Christian E.Elger, and1,2KlausLehnertz(1Dept. of Epileptology, University of Bonn, Bonn, Germany; and2 Helmholtz‐Institute for Radiation and Nuclear Physics, University of Bonn, Bonn, Germany ) Rationale: Patients suffering from intractable focal epilepsy may benefit from neurosurgical resection of the epileptic focus provided it can be identified via electroencephalographic recording of the seizure onset. A question of interest is whether the focus can be reliably identified from interictal recordings, i.e. without the necessity of observing seizures. Measurements of the synchronization between different EEG channels have been found to be promising for this purpose. Up to now, a number of measures for synchronization have been proposed based on different approaches. In this study a comparison between different linear and nonlinear synchronization measures was performed with respect to two aspects: 1. general performance of the measures in terms of their ability to discriminate the ipsi‐ from the contralateral hemisphere, 2. comparison of different measures to investigate whether the different approaches render complementary or redundant information. Methods: We analyzed intracranial EEG recordings from the seizure‐free intervals of 30 patients with medically intractable medial temporal lobe epilepsy undergoing invasive presurgical diagnostics. EEG signals were recorded via bilateral intrahippocampal depth electrodes, each equipped with 10 recording contacts and implanted stereotactically along the longitudinal axis of the hippocampal formation. The total recording time comprised more than 83 hours. Several measures of synchronization were calculated for all combinations of depth electrodes within each hemisphere using a moving window technique, and then averaged over time. For each of the measures the average spatial synchronization in each hemisphere was calculated. A detection of the focal hemisphere was carried out for all measures and patients based on the assumption of a higher average synchronization in the focal hemisphere. True and false lateralizations were compared for different measures. Results: Correct lateralization based on higher synchronization of the focal hemisphere was found to be statistically significant for all measures investigated, although actual performance differed among the investigated measures. False lateralizations for different measures ranged between 6 and 8 out of 30. A comparison of correct and false lateralizations showed a strong similarities between measures. Conclusions: The focal hemisphere in an epileptic brain is characterized by a pathological increase in synchronization. The different measures investigated in this study appear to be sensitive to this increased synchronization. Lateralization analysis using synchronization measures could render additional diagnostic information, although the false lateralizations found in this study could limit its clinical applicability. (Supported by the Deutsche Forschungsgemeinschaft and the intramural research fund BONFOR of the University of Bonn.) 1HoumanKhosravani,2RobertPinnegar,2RossMitchell, and1PaoloFederico(1Department of Radiology, University of Calgary, Calgary, AB, Canada; and2 Department of Clinical Neurosciences, Univeristy of Calgary, Calgary, AB, Canada ) Rationale: Visualization of time‐frequency changes in EEG has traditionally been visualized using Fourier‐based techniques (i.e. Short Time Fourier Transform, STFT). This approach is limited given its fixed‐length analysis window and the non‐stationary nature of EEG signals. The S Transform is a Fourier‐based multi‐resolution transform that has a variable length analysis window, thus circumventing the shortcomings of the STFT. In the present study, we introduce a novel application of the S Transform for visualization of normal and epileptiform EEG phenomena in humans. Methods: Selected recordings from five normal subjects and five patients with either focal or generalized epileptiform discharges were used. We identified a single channel that had a clear example of an electrographic feature of interest (e.g. alpha, mu, sleep spindles, eye blink, spike and wave, sharp wave) and applied the S Transform to it. Results: The S Transform provided clear visualization of the frequency components, in time, of the electrographic events. Some of the spectral attributes of these events were not immediately apparent on conventional review of the EEG. Conclusions: The application of the S Transform to EEG is novel and can potentially be a useful tool for visualizing clinically relevant features in EEG. (Supported by Canadian Institutes of Health Research and Alberta Heritage Foundation for Medical Research.) 