• Home
• Preliminary Program

14:00   Brain Chair: Sofia-Eirini Kotti 14:00 15 mins Enabling large-scale seizure detection with a tensor-network Kalman filter for LS-SVM Seline de Rooij, Kim Batselier, Borbála Hunyadi Abstract: Due to recent advances in wearable EEG devices, accurate seizure detection algorithms have become even more important. The ever-increasing size of the generated datasets poses a significant challenge to many existing seizure detection methods based on kernel machines such as SVMs. Typically, this problem is mitigated by significantly undersampling the majority class, but in practice, these methods tend to suffer from too many false alarms. Recent works have proposed tensor networks to enable large-scale classification with kernel machines. In our work, we explore the use of one such method, the tensor-network Kalman filter for LS-SVMs (TNKF-LSSVM), in seizure detection, as we hypothesize that using more data will improve the detection performance. This hypothesis was tested using data from the Temple University Hospital Seizure Corpus (TUSZ). Our results show that the TNKF-LSSVM performs comparably to a regular LS-SVM in detecting seizures when both are trained on the same (smaller) dataset (AUCLS-SVM = 0.797 vs. AUC¬TNKF-LSSVM¬ = 0.811). We can also show that it can be trained on truly large-scale data where LS-SVM cannot (N>10^5 ). However, for the presented model configuration detection performance does not seem to improve with more input data. 14:15 15 mins Influence of anisotropic electrical conductivity in white matter tissue on the EEG source reconstruction accuracy Stefan Dukic, Boudewijn T.H.M. Sleutjes, Leonard H. van den Berg Abstract: Source reconstruction of brain activity using EEG/MEG data is becoming an established tool in neuroscientific research. Recently, using source analysis on EEG data we have shown multiple networks that are affected in amyotrophic lateral sclerosis (ALS)[1], some of which we have not observed using sensor analysis[2]. This approach, however, requires a volume conductor model of the human head that mimics the electromagnetic properties of the investigated subject as accurately as possible. These models often assume isotropic conductivity tensors across different tissues, which is known to be incorrect in particular for the white matter and skull. Here, we investigate the influence of white matter anisotropy derived from diffusion tensor imaging (DTI) data on the dipole estimation error. To construct an anisotropic head model, a T1-weighted and a DTI dataset were acquired from a healthy volunteer (female, 60 years). We segmented the MRI scan into six compartments (i.e. grey and white matter, cerebrospinal fluid, skull, scalp and air cavities). For the finite element mesh generation hexahedral elements were used. Based on this geometry, two models were made: 1) a simple model with isotropic conductivity tensors assigned to elements belonging to the same tissue type and 2) an advanced model with anisotropic conductivity tensors assigned to elements belonging to the white matter. Anisotropic conductivity tensors were estimated using DTI-derived diffusion tensors and the linear relationship between the two estimates[3]. The dipole location error is determined by applying dipole fitting 125 times on each model using simulated EEG data (15 Hz sinusoid originating from the right putamen). Using Euclidean distance, the localisation error was on average 6.28 mm (range: 0.13 - 44.43 mm) for the simple model and 6.07 mm (range: 0.09 - 32.71 mm) for the advanced model indicating slight advantages for the advanced model. This study shows evidence of the importance of white matter anisotropy modelling in healthy individuals. Accounting for white matter anisotropy is likely to have an even greater impact in diseases that affect white matter, such as ALS. Additional analyses that use more repetitions (>125) and that assess other dipole locations (beyond the putamen) are warrant. [1] Dukic, S. et al. Patterned functional network disruption in amyotrophic lateral sclerosis. Hum. Brain Mapp 40, 4827–4842 (2019). [2] Nasseroleslami, B. et al. Characteristic increases in EEG connectivity correlate with changes of structural MRI in amyotrophic lateral sclerosis. Cereb. Cortex 29, 27–41 (2019). [3] Rullmann, M. et al. EEG source analysis of epileptiform activity using a 1 mm anisotropic hexahedra finite element head model. Neuroimage 44, 399–410 (2009). 