The experiments described in this thesis aimed to investigate the neurophysiological consequences, at the cortical level, of the neurodegenerative condition, amyotrophic lateral sclerosis (ALS). A principle tenet of this study was that ALS is, first and foremost, a disorder of the cortical motor system, the precise pathological mechanisms of which remain incompletely understood. Furthermore, the degree to which neurodegeneration can be evidenced before the onset of symptoms is thus far uncertain, and the optimal means by which to measure therapeutic response has yet to be determined. Chapter 1 introduces relevant key concepts in ALS and briefly summarises three studies completed in the early phases of pursuit for the above degree. These studies respectively considered presmyptomatic cellular ALS pathology, quantitation of disease progression and eyetracking assessment of cognitive dysfunction. Chapter 2 describes magnetoencephalography, the investigative technology utilised in the subsequent experimental chapter. In chapter 3, the effects of ALS on movement related modulation of neuronal oscillations are determined. An excessive peri-movement desynchronisation and delayed post-movement rebound was described. Functional connectivity between cortical regions at rest is appraised in chapter 4. ALS appeared to result in quite striking increases in functional connectivity, in keeping with the fMRI literature and in support of diminished intracortical inhibitory influences. The functional communication from the motor cortices is directly considered during active motor performance in chapter 5. ALS related reductions in beta-band coherence were noted in both corticospinal and inter- hemispheric communication. In conclusion, the results demonstrated considerable support for proposed excitotoxic disease mechanisms and were in alignment with reported findings in other neurodegenerative diseases. Finally, a pilot study by which the neural mechanisms for cognitive impairment in ALS are explored via antisaccade performance is described. While underpowered, the experimental design showed promise for future application.

A Neuroimagem Funcional evoluiu muito nos últimos anos com o aparecimento de técnicas como Positron Emission Tomography ou PET (Tomógrafo por Emissão de Pósitrons) e Functional Magnetic Ressonance Image ou fMRI (Imagem de RessonÂncia Magnética Funcional) [Belliveau et al., 1991]. Elas permitem a observação de atividade no cérebro com uma resolução de alguns milímetros, e devido a natureza do sinal medido, com uma resolução temporal da ordem de 5 segundos [Kim et al., 1997]. Magnetoencefalografia e Eletroencefalografia (M/EEG), por outro lado, possuem uma resoluçao temporal da ordem de milissegundos, já que o sinal é produzido pela movimentação do íons através das membranas celulares [Nunez and Srinivasan, 2006]. Porém a sua resoluçeo espacial é muito baixa jé que tipicamente são problemas mal postos, com muito mais variáveis do que dados. Um equipamento de M/EEG de alta resolução possui da ordem de O(200) canais, que permitem medidas do campo magnético (para o MEG) ou do potencial elétrico (para o EEG) em O(200) posições em torno da cabeça. Para uma escala com resolução de ordem l existem (L /l )3 variáveis, onde L = aprox. 15cm. Neste trabalho procuramos estudar métodos para aumentar a resolução espacial das técnicas de EEG, pois o mapeamento funcional do cérebro humano esta intimamente relacionado à localização da atividade no espaço bem como no tempo [Friston, 2009] (muitas relativo ao momento de um estímulo externo). Todo o trabalho de localização de fontes para EEG pode ser facilmente estendido para MEG. Métodos Bayesianos são o cenário natural para lidar com problemas mal postos [Wipf and Nagarajan, 2009]. Existem, essencialmente, duas direções nas quais os algoritmos Bayesianos podem ser melhoradas, através da construção de uma melhor verossimilhança ou uma distribuição a Priori. Embora reconheçamos que avanços importantes podem ser feitos no direção anterior, aqui nos concentramos na segunda. Neste trabalho nós introduzimos um método multiescala para construir uma melhor distribuição a Priori. Uma idéia similar foi estudada dentro do contexto mais simples de fMRI [Amaral et al., 2004]. Muitos novos problemas aparecem ao lidar com o caráter vetorial do EEG. O mais importante, é a construção de um conjunto de superfícies renormalizadas que aproximam a região cortical onde a fonte de atividade esta localizada e o problema relacionado de de nir as variáveis relevantes para representar o cérebro em uma escala com menor resolução. A validação do novo algoritmo é sempre um problema essencial. Nós apresentamos resultados que sugerem, em dados simulados, que nosso método pode ser uma alternativa válida para os atuais algoritmos, julgando ambos pela taxa de erros na localização de fontes bem como pelo tempo que eles levam para convergir. / Functional Neuroimaging has evolved in the last few decades with the introduction of techniques such as Positron Emission Tomography or PET and Functional Magnetic Ressonance Image or fMRI [Belliveau et al., 1991]. These allow observing brain activity with a resolution of a few millimeters and, due to the nature of the signal, a time resolution of the order of 5 seconds [Kim et al., 1997]. M/EEG, on the other hand, have a millisecond time resolution, since the signal is produced by the transport of ions through cell membranes [Nunez and Srinivasan, 2006]. However their space resolution is much lower since these are typically ill posed problems with many more unknowns than data points. A high resolution M/EEG has of the order of O(200) data channels, which allow measuring the magnetic or electric field at O(200) positions around the head. For a resolution scale of order l there are O(L l )3 variables, where L = 15cm. In this work we aim at studying methods to increase the spatial resolution of EEG techniques, since functional mapping of the human brain is intimately related to the localization of the activity in space as well as in time [Friston, 2009] (often relative to the time of external stimuli). Any advance in the inverse problem of source localization for EEG can rather easily be extended to deal with MEG. Bayesian methods are the natural setting to deal with ill posed problems [Wipf and Nagarajan, 2009]. There are essentially two directions in which Bayesian algorithms can be improved, by building a better likelihood or a prior distribution. While we recognize that important advances can be done in the former direction we here concentrate in the latter. In this work we introduce a multiscale method to build an improved prior distribution. A similar idea has been studied within an easier context of fMRI [Amaral et al., 2004]. Several new problems appear in dealing with the vectorial character of EEG. The most important, is the construction of a set of renormalized lattices that approximate the cortex region where the source activity is located and the related problem of de ning the relevant variables in coarser scale representation of the cortex. Validation of a new algorithm is always an essential problem. We present results which suggest on simulated data, that our method might be a valid alternative to current algorithms, judged both by the rate of errors in source localization as well as by the time it takes to converge.