1CatherineSchevon,2JoshuaCappell,3WernerDoyle,3HowardWeiner,2RobertGoodman,1RubenKuzniecky, and2RonaldEmerson(1Department of Neurology, New York University School of Medicine, New York, NY;2Neurological Institute, Columbia College of Physicians and Surgeons, New York, NY; and3 Department of Neurosurgery, New York University School of Medicine, New York, NY ) Rationale: The presence of localized synchronization of neuronal activity in epilepsy has been well documented in microelectrode and clinical intracranial EEG (ICEEG) recordings. An intriguing hypothesis about the development of epilepsy is that a cortical region, when potentiated by repeated seizures, becomes chronically biased, leading to hypersynchrony. Thus, brain regions most involved in epileptogenesis are marked by chronic local hypersynchrony (LH). In this study, we discuss and illustrate the use of the mean phase synchrony measure to detect and map patterns of locally hypersynchronous neuronal activity in long term ICEEG recordings obtained during the epilepsy presurgical evaluation. Methods: Seven patients with different epilepsy syndromes were studied. Three 5 minute interictal samples from different monitoring days were tested in each case. Two seizures from each patient were studied, with 30 second samples taken from the preictal, ictal, and postictal periods. Interelectrode synchrony was measured using the mean phase coherence algorithm, which calculates the degree of phase locking independent of amplitude. Pairwise synchrony over one‐second epochs was calculated for all adjacent channels of a subdural grid located over or near the clinically determined seizure onset zone. LH regions were defined as a set of contiguous channel pairs with average synchrony greater than one standard deviation above the mean for the entire grid. Results: Chronic LH regions were found in all seven patients, with these characteristics: (1) LH persisted over long recording periods and was state‐independent. (2) Each patient had a unique pattern of LH regions with different anatomical distributions. (3) In each case, an LH area was associated with the clinically identified the seizure onset zone. (4) In peri‐ictal recordings, there was a relative focal desynchronization preictally, a prominent finding in two patients but present to some degree in all seven, and enhanced hypersynchrony postictally. Conclusions: These findings suggest that local hypersynchrony in the EEG is related to abnormalities that underly partial epilepsy. This identification may prove useful in several areas. First, clinical EEG interpretation may be enhanced by using synchrony measurements to help define epileptogenic networks. Second, the described properties may be investigated to shed light on epileptogenic mechanisms. Third, knowledge of network structure and function may lead to improved techniques for disrupting the network to prevent seizures through resection, on‐site neurostimulation, or local drug delivery. Fourth, these observations may lead to improved techniques for automated seizure prediction and detection. FACES 1,7Deng‐ShanShiau,2,7Chang‐ChiaLiu,2,7WichaiSuharitdamrong,2Panos M.Pardalos,1,3–5Paul R.Carney, and1,3–7J. ChrisSackellares(1Neuroscience, University of Florida;2Industrial and Systems Engineering;3Pediatrics;4Biomedical Engineering;5Neurology;6Psychiatry, University of Florida, Gainesville, FL; and7 Malcolm Randall VA Medical Center, Gainesville, FL ) Rationale: Rapid identification of patients with independent bitemporal seizure onset zones could greatly reduce presurgical evaluation costs. We describe a novel method to identify patients with independent bilateral seizure foci based on a nonlinear characteristic of EEG signals. Signal nonlinear characteristic is quantified by the statistical difference of Short‐Term Maximum Lyapunov Exponent (STLmax) values estimated from the original EEG signals and its surrogate dataset. Methods: Eight adults with temporal lobe epilepsy (5 with unilateral temporal onsets and 3 with independent bilateral onsets) were included in this study. Long‐term EEG recordings from bilaterally placed depth and subdural electrodes (left and right temporal depth, subtemporal and orbitofrontal) were analyzed. Initial analyses involved the following steps: (1) each EEG signal was divided into 10.24 sec epochs and 10 surrogate EEG time series were generated, (2) STLmax values were estimated for each epoch of the original EEG signal and each of the 10 surrogate EEG signals, (3) T‐index, derived from pair‐T statistics, was calculated to measure the statistical difference in STLmax values between the original signal and mean values estimated from the 10 surrogate EEGs. Nested two‐way ANOVA was applied to test the significance of the brain region effect (patients were random blocks; seizures were nested within patient). Significance level = 0.05 was applied for rejection of the null hypothesis. Results: For each of the five patients with unilateral seizure onset, statistical test revealed that the nonlinear characteristics of EEG signals derived from mesial temporal electrodes within the seizure onset zone are significantly different from those from other areas (p < 0.01). In contrast, the nonlinear characteristics of EEG are not different (p > 0.05) among recording areas in all three patients who experienced bilateral seizure onsets. These differences were present throughout the recording, even before the first seizure was recorded. Conclusions: The results of this study suggest that it may be possible to efficiently and quantitatively determine whether a patient is likely to have unilateral temporal or independent bitemporal seizure onsets, based on analysis of the nonlinear characteristics of the EEG signal. In these patients, the findings were robust and persistent. Further studies in a large sample of patients will be required to determine whether or not this approach will have utility in selecting candidates for resective surgery. (Supported by NIH Grant RO1EB002089 and Department of Veterans Affairs.) 1StevenSchachter,2AliShoeb,3BlaiseBourgeois,4S. TedTreves, and2JohnGuttag(1Departments of Neurology, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, MA;2Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Boston, MA;3Departments of Neurology, Children's Hospital Boston, Boston, MA; and4 Nuclear Medicine, Children's Hospital Boston, Boston, MA ) Rationale: The accurate detection of electrographic seizure onsets with an ambulatory system could potentially be used to trigger an acute intervention to limit seizure progression/duration, and could also serve to notify a caregiver or disable harmful elements in the patient's environment (eg, a gas stove that is lit). Methods: We implemented a patient‐specific seizure detection algorithm (Shoeb A, et al. Epilepsy Behav 2004; 5:483–498) on a small digital signal processor (DSP; TM320C6711, Texas Instruments), and streamed previously recorded ambulatory EEGs from two patients to the DSP at a rate of 200 samples/sec per channel. For each EEG, we noted the sensitivity and specificity of seizure detections, and the latency with which the algorithm declared seizure onset following the initial electroencephalographic changes. Results: The algorithm trained on seizure‐free EEG epochs and 7 electrographic seizures uniformly sampled over the duration of the two EEG studies. The figure shows the average latency for detection of seizure onsets once the electroencephalographic changes began. The algorithm detected 70/72 seizure onsets (30/31 for patient #1, and 41/42 for patient #2), and declared 2 false detections (0 for patient #1, and 2 for patient #2) in 5.5 hours (2 hours for patient #1, 3.5 hours for patient #2) of ambulatory EEG epochs uniformly sampled over a 36‐hour period (20 hours for patient #1, and 16 hours for patient #2). The figure also compares how soon following electrographic seizure onset patient #1 became symptomatic (indicated by pressing the event button). Patient #2 was unaware of his seizures and therefore did not press the event button. Conclusions: These preliminary results demonstrate the high sensitivity and specificity of seizure onset detections using our algorithm and the feasibility of detecting seizures in the ambulatory setting within seconds of electrographic seizure onset and before symptoms occur. Further evaluation of our detection methodology on pre‐recorded as well as on‐line ambulatory EEG studies is underway. (Supported by Center for Integration of Medicine and Innovative Technology (CIMIT); MIT Project Oxygen Partnership.) 1ElenaUrrestarazu,1Jeffrey D.Jirsch,1PierreLeVan,1FrançoisDubeau, and1JeanGotman(1 EEG, Montreal Neurological Institute,McGill University, Montreal, QC, Canada ) Rationale: High frequency activity in the intracerebral EEG of epileptic patients has been described in recent studies using microelectrodes. We report changes in high frequency activity immediately following epileptic spikes in intracerebral recordings obtained from macroelectrodes. Methods: Eight patients with intractable epilepsy were evaluated; 4 had mesial temporal epilepsy (MT) and 4 neocortical epilepsy (N). EEGs from stereotaxically placed depth electrodes were filtered at 500Hz and sampled at 2000Hz. Spikes were classified according to their localization and morphology, with 60 spikes in each type. Segments of 256ms were selected immediately following each spike (post‐spike), and 2s before each spike (baseline). Log‐spectral power was estimated in 4 frequency bands: slow (0–40Hz), gamma (40–100Hz), high frequency (HF, 100–200Hz), and very high frequency (VHF, 250–500Hz). Changes between post‐spike periods and baselines were assessed for each channel (Wilcoxon test, p < 0.01). Results: 25 types of spike were identified (16 MT, 9 N). Changes were seen in 14 types of spike in HF (10 MT, 4 N) and in 11 types in VHF (8 MT, 3 N). Changes in VHF were always associated with changes in HF. In 9 types of spike (7 MT, 2 N), there was an increase in slow activity corresponding to a visible slow wave, but also a decrease in fast activity. These changes occurred in the hippocampus and neocortex. In 6 types of spike (5 MT, 1 N), there was an increase in slow activity despite the absence of a visible slow wave and an increase in fast activity. These changes occurred mostly in the amygdala, even though the spike was minimal in that area. The same pattern was observed in the hippocampus in two types of spikes in one patient. Finally, there was one type of spike with a decrease in fast activity without changes in slow activity. Overall, changes in VHF activity were spatially more restricted than changes in HF and gamma bands. Conclusions: A significant proportion of spikes were followed by changes in high frequency activity, more often in mesial temporal than neocortical spikes. Slow waves following spikes were generally associated with a decrease in HF and VHF compared to baseline. Changes in high frequency were clearest in channels where the spikes were most prominent, while changes at lower frequencies were more widespread. In many instances, there was an increase in high frequencies in the amygdala, even though the spike was not prominent in this region. This study demonstrates that high frequency activity can be recorded with macroelectrodes and has significant changes in epileptic discharges. The reduction in HF and VHF following spikes may reflect a profound inhibition that is more marked in the region of spike generation. The amygdala behaves differently from the hippocampus and neocortex, not showing this reduction in fast activity. (Supported by Canadian Institutes of Health Research Department of Education of Basque Government.) 1AnneleenVergult,1WimDe Clercq,1BartVanrumste,1SabineVan Huffel,2JohanVan Hees,2AndréPalmini, and2WimVan Paesschen(1Electrical Engineering, Katholieke Universiteit Leuven, Leuven, Belgium; and2 Neurology, University Hospital Gasthuisberg, Katholieke Universiteit Leuven, Leuven, Belgium ) Rationale: The aim of this study was to develop and investigate the performance of automated muscle artifact removal from ictal electroencephalograms (EEG) based on canonical correlation analysis (CCA). Methods: CCA separated a 10s EEG epoch in a set of components (or sources) with a decreasing autocorrelation index (AI). It has been noticed that muscle artifact components have a lower AI than genuine EEG components. A neurologist removed muscle artifact from 317 epochs of ictal EEG of 40 patients by gradually removing the components starting from the one with the lowest AI. The neurologist determined subjectively the separation point where as much muscle activity was removed as possible, without affecting the brain activity significantly. This yielded a gold standard binary classification of the CCA components into muscle artifact components and EEG components. Next a fully automated classification of each CCA component was performed based on the ratio of spectral energy of the frequency band 25–50Hz to the band 10–15Hz. It is anticipated a large value of this feature will be observed when muscle activity is present in a component. The optimal feature threshold was selected by maximally remaining EEG components and maximally suppressing muscle artifact components. Results: The developed automated method was able to remain on average 99±0.3% of the EEG component energy, while removing on average 71±33% of the muscle artifact component energy. Conclusions: An automated muscle artifact removal technique was presented, that could be used as pre‐processing step in the early detection of ictal activity or as a filter in the visual evaluation of ictal EEG.Currently the same EEG's are being processed by other neurologists, to investigate the user dependency of the optimal separation point to obtain a more generalized gold standard which can improve the robustness of the automated muscle artifact elimination. (Supported by European network of excellence BIOPATTERN (FP6‐2002‐IST‐508803) and Fund for Scientific Research Flanders FWO research project nr G.0360.05. Bart Vanrumste is funded by the ‘Programmatorische Federale Overheidsdienst Wetenschapsbeleid’ of the Belgian Government.) 1CesarViteri,1ElenaUrrestarazu,1JorgeIriarte,1ManuelAlegre,1MiguelValencia, and1JulioArtieda(1 Neurology and Clinical Neurophysiology, CUN, CIMA, Universidad de Navarra, Pamplona, Spain ) Rationale: Independent component analysis (ICA) is a system that differentiates components of complex signals. One of its capabilities is to study the epileptiform discharges to demonstrate similar or different origin of its components. The goal of this study was to discover how ICA analyse the discharges in the benign centrospikes, to understand better this kind of idiopathic epilepsy. Methods: We analysed 15 discharges of 3 patients with benign centrotemporal epilepsy. Their morphology was sharp wave, spike or spike‐and‐wave. The samples were recorded digitally with a 32‐channels Lamont amplifiers and Harmonie 5.2b program. ICA was applied using the JADE algorithm implemented in a Matlab platform. The components were identified visually. The suspected components were selected and the EEG of these components by itself and together were reconstructed. The topography in each component of the discharge was compared with the original using the BESA program in the same period. Results: The spikes and waves were located in different components. Spikes were separated in one (n = 10), two (n = 4) or three (n = 1) components, with delay between their maximums of 5–10 ms. The waves were isolated in other one or two components. Their topography was centrotemporal, very similar but not identical. The lateralization of each component was variable between components of the same spike, and it could be right, left or bilateral. Each component of the spike fit very well with the EEG in the correspondent time. The topography of the components from different spikes of the same patient varied slightly in the localisation. Conclusions: ICA differentiates several components with temporal evolution in the epileptiform discharges in centrotemporal epilepsy. In most of the discharges, the spike and the wave had different components suggesting different origin. Their topography was centrotemporal, similar but variable in the lateralization and localisation. In each patient the way of analysis of the discharge was very similar but not identical. 1ChunmaoWang,2IstvanUlbert,1Werner K.Doyle,1OrrinDevinsky,1RubenKuzniecky, and3,4EricHalgren(1Department of Neurology, New York University Comprehensive Epilepsy Center, New York, NY;2Institute of Psychology, Hungarian Academy of Sciences, Budapest, Hungary;3Massachusetts General Hospital NMR Center, Harvard Medical School, Charlestown, MA; and4 INSERM, Marseille, France ) Rationale: To study spatio‐temporal stages in word processing, both current source density in cortex laminars and subdural potentials were recorded. Methods: Laminar multiple microelectrodes and clinical macroelectrodes were chronically implanted in epilepsy patients undergoing subdural grid and depth implantation for seizure localization. The patients participated in several cognitive tasks during their normal alert status. Three of them were visual word recognition tasks, in which the patients were requested to decide if a word: (1) presented an animal or object longer than one foot; (2) rhymed in “AY; ” or 3) ended in “ED” for its past tense. Another task was auditory version of the size judgment task, in which a computer speaker instead of a monitor presented the stimuli. In all four tasks, half of the words were presented during practice stages prior to test, so repetition effect was expected. Results: Extensive overlapping responses were recorded in frontal, temporal, and occipital cortices. In visual cortex, the earliest response was peaking at ∼140 ms. For auditory stimuli, the earliest auditory response was in superior temporal cortex, peaking at ∼100 ms, and the largest response was in adjacent area [still in superior temporal cortex], starting from ∼120 ms and peaking at ∼300 ms. These responses were equivalent to new and repeated words. However, there were significant differential responses from ∼500–1000 ms to new versus repeated words, for both auditory and visual stimuli, mainly in medial and ventral temporal areas. Laminar electrodes in adjacent areas measured a sink and source complex, with a greater response to new words during the same time range. Conclusions: For both micro and macro recording, word repetition does not affect the initial processing stage but decreases later activity. These differential responses appeared to represent the N4‐P3b. (Supported by NS18741 and NS44623.) 1,2MatthiasWinterhalder,1,2BjoernSchelter,2ThomasMaiwald,2ArianeSchad,3ArminBrandt,1,2JensTimmer, and1,3AndreasSchulze‐Bonhage(1Bernstein Center for Computational Neuroscience, University of Freiburg, Freiburg, Germany;2Center for Data Analysis and Modeling, University of Freiburg, Freiburg, Germany; and3 Epilepsy Center, University Hospital Freiburg, Freiburg, Germany ) Rationale: Reliable and early prediction of epileptic seizures would open new routes for therapeutic interventions in patients with pharmacorefractory epilepsy. As seizures are characterized by an abnormal synchronization of neurons, multivariate time series analysis techniques detecting synchronized dynamics in invasive EEG recordings of epilepsy patients are a promising approach. Here, we have investigated two synchronization measures originating from the theory of Nonlinear Dynamics with respect to their potential role for epileptic seizure