Prof. Dr. Ahmet Ademoğlu danışmanlığındaki tezler
12 tez · Boğaziçi University
Neuroimaging of brain activity using spatio-temporal signal modelling
Functional neuroimaging enables us to obtain information about how the brain responds to cognitive and/or emotional tasks.Neuroimaging of brain activity requires spatio-temporal modelling of measured electrical and/or hemodynamic data and integration of the measurements obtained at different spatial and or temporal scales. In this thesis, new techniques are employed for the investigation of spatio-temporal dynamics of different functional data as the EEG-ERP, the invasive/non-nonivasive recordings of epileptic EEG, and simultaneously recorded steady state EEG-fMRI. Spatio-temporal wavelet decompositions using realistic head models are applied in order to produce simple stationary input subtopographies for the source localization. Besides, a spatial decomposition method based on radial basis functions is used. The usage of the subtopographies facilitate the inverse solution and it is shown that even the temporally correlated EEG sources can be localized by this approach.Integration of the data obtained from different spatial scales is an important problem in epileptic EEG. To assess their reliability, the spatial performance of the scalp EEG based inverse solutions are compared with deep or cortical measurements and their simultaneously measured datasets. The multimodal functional information integration is proposed to compare the dynamics deduced by the simultaneously recorded SSVEP and fMRI. The temporal correlation between the time series of EEG and fMRI is calculated via the GLM. It is observed that the SSVEP source maps are the spatial subsets of the fMRI activity.The study demonstrates the applicability and potential of new spatio-temporal methods in EEG research which can be used to study cognition, attention, memory, and perception. Proposed methods can also be used as tools in more practical areas like brain computer interfacing, neurosurgical planning and neuro-psychological assessment of certain disorders.
Functional connectivity network analysis of alzheimer and mild cognitive impairment patients
In our era, while the life span is expanding, neurodegenerative diseases, such as Alzheimer's disease (AD), pose a great threat upon the quality of life. In such a case, the best course of action would be to detect, modify or treat the pathologies before they become too severe. Since the main cause of AD is still unknown, further studies for possible biomarkers are needed. Therefore, in this study, the objective is to find a distinctive agent for AD and mild cognitive impairment (MCI) from an optimized auditory oddball task fMRI data via functional connectivity analysis. In order to achieve that, a group ICA approach using temporal concatenation of the subject data is adopted. Since, there are no studies investigating functional connectivity of AD and MCI during an oddball task, especially via group ICA, this study can enrich the literature. As the results are concerned, in group comparisons, no significant differences are found in spatial maps. On the other hand, there are promising findings in temporal course analysis of the components such as the multiple regression outcomes. Therefore, our next aim will be to perform a longutidinal study including both resting state and task related data for finding a better biomarker.
Investigation of oscillatory mechanisms and thalamo-cortical circuitry of the visual system by simultaneous EEG-FMRI
Neural oscillation is an indispensable phenomena in the functioning of the cortical networks. Evoked neural oscillations triggered by external rhythmic stimulation mimic spontaneous ongoing oscillations, thus could shed light on the intrinsic specialization and tuning of the cortical networks. In this thesis, flickering light stimulation is used to constitute steady state for a wide range of temporal frequencies (6-46 Hz) during simultaneous electroencephalography (EEG) and blood oxygenation level dependent (BOLD) functional magnetic resonance imaging (fMRI) scans of 40 healthy volunteers. Firstly, thalamo-cortical loop of the visual system is the subject of interest. Our finding about high correlation between the frequency response characteristics of the lateral geniculate nucleus (LGN) and the primary visual cortex (V1) supports the oscillatory tuning property of the thalamo-cortical interactions. Secondly, contribution of oscillations in the modeling of hemodynamic response is discussed based on the sensitivity of BOLD components (phasic and tonic) to temporal frequency. Our results show that, tonic BOLD component is decreasing more sharply than phasic component with increasing frequencies pointing higher dependency of tonic BOLD response to the stimulation frequency. Finally, EEG informed fMRI analysis is conducted for the sake of testing resonance phenomena. The correlation maps between the BOLD responses and the steady state visually evoked potential (SSVEP) amplitudes show significant correlation for the beta and gamma bands but not for alpha band. This result supports the view that the global amplitude maximum of the SSVEP in the alpha band is due to the synchronization without synaptic activity (BOLD) increase.
Nörogörüntülemede tensör analizi
Acquisition of large amounts of data in neuroimaging research requires development of new methods that can disentangle the underlying information and reveal the features related to cognitive processes. This thesis attempts to propose new methods that favor the multimodality and multidimensionality of the brain data. The main difficulty for the fusion of imaging modalities is the discrepancies in their spatial and temporal resolutions as well as the different physiological processes they reflect. This problem is addressed by decomposing the EEG and fMRI data cast as tensors on both common and discriminant subspaces and computing the common spatial profile from the data on the cortical surface. The Granger causality analysis of brain connectivity is reformulated on tensor space enabling incorporation of tools developed in that area of research. The first approach on this analysis facilitated tensor methods for sparse representation of the connectivity patterns whereas the second method resolved them as atomic structures. General theory and computationally efficient algorithms are presented. The techniques are illustrated on the simultaneous EEG/fMRI recordings for the fusion model and on the fast fMRI data for the connectivity analysis. The proposed approaches may have a wide application area ranging from the early diagnosis of neurological diseases to the brain-computer interface studies.
