• Home
• Preliminary Program
• Keynote speakers
• Registration & Venue
• Company Market/sponsoring
• Important Dates
• Awards
• Organizing Board
• Conference Topics
• Call for Abstracts
• Contact

10:30   EEG/fMRI I Chair: John van Opstal 10:30 15 mins A NEW TOOL FOR AUTOMATIC DETECTION AND VISUALIZATION OF INTRACRANIAL ELECTRODES ON SUBJECT CORTICAL SURFACE SPACE Mariana Branco, Anna Gaglianese, Daniel Glen, Ziad Saad, Dora Hermes, Natalia Petridou, Nick Ramsey Abstract: Electrocorticographic (ECoG)-based Brain-Computer Interface (BCI) systems require to accurately localize implanted cortical electrodes with respect to the subject’s neuroanatomy. Electrode localization is particularly relevant to understand the area recorded from, hence providing an optimal control of neuroprosthetic devices. Yet, this problem has been shown to be non-trivial [1-3]. Current procedures require either a time-consuming detection and transcription of the electrodes coordinates from the CT volume scan or a combination of several different software programs. Here we propose an improved, and faster method that automatically detects electrodes on the post-operative high-resolution 3D CT scan using a 3D clustering algorithm from AFNI (http://afni.nimh.nih.gov/afni) [1]. In the pipeline we also incorporate individual cortical surface estimation using FreeSurfer (http://surfer.nmr.mgh.harvard.edu) and brain shift correction as in [3]. The current pipeline consists of the following steps: 1) high-resolution post-operative CT alignment to the pre-operative MRI anatomy using the local Pearson correlation cost function implemented in AFNI (align_epi_anat.py function); 2) automatic electrode localization on the aligned CT scan using 3D clustering detection function provided by AFNI (3dclust function); 3) electrode coordinates extraction and simultaneous labeling obtained by computing the center-of-mass of each cluster in a 3D space (implemented in SUMA); 4) brain shift correction by projecting the electrodes on the surface of the cortex [3] obtained by the FreeSurfer segmentation; 5) projected electrodes visualization on the brain surface rendering. Validation of the pipeline was carried out in five patients with intractable epilepsy implanted with standard clinical grids (8x8 grid array, 2.3mm exposed surface, 1cm center-to-center inter-electrode distance; Ad-Tech, Racine, USA). Results were compared to the ones obtained with the currently available method developed by Hermes et al. [3] on the same subjects. Mean Euclidean distance and standard errors between electrodes position were 1.03±0.10mm, 2.08±0.11mm, 1.39±0.11mm , 1.52±0.14mm and 2.7±0.1mm , for each subject respectively. Our pipeline allows for an automatic and accurate localization of ECoG clinical grid electrodes on the brain surface. Moreover, the method has excellent potential to automatically detect electrodes on high-resolution ECoG grids using the post-operative CT scan, hence allowing for a more accurate projection. This procedure will provide an important tool for ECoG in neuroscience allowing an easy and accurate overlap between electrophysiological recordings and non-invasive brain mapping techniques to better understand brain functioning. 10:45 15 mins AN AGE PREDICTION MODEL FOR PRETERM NEONATES BASED ON BRAIN CONNECTIVITY Mario Lavanga, Ofelie De Wel, Alexander Caicedo, Sabine Van Huffel, Gunnar Naulaers, Katrien Jansen, Anneleen Dereymaeker Abstract: Recent studies have shown that brain connectivity is an important feature to assess maturity and development in preterm neonates [1]. In this work, we investigate how functional connectivity (FC) evolves during early brain maturation. The objective is twofold: we aim to provide a model able to predict age in premature neonates, as well as to understand how brain connectivity is affected by maturation in first stages of life. This study uses data from a previous study published elsewhere [2]. FC is assessed through the means of mean squared coherence (MSC), phase locking value (PLV) and activity synchrony index (ASI). Each connectivity measure is used, separately, to predict the postmenstrual age (PMA) of the infant. This procedure reveals that coherence in beta bands and ASI are the best predictors for PMA. These features are then combined in a single multivariate linear regression model to produce a single multivariate model for the prediction of PMA. The prediction performance shows the root mean square error equal to 2.05 weeks for the regression model using the best predictors. Further, the results indicates a decrease/increase in coherence/ASI with age. This can be due to the shift from the thalamo-cortical connections to cortico-cortical connections, leading to more localized and task dedicated networks. Finally, a larger correlation of ASI and coherence is found in the left hemisphere compared to the right hemisphere. As expected multimodal approaches outperforms the models produced using a single feature modality. REFERENCES [1] E. J. Meijer, K. H. M. Hermans, A. Zwanenburg, W. Jennekens, H. J. Niemarkt, P. J. M. Cluitmans, C. Van Pul, P. F. F. Wijn, and P. Andriessen, “Functional connectivity in preterm infants derived from EEG coherence analysis.,” Eur. J. Paediatr. Neurol., vol. 18, no. 6, pp. 780–9, 2014. [2] N. Koolen, A. Dereymaeker, O. Räsänen, K. Jansen, J. Vervisch, V. Matic, G. Naulaers, M. De Vos, S. Van Huffel, and S. Vanhatalo, “Early development of synchrony in cortical activations in the human,” Neuroscience, vol. 322, pp. 298–307, 2016. 11:00 15 mins TMS EVOKED EEG POTENTIALS DEMONSTRATE ALTERED CORTICAL EXCITABILITY IN MIGRAINE WITH AURA Matthijs Perenboom, Robert Helling, Prisca Bauer, Else Tolner, Gerhard Visser Abstract: Migraine is associated with altered processing of sensory input that may be due to cortical hyperexcitability. The visual cortex is believed to be particularly affected, based on the hypersensitivity of migraine patients to light and occurrence of visual premonitory signs (aura) preceding the headache phase in a proportion of migraine patients. Transcranial magnetic stimulation with concomitant electroencephalography recordings (TMS-EEG) is a new method to measure cortical excitability from the direct response to non-invasive stimulation over the skull. Recent studies have shown that phase clustering in EEG responses is linked to cortical excitability. We quantified differences in TMS evoked EEG potentials (TEP) between healthy controls and patients with migraine with aura, to study TEP’s possibility as cortical excitability biomarker in migraine. We included nine patients with migraine with aura and nine age- and sex-matched healthy controls. All underwent single-pulse TMS on the vertex with simultaneous 64-channel EEG recording. Migraine patients were recorded interictally (at least three days before and after an attack). On average 300 pulses were delivered between -8% and +8% of the resting motor threshold. We compared averaged TEP waveforms, power spectrum and phase clustering over trials between the groups of participants. TEP waveforms differed between migraine patients with aura and healthy controls around the N100 and P150 peaks, mostly located at frontal and occipital regions respectively. Hundred ms after the stimulus, phase clustering in the occipital lobe remains stronger in healthy controls than in patients, indicating reduced phase consistency after the N100 peak in migraine patients. Patients with migraine with aura show different cortical responses to non-invasive magnetic stimulation compared to healthy controls. This suggests that cortical excitability is altered, also between migraine attacks. Our findings are in line with studies that used indirect cortical stimulation with e.g. visual or somatosensory inputs and magnetic stimulation with peripheral readouts. We conclude that TMS-EEG possibly is a suitable method to directly study changes in cortical excitability during the migraine cycle. 11:15 15 mins FLEXIBLE HEMODYNAMIC COUPLING IN EEG-FMRI: CIRCUMVENTING THE GLM BY STRUCTURED COUPLED MATRIX-TENSOR FACTORIZATION Simon Van Eyndhoven, Borbála Hunyadi, Lieven De Lathauwer, Sabine Van Huffel Abstract: Simultaneous measurement of EEG and fMRI is gaining importance in brain studies, thanks to the highly complementary spatiotemporal resolution of the two involved modalities. The General Linear Model (GLM) has been a standard workhorse to jointly analyse the heterogeneous datasets that stem from these measurements, and although this model is widely applied, it relies on several assumptions. In the ‘standard’ approach, a single ‘canonical’ hemodynamic response function (HRF) is used, which describes the expected BOLD signal in fMRI resulting from an impulse event in a reference stimulus time series. Hence, the HRF waveform is considered invariant over the different brain regions and even over different subjects in a group study. We argue that, although the GLM has its merit as a simple and attractive model to make sense of multimodal neural data, this restriction obfuscates the interpretations that can be drawn through its use. On the other end of the spectrum, blind source separation (BSS) techniques such as independent component analysis can be used to decompose both EEG and fMRI data into a set of constituting sources, in which case (by definition) no model of the neurovascular coupling is assumed. We aim to tackle this issue by taking a hybrid data-driven estimation approach, that still models the fMRI time series analogously to the GLM. That is, we first structure the EEG and fMRI data as a third-order tensor with temporal, spatial and spectral modes, and a matrix with temporal and spatial modes, respectively. Through a coupled matrix-tensor factorization (CMTF), we can blindly identify spatial, temporal and spectral patterns of the underlying neural sources of the data. However, by imposing a structural coupling relationship between the temporal patterns of the EEG sources and those of the fMRI sources (convolution with an unknown HRF) we can incorporate extra knowledge about the signals’ characteristics, compared to more simple BSS techniques. In this way, the ‘true’ HRF itself is found in a data-driven fashion, without requiring prior knowledge. We provide extensions to our method, to account for spatial modulation and smoothness of the HRF wavevorm. As a proof of concept, we show the feasibility of the proposed approach in the case of synthetic, multimodal datasets. As future work, we will apply this method on a real dataset of EEG-fMRI recordings from epileptic patients, where we aim to localize the irritative zone, responsible for the generation of interictal epileptic discharges. Results will then be compared to those which are obtained through the standard GLM approach. REFERENCES [1] M. Monti, “Statistical analysis of fMRI time-series: a critical review of the GLM approach”, Frontiers in Human Neuroscience, 5.28, (2011). [2] E. Karahan et al., “Tensor analysis and fusion of multimodal brain images”, Proceedings of the IEEE, 103.9, (2015), 1531-1559. 11:30 15 mins EEG-FMRI FUSION FOR EXPLORING EPILEPTIC NETWORK ACTIVITY Borbála Hunyadi, Maarten De Vos, Wim Van Paesschen, Sabine Van Huffel Abstract: The ultimate goal of the EEG-fMRI analysis in refractory focal epilepsy is the precise localization of the epileptogenic zone (EZ) to facilitate successful surgery. Many studies have shown that EEG-correlated fMRI analysis can identify fMRI voxels which covary with the timing of interictal spikes assessed on EEG. However, this type of analysis often does not reveal a single focus but an extensive epileptic network. We propose [1] to unravel the extensive network using a symmetric data fusion approach, based on the following assumptions. We consider various brain sources which are active during the recordings. Both modalities capture a linear mixture of the true underlying source activities. The mixing system is determined purely based on the relative strength of each source, therefore, it is the same for both modalities. In this case, the underlying sources can be retrieved using joint blind source separation of the EEG and fMRI data. We solve the source separation problem using different variants of joint independent component analysis (jointICA) and coupled matrix-tensor factorization (CMTF). The feasibility of the proposed approach was evaluated on a dataset recorded from 10 refractory temporal lobe (TL) epilepsy patients. Interictal epileptic discharges (IEDs) were marked by an experienced neurologist. Multichannel average IED matrices from all patients were stacked together in a tensor, or were vectorised and concatenated into a matrix. Subsequently, traditional EEG-correlated fMRI analysis was performed to obtain fMRI activation maps for each patient. These maps were vectorised and stacked together in a matrix. The EEG tensor (matrix) and fMRI matrix was jointly decomposed using CMTF (jointICA) into 2 terms, after automated rank estimation. Using either approach, the first component captured the initial spike part of the IED (i.e. onset), as assessed based on both the temporal and the channel distribution. The corresponding voxel signature shows activation in the right TL (corresponding to the EZ) and deactivations in the default mode network. The second component captures slow wave activity (i.e. propagation or inhibition), and the corresponding voxel signature shows, besides weaker activation in the right TL, activations in the left TL and the occipital lobe; and a deactivation pattern resembling the executive control network. Finally, we showed that CMTF produces the most robust results in terms of the reproducibility of the signatures against different input datasets, created by excluding randomized subsets of patient data. We conclude that CMTF can provide a reliable and detailed spatiotemporal characterization of the interictal epileptic networks, which may lead to new and important insights in epileptic network behaviour. Future work will aim to extend the approach to characterizing single-subject EEG-fMRI data for a prospective localization of the EZ. REFERENCES [1] Hunyadi, B., Van Paesschen, W., De Vos, M., Van Huffel, S. (2016). Fusion of electroencephalography and functional magnetic resonance imaging to explore epileptic network activity. Proc. of the European signal processing conference. EUSIPCO 2016. Budapest, Hungary, Aug. 2016 (pp. 240-244) 11:45 15 mins A FULLY IMPLANTED BRAIN-COMPUTER INTERFACE FOR COMMUNICATION IN LOCKED-IN PATIENTS BASED ON ELECTROCORTICOGRAPHY Mariana Branco, Zachary Freundenburg, Elmar Pels, Erik Aarnoutse, Sacha Leinders, Mariska van Steensel, Nick Ramsey Abstract: People with severe paralysis who have lost the ability to communicate have only limited options to regain this ability. In the last decade Brain-Computer Interfacing (BCI) has been proposed as an assistant technology to reestablish this lost communication [1-2]. The principle behind a BCI is to record activity directly from the brain and to translate it into an input signal to control a device. For optimal usability in daily life at the homes of the target population, such a system should be accurate and intelligent (i.e., it incorporates smart decoding algorithms that dynamically adjust to e.g. slow signal changes), fully implantable (i.e., permanently available and invisible), safe, stable, easy and comfortable to use [3-5]. However, even though technology advances fast, many of these requirements have not been met so far. One approach that could, conceptually, meet the requirements of a BCI for home-use is an intracranial electrocorticography (ECoG) based system. ECoG in humans has recently received an increased interest since it provides a direct measure of neuronal activity in the human brain [6]. This technique benefits from a very good spatial and temporal signal resolution [6], which allows for an accurate distinction between complex mental paradigms. Here we present a fully implantable system designed to restore communication in completely paralyzed patients, based on ECoG signals recorded from the sensorimotor cortical hand region. This system translates the acquired signal, recorded from electrodes spaced 1 cm apart, to a real-time “click” control signal that is wirelessly sent to a computer to control a spelling program. This “click” control signal allows to make selections in a spelling matrix, thereby allowing the user to self-initiate communication using brain signals. However, usability and speed of communication could be improved by increasing the degrees-of-freedom of the BCI, by for example using two separate clicks to control a row and column on the spelling matrix. In order to determine the feasibility of a multiple degrees-of-freedom ECoG-based BCI, we are currently investigating the possibility of using small high density ECoG grids (electrodes only 3 mm apart) to translate finer mental acts (than just coarse hand movement) into independent control signals. In fact, we show a proof-of-principle strategy where four different hand gestures could be decoded [7] from a high-density grid, placed subdurally over the primary sensorimotor hand area. In a nutshell, the research of BCI technology, together with the increasing knowledge about detailed organization of brain functions, promises to provide a new assistive technology for patients who lost the ability for verbal communication.