14:30 15 mins Near-infrared spectroscopy for the measurement of cerebral autoregulation during carotid end-arterectomy (NICACEA-study) – initial experiences Nick Eleveld, Gea Drost, Anthony Absalom, Clark Zeebregts, Jean-Paul de Vries, Natasha Maurits, Jan Willem Elting Abstract: Background: Dynamic cerebral autoregulation (DCA) is the brain’s ability to maintain adequate cerebral perfusion in the face of blood pressure changes over time. DCA is traditionally quantified with transcranial doppler (TCD) and arterial blood pressure (ABP) measurements. We have recently developed a simpler DCA-method based on near-infrared spectroscopy (NIRS-DCA) measurements1 and are currently validating this method in a clinical population undergoing carotid end-arterectomy (CEA, plaque removal). CEA is an interesting clinical model for validation because cross-clamping of the carotid artery can be required, which leads to a profound change in cerebral perfusion. Methods: This is an ongoing two-centre prospective observational study in 50 patients undergoing CEA. After the induction of anaesthesia, we perform continuous-wave NIRS-measurements on the bilateral frontotemporal forehead with a multi-distance device (Brite MKII). We obtain the blood flow velocity in both middle cerebral arteries with TCD and intra-arterial ABP waveforms. End-tidal CO2 and other haemodynamic variables are monitored intermittently. Our primary comparison is TCD+ABP-based DCA (low-frequency phase shift) versus NIRS-based DCA (corrected low-frequency phase shift), before and after cross-clamping of the carotid artery. To investigate the influence of extracerebral tissue (scalp, skull) on the NIRS-measurements, we compare the results of source-detector (SD) pairs at four SD distances (1, 3, 4, 5 cm). Initial results and discussion: To date 15 measurements have been performed. Our initial results show that TCD and NIRS data quality have been variable. Ten measurements showed movement related artifacts and signal loss in TCD. Movement artifacts were present to a variable degree in most NIRS-measurements. ABP-data was of excellent quality. Artifacts were mostly related to movement of the sensors in the surgical process. To allow reliable identification and correction of the artifact segments, we currently explore multivariate signal identification techniques.2, 3 Multivariate techniques should be able to exploit the strong overlap in signal content that is present in physiological artifact-free TCD, NIRS, and ABP data. We will show the initial results on our obtained data. References: 1. Elting JWJ, Tas J, Aries MJH, Czosnyka M, Maurits NM. Dynamic cerebral autoregulation estimates derived from near infrared spectroscopy and transcranial Doppler are similar after correction for transit time and blood flow and blood volume oscillations. J Cereb Blood Flow Metab SAGE Publications Sage UK: London, England; 2020; 40: 135–49 2. Rehman N, Mandic DP. Multivariate empirical mode decomposition. Proc R Soc A Math Phys Eng Sci [Internet] Royal Society; 2010; 466: 1291–302 Available from: https://doi.org/10.1098/rspa.2009.0502 3. Rehman N u., Aftab H. Multivariate Variational Mode Decomposition. IEEE Trans Signal Process 2019; 67: 6039–52 14:45 15 mins Prediction of depression symptom improvement based on multi-echo functional MRI Jesper Pilmeyer, Rolf Lamerichs, Marcel Breeuwer, Sveta Zinger Abstract: Objective prognosis of major depressive disorder (MDD) based on functional MRI (fMRI) biomarkers remains problematic due an abundance of physiological and motion confounders and susceptibility artifacts in deeper located subcortical and inferior anterior regions. Increased amygdala activity during negative emotional face-matching tasks is often reported in patients with MDD. Yet, studies predicting longitudinal symptom improvement in MDD are scarce. Multiband multi-echo acquisitions improve the BOLD sensitivity, reduce signal losses in regions prone to susceptibility artifacts and allow for improved spatial or temporal resolution. In this work we acquired fMRI scans of an emotion-related task with the aim of predicting 3- months and 6-months MDD symptom improvement. Thirty-two MDD patients underwent MRI examination at baseline, including a T1-weighted and a multiband multi-echo fMRI acquisition. During the fMRI scan, patients performed the Hariri task, a well-validated emotional facematching paradigm that includes blocks of rest, shapes and sad or angry faces. The Hamilton Depression Rating Scale (HDRS) was obtained at baseline, 3-months and 6-months follow-up to assess depression severity. Patients with a HDRS_follow-up ≤ 50% compared to HDRS_baseline were classified as responder, whereas the others were labelled as non-responder. Temporal signal-to-noise ratio (tSNR) and t-values were calculated for several multi-echo combinations and single-echo (echo 2) to compare data quality and face-related activation contrast, respectively. Binary classification between response groups was performed using polynomial support vector machine classifiers and validated by leave-one-out cross validation. The features were activation contrasts of faces-rest and faces-shapes in both amygdalae and hippocampi. The tSNR was the highest for multi-echo combinations in all regions-of-interest. The majority of subjects showed contrast enhancement of minimally 10-20% for multi-echo combinations compared to single-echo. Furthermore, multi-echo based features predicted 3-months response with 91% accuracy. The 6-months response could be predicted with 87% accuracy by singleecho derived features. Based on a multiband multi-echo sequence, we showed overall improvement in signal quality and emotion-related activation contrast in the amygdala and hippocampus compared to singleecho. 3-months and 6-months response in MDD could be predicted with high accuracy based on these features. This demonstrates the potential of multiband multi-echo fMRI for prognosis in psychiatric disorders. 15:00 15 mins Modeling and inference of dynamic functional connectivity networks from functional ultrasound data Ruben Wijnands, Justin Dauwels, Ines Serra, Pieter Kruizinga, Aleksandra Badura, Borbála Hunyadi Abstract: Functional ultrasound (fUS) is a novel large-scale brain imaging technique that measures hemodynamic responses as a time series of images. Thereby, fUS measures neural activity indirectly through the neurovascular coupling (NVC). Often, such a time series of images is used to analyze dynamic functional connectivity (dFC) by directly computing a connectivity metric between the measured hemodynamic signals, ignoring the functional connectomics of underlying neural populations. This work proposes a novel fUS signal model, consisting of a hidden Markov model (HMM) cascaded with a convolutive model, that captures how fUS signals arise from a generative perspective while incorporating high-level biological functioning of neural populations. Consequently, the developed model enables inference of functional connectivity networks that are here defined as coactivation patterns (CAPs) of neural populations at a certain time point. CAPs are inferred by first estimating the activity of neural populations that underlie the observed fUS signals through a deconvolution procedure using the non-negative least absolute shrinkage and selection operator (NNLASSO). Then, using expectation maximization (EM), recurring neural CAPs and their transition preferences are learned from the reconstructed activity of neural populations. Our results show that our model and corresponding methods can identify biologically plausible networks of functional connectivity. Furthermore, this method captures a difference in brain dynamics between wild-type and Shank2-/- mouse mutants. 15:15 15 mins Simulation of electric fields for dual-site transcranial alternating current stimulation in motor cortices Silvana Huertas-Penen, Marieke Rona, Tjitske Heida, Bettina C. Schwab Abstract: In the past years, transcranial alternating current stimulation (tACS) has been investigated as an intervention for different neurological conditions/diseases. There is high variability between tACS studies depending on the subjects and the settings of stimulation. This has no clear explanation, but understanding the physiological mechanisms behind the effects of tACS could help compensate for the variability. Analysing the variations in brain functional connectivity when changing the settings of dual-site tACS in the motor cortices, could help to identify the physiological mechanisms behind it as well as understanding the role of stimulation settings in steering the brain's functional connectivity. As a first step toward studying the functional connectivity of the brain in dual-site tACS, our research will determine which stimulation montage to use, including the positioning and size of electrodes to avoid spatial overlap of the electric fields of the motor cortex. The stimulation intensity used was 1.5 mV (zero-to-peak). This value was chosen because multiple studies have shown that to influence the brain, the intensity needs to be above 1mV. However, without local anaesthetics, participants feel high discomfort at intensities equal to or above 2mV. The criteria to select the position and size of electrodes were to maximize the field strength and focality of the electric field in the motor cortices. In addition, the field strength must be the same in both cortices. Finite element method (FEM) simulations were performed using SimNIBS. We simulated more than 20 different montages of stimulation electrodes (ring-electrodes, circular electrodes and pad electrodes) using the ICBM152 MRI head model template. We found that having 110/90 mm outside ring electrodes and 25 mm inside electrodes, located in the C3 and C4 EEG electrode positions, had the most favourable compromise between electric field intensity in the motor cortices and avoidance of spatial overlap of electric fields. With these results, the next steps are to experimentally stimulate participants while changing the settings of the stimulation between cortices. We will also generate computational neural network models based on the hypothesized physiological mechanisms of tACS and compare their results with the experimental ones.

end %-->