Inter and intra-individual behavioral variability predicted by neural activity in the multiple demand network
A wide range of cognitive tasks consistently identify a Multiple Demand (MD) network in frontal and parietal brain regions. Its activity is closely linked to executive functions (EFs) such as attention, task switching, solving novel problems and manipulating information in the working memory. We here investigate the relation between MD neural activity and EFs using a large fMRI dataset (n=120). We examine this relation through two approaches (1) inter-individual variability, addressing several methodological challenges: We find that MD activity - which varies substantially across individuals, but is consistent within individuals across time - can explain a substantial proportion of variance in individual performance on a spatial working memory task such that individuals who find the task challenging, increase their MD activity substantially to improve their performance. This suggests that MD activity tightly reflects the executive demand of an individual. In the second approach we examine (2) trial-by-trial variability by employing three different models to fuse reaction time (RT) data with the BOLD time-series. We find that BOLD amplitude increases with longer RTs. This is consistent with the findings from the first approach showing increased MD activity for slower individuals. Together both findings support the view that within and between individual differences are manifested in the same brain regions. These results have implications for (1) understanding brain processes of EFs through ID studies (2) given that ID in EF are largely genetically determined, genetic variability can be linked to the neural activity of the MD network as an intermediate stage to link genetics with behavior (3) using ID in fMRI responses as clinical biomarkers.
EEG-based brain-computer interface using wavelet coherence and graph-theoretic methods
Emotion and cognition, two fundamental aspects of human mental processes, have traditionally been investigated independently. A comprehensive understanding of human mental functioning requires exploration of emotion and cognition. Among various physiological methods, electroencephalography (EEG), stands out as a reliable, non-invasive tool to evaluate emotional and cognitive processes. Studying neural oscillations across various EEG frequency bands and evaluating event-related potentials (ERPs) helps identifying distinct patterns linked to various emotional and cognitive states. In this thesis, pleasant/unpleasant emotional moods as well as low/high cognitive states were discriminated using graph-theoretic features extracted from spatio-temporal components. Wavelet coherence values corresponding to the N100, N200, and P300 single-trial ERP components were examined across the delta, theta, alpha, and beta frequency bands. Subsequently, graph-theoretic analyses were performed on connectivity matrices constructed from wavelet coherence-based functional networks to obtain several global and local graph metrics. These metrics were employed as features in classification using support vector machines (SVM), k-nearest neighbors (k-NN), and linear discriminant analysis (LDA) to enhance the classification accuracy by leveraging graph-theoretic features. Delta band wavelet coherence based graph features enabled accurate classification using SVM. Pleasant/unpleasant emotional moods and low/high cognitive states were classified with average accuracies of up to 92% and 89%, respectively. These findings emphasize the robustness of the proposed approach, highlighting its potential to improve EEG-based assessment of emotional and cognitive states, while paving the way for future brain–computer interface (BCI) applications.
Surface based morphometry in alzheimer's disease
Alzheimer's disease (AD) is a neurodegenerative disorder especially affecting the elderly population which is growing worldwide. In this study, surface-based morphometry analysis was performed on anatomical MR images of patients with Alzheimer's Disease (AD) and healthy control (HC) subjects using a computational anatomy toolbox called CAT(Computational Anatomy Toolbox) on SPM (Statistical Parameter Mapping) platform. MR images were obtained from a database named Minimal Interval Resonance Imaging in Alzheimer's Disease (MIRIAD) consisting of 46 AD patients and 23 HC subjects. The cortical thickness measurements were performed over 34 different regions on each hemisphere defined by Desikan-Killiany anatomical atlas. The t-statistics parameters of the cortical thickness values were found to be decreased in 24 regions in AD patients compared with the HC subjects. Additionally, the linear correlation values between the MMSE scores and cortical thickness values of AD and HC individuals were estimated for each atlas region. Accordingly, 28 regions exhibited a significant correlation between MMSE(Mini Mental State Examination) scores and cortical thickness values. Significant regions that were affected by AD were observed to be as parietal, temporal, frontal, cingulate and occipital lobes as reported in previous studies.
EEG data classification using multilinear regression model
Brain Computer Interfaces (BCI) are systems that facilitate people to use a computer, to control an electromechanical or a neuroprosthetic device without using their motor nervous system. It is possible to obtain an information about the brain tissues with electrodes placed on the skull which record the electrical activity called electroencephalogram (EEG) . The electrodes placed in di erent regions capture the activity in their neighborhood. BCI systems combining the electrical signals from these electrodes use signal processing and machine learning algorithms to identify the motor or the cognitive activity that is embedded in the brain signals so as to mobilize the peripheral devices according to the information gathered. Emotion estimation is often used in brain computer interface applications to improve and control the communication between man and machine. In recent years, emotion estimation studies based on brain electrical activity, which is the most widespread method used for accurate emotion analysis, have gained momentum. In this thesis study, multichannel EEG data taken from normal subjects who encountered emotionally pleasant and unpleasant pictures were classi ed with a multilinear regression algorithm. The results were compared with those of the Support Vector Machine (SVM) and proved to be better in accuracy.
Epileptic seizure prediction using machine learning and deep learning methods
Epilepsy is one of the most common neurological diseases in the world which negatively affects the daily life of a patient. Predicting epileptic seizures is of great importance for healthcare professionals and patients. The electroencephalography (EEG), which allows for registering brain activity with the help of electrodes placed on the scalp, is generally used to diagnose and monitor epilepsy. In this study, automatic seizure prediction was performed using CHB-MIT dataset which contains EEG data recorded at Boston Children's Hospital. Support Vector Machines (SVM), a common machine learning algorithm chosen as the primary method within this thesis's scope, and three different deep learning methods were compared. The first of these methods was long short term memory (LSTM) classifier with convolutional autoencoder which did not need any feature extraction. The second method used the spectrograms obtained by preprocessing the EEG data which were fed into a convolutional neural network (CNN) based classifier. The last method was based on converting the EEG data into three-dimensional images by applying source localization and performing classification with CNN. Among the methods used, the best result was obtained using source localization based CNN classification with 89.06% specificity, 92.58% sensitivity and 90.41% accuracy. Computational cost of three methods in terms of runtime efficiency were also compared, and it was observed that the SVM, which yielded the lowest classification performance with 74.07% accuracy, worked significantly faster than other methods.
Emotion recognition of EEG data using tensor logistic regression
Emotion recognition is a research area gaining momentum in the last three decades with a strong impact on our daily life. One of the most widely used methods to study emotion recognition is using physiological signals such as EEG data. However, using physiological signals requires using feature extraction and selection methods. Moreover, there is no gold standard for choosing the best methods. Therefore, this study aims to compare the sensor space and source space EEG data for emotion recognition using tensor based methods. In order to achieve that, different frequency bands were used as features of EEG data. In addition, support vector machine (SVM) as a conventional method, and logistic tensor regression (LTR), which was a tensor-based method, were used as two different classification methods. The results showed that the gamma was the most discriminating frequency band. Also, source space data improved the accuracy rates when compared with sensor space data. Moreover, TLR was superior in the source space than SVM. In the sensor space, both methods performed similarly.
Spatiotemporal graph convolutional neural networks for motor imagery EEG classification
Electroencephalography (EEG) has various applications in medicine, neuro-science, and neural engineering. It records the electrical activity of the brain tissues caused by the interaction among different neuronal communities. Numerous algorithms for the automatic classification of EEG signals have been developed. These algorithms work via extracting unique and non-redundant features from EEG signals. However, the majority of the proposed algorithms employ temporal components of EEG signals while disregarding the rich spatial network structure that underlies in the EEG. In this study, we propose a classification pipeline that uses the network structure of EEG data for a simultaneous representation of spatial and temporal features in the EEG signals. First, graph theory is utilized to model the EEG networks in two spatial domains; i) the sensor space and ii) the cortical source space. Second, a spatiotem-poral graph convolutional neural network (STGCNN) classification model is employed combining both temporal and spatial features of EEG data for its classification under motor imagery conditions. Additionally, the model is tested using the cortical source space data in each of the seven resting state brain networks individually to estimate their performance on classification accuracy. The results show that STGCNN model performs slightly better than the temporal convolutional neural network (CNN) models by 1.25%.
Motor imagery classification using multivariate empirical mode decomposition (MEMD) based functional connectivity features
Electroencephalogram (EEG) motor imagery signals are widely used for the implementation of brain-computer interfaces (BCI). Recently, functional connectivity measures have attracted attention as they can be used to capture statistical dependencies among EEG channels. However, functional connectivity during motor imagery tasks have not been fully explored. This study utilizes Instrinsic Mode Function (IMF) level phase locking value (PLV), coherence, and imaginary part of coherency in lefthand/right-hand motor imagery classification. EEG signals are decomposed into IMFs via noise-assisted multidimensional empirical mode decomposition (NA-MEMD), and connectivity metrics over the selected four channels are calculated for each trial as raw data and as a function of time and frequency. Resulting features are used to train multiple classifiers and their accuracy scores are analyzed. Best results are obtained using features derived from time-frequency functions of imaginary part of frequency where the overall accuracy score of 0.77 is achieved. This study shows that the change of connectivity throughout the duration of the task provides a more effective feature than connectivity calculated using each trial as raw data. Achieved accuracy scores are comparable to similar studies with the additional advantage of using few number of channels.