prediction. Methods: Two quantities measuring phase and lag synchronization have been applied to an invasive EEG data pool of 21 patients each with 24 hours of seizure‐free recordings and 2–5 pre‐seizure periods. Synchronization changes have been examined in different brain structures analyzing focal electrode contacts, i.e. early involved in ictal activity, as well as extra‐focal electrode contacts, i.e. not involved in ictal activity or only involved lately during seizure spread. The seizure prediction performance has been assessed by the methodology of the seizure prediction characteristic. The seizure prediction characteristic evaluates sensitivity of a prediction method with respect to its specificity and temporal aspects of a prediction. Comparing with the probability to predict seizures by chance, a test is utilized to decide about the statistical significance. Results: Compared to a random prediction, a significantly better seizure prediction performance has been shown for half of the patients. Both decreasing and increasing synchronization in the EEG dynamics could precede seizures. Regarding the topography of brain structures investigated, a statistical significant superiority of combinations between one focal and one extra‐focal electrode contact could be demonstrated for the lag synchronization measure (p < 0.05). Conclusions: Preictal changes in the synchronization of the EEG dynamics may offer a chance for epileptic seizure prediction. Our study shows that the seizure prediction performance differs between patients and strongly depends on the analysis technique applied and the brain structure investigated. (Supported by German Federal Ministry of Education and Research (BMBF grant 01GQ0420) and the Deutsche Forschungsgemeinschaft (Ti 315/2‐1).) 1JohnZempel,2Justin L.Vincent,2Linda J.Larson‐Prior, and2Abraham Z.Snyder(1Neurology, Washington University, St. Louis, MO; and2 Radiology and NeuroImaging Laboratory, Washington University, St. Louis, MO ) Rationale: The ballistocardiogram (BKG) is a significant source of artifact in scalp EEG recorded in the static magnetic field characteristic of human MRI scanners. Multiple methods (averaging, ICA, adaptive filtering) have been successfully employed by others to reduce the BKG. A new method of identifying BKG is used to biologically characterize the BKG across multiple electrodes and subjects. Methods: Simultaneous EEG and fMRI recordings were made from twelve subjects. Greater than than ten hours of EEG containing BKG. EEG‐fMRI was recorded in a 3 Tesla Siemens Allegra MR scanner using a Neuroscan (El Paso, TX) Synamps/2 amplifier and Maglink cap and cabling system with Scan 4.5 software. Standard 10/20 electrode positions were collected at a 20 kHz sampling rate and the gradient artifact greatly reduced with Scan 4.5 with decimation of the EEG signal to 500 Hz. 10–20 electrodes were referenced to a common electrode halfway between CZ and CPZ). Two additional processing steps were developed to reduce the BKG (Vincent, et al, Soc Neuroscience Abstracts 2005). First, a custom template based beat detector program, which identifies EKG patterns that represent single heart beats, was used to provide accurate triggering. Second, a moving general linear model was used to generate models of the BKG. The extracted BKG signal was characterized by standard averaging and forward FFT techniques. Results: The BKG waveforms recovered by the GLM technique consistently included components outlasting the mean inter‐beat interval in all (12/12) subjects and all 10–20 electrodes. Thus, as the waveforms due to successive beats inconsistently overlap, simple averaging cannot adequately characterize and reduce the BKG artifact. The BKG is large at 3 T. The positive to negative peak variation in the BKG is always more than 100 microvolts, and the largest components are most frequently several hundred microvolts in size in the O1‐Ref electrode. The frequency content of the BKG is complicated. In addition to the frequency of the heart rate itself, bimodal peaks at 3–5 Hz and 8–10 Hz are present in the BKG, which contaminate frequency ranges needed to recover standard EEG. When the power spectral density of each electrode is averaged across subjects, the frequency bin with peak power varies over the scalp. Conclusions: BKG recorded at 3 Tesla represents a significant electrical signal that must be successfully removed to obtain EEG that represents brain activity. In addition to removing the BKG, biological characterization of the BKG across multiple subjects allows determination of the nature and variability of this signal. BKG itself is spatially variable in size and frequency content across the scalp. These observations constrain theoretical accounts of BKG origin, which presently remain incomplete. (Supported by K12NS01690 for John Zempel and NS06833 (laboratory of Marcus Raichle).)