Simultaneous EEG-fMRI is a multi-modal neuroimaging technique where hemodynamic activity across the brain is measured at millimeter spatial resolution using functional magnetic resonance imaging (fMRI) while electrical activity at the scalp is measured at millisecond resolution using electroencephalography (EEG). EEG-fMRI is significantly more challenging to collect than either modality on its own, due to electromagnetic coupling and interference between the modalities. The rationale for collecting the two modalities together is that, given the complementary spatial and temporal resolutions of the individual modalities, EEG-fMRI has the potential as a non-invasive neuroimaging technique for recovering the latent source space of neural activity. Inferring this latent source space and fusing these modalities has been a main challenge for realizing the potential of simultaneous EEG-fMRI for human neuroscience. In this thesis we develop a principled and interpretable approach to address this inference problem. We build on the knowledge of the generative processes underlying image/signal formation of each of the modalities, traditionally viewed via linear mappings, and recast this into a framework which we refer to as “neural transcoding”. The idea of neural transcoding is to generate a signal of one neuroimaging modality from another by first decoding it into a latent source space and then encoding it into the other measurement space. We implement this transcoding via deep architectures based on convolutional neural networks. We first develop a basic transcoding architecture and test it on simulated EEG-fMRI data. Evaluation on simulated data enables us to assess the model’s ability to recover a latent source space given known ground truth. We then extend this architecture and add a cycle consistency loss to create a cycle-CNN transcoder and show that it outperforms, in terms of the fidelity of recovered source space, both the basic transcoder as well as traditional source estimation techniques, even when we provide those techniques detailed information about the image generation process. We then assess the performance of the cycle-CNN transcoder on real simultaneous EEG-fMRI datasets including an auditory oddball dataset and a three-choice visual categorization dataset. Without any prior knowledge of either the hemodynamic response function or leadfield matrix, the transcoder is able to exploit the temporal and spatial relationships between the modalities and latent source spaces to learn these mappings. We show, for real EEG-fMRI data, the modalities can be transcoded from one to another, and that the transcoded results for unseen test data, have substantial correlation with the ground truth. In addition, we analyze the source space of the transcoders and observe latent neural dynamics that could not be observed with either modality alone--e.g., millimeter by millisecond dynamics of cortical regions representing motor activation and somatosensory feedback for finger movement. Collectively, this thesis demonstrates how one can incorporate a principled understanding of the generative process underlying biomedical image/signal formation in a deep learning framework to build models that are interpretable and symmetrically combine both modalities. It also potentially enables a new type of low-cost computational neuroimaging -- i.e., generating an ``expensive" fMRI BOLD image from ``low cost" EEG data.

京都大学 / 新制・課程博士 / 博士(人間健康科学) / 甲第24143号 / 人健博第106号 / 新制||人健||7(附属図書館) / 京都大学大学院医学研究科人間健康科学系専攻 / (主査)教授 十一 元三, 教授 澤本 伸克, 教授 村井 俊哉 / 学位規則第4条第1項該当 / Doctor of Human Health Sciences / Kyoto University / DFAM

Introduction: The brain can change its structure and functionality as a result ofexternal factors. The working memory (WM) of the brain is where informationcan be held and manipulated during a short period of time, with the purpose ofachieving higher cognitive functions such as reasoning and learning. The WMimproves in capacity during the development from childhood into adulthood,and variation of improvement is possible as an effect of situational factors andstimuli.Goal: The main goal of this project was to examine the effects of a WMtraining program on power distribution, connectivity and synchronicity withinbrain networks, using an intra-individual analysis approach.Method: A series of magnetoencephalography (MEG) measurements wasacquired for four subjects while they were performing WM and control tasks,during a WM training program, along with an MRI image of the brain for eachof the participants. The data was preprocessed for noise and artifact removaland a source reconstruction was performed. Time-frequency representationsof the data were created and the frequencies were categories into alpha,beta and gamma bands. The power difference between the WM and controltask was calculated as a function of cognitive load of each frequency band,and its variation over load was calculated as a constructed metric called’area under power difference curve’ (AUPDC), and visualised using colourscale representation upon the brain MRI of each subject. Brain parcels thatsignificantly deviated from a random distribution of AUPDC values wereidentified using a Gaussian distribution fit.Results and discussion: All subjects showed a clear improvement inperformance accuracy of the tasks, but as the effect on the power distributionsvaried considerably for each subject and frequency band, other aspects besidepower need to be investigated in order to understand the mechanisms behindthe improvement. However, the overall results indicate that many significantAUPDC values seem to have decreased during the WM training, both forthe positive and negative significant AUPDC values, suggesting a strongerdecreasing trend in power difference over cognitive load and a weaker increasingtrend. This could suggest an improved brain activation efficiency as an effectof the WM training.

Alzheimer's disease (AD) is developing into the single greatest healthcare challenge in the coming decades. The development of early and effective treatments that can prevent the pathological damage responsible for AD-related dementia is of utmost priority for healthcare authorities. The role of the APOE-ε4 genotype, which has been shown to increase an individual's risk of developing AD, is of central interest to this goal. Understanding the mechanism by which possession of this gene modulates brain function, leading to a predisposition towards AD is an active area of research. Functional connectivity (FC) is an excellent candidate for linking APOE-related differences in brain function to sites of AD pathology. Magnetoencephalography (MEG) is a neuroimaging tool that can provide a unique insight into the electrophysiology underpinning resting-state networks (RSNs) - whose dysfunction is postulated to lead to a predisposition to AD. This thesis presents a range of methods for measuring functional connectivity in MEG data. We first develop a set of novel adaptations for preprocessing MEG data and performing source reconstruction using a beamformer (chapter 3). We then develop a range of analyses for measuring FC through correlations in the slow envelope oscillations of band-limited source-space MEG data (chapter 4). We investigate the optimum time scales for detecting FC. We then develop methods for extracting single networks (using seed-based correlation) and multiple networks (using ICA). We proceed to develop a group-statistical framework for detecting spatial differences in RSNs and present a preliminary finding for APOE-genotype-dependent differences in RSNs (chapter 5). We also develop a statistical framework for quantifying task-locked temporal differences in functional networks during task-positive experiments (chapter 6). Finally, we demonstrate a data-driven parcellation and network analysis pipeline that includes a novel correction for signal leakage between parcels. We use this framework to show evidence of stationary cross-frequency FC (chapter 7).

Explorations into the organisation of spontaneous activity within the brain have demonstrated the existence of networks of temporally correlated activity, consisting of brain areas that share similar cognitive or sensory functions. These so-called resting state networks (RSNs) emerge spontaneously during rest and disappear in response to overt stimuli or cognitive demands. In recent years, the study of RSNs has emerged as a valuable tool for probing brain function, both in the healthy brain and in disorders such as schizophrenia, Alzheimer’s disease and Parkinson’s disease. However, analyses of these networks have so far been limited, in part due to assumptions that the patterns of neuronal activity that underlie these networks remain constant over time. Moreover, the majority of RSN studies have used functional magnetic resonance imaging (fMRI), in which slow fluctuations in the level of oxygen in the blood are used as a proxy for the activity within a given brain region. In this thesis we develop the use of magnetoencephalography (MEG) to study resting state functional connectivity. Unlike fMRI, MEG provides a direct measure of neuronal activity and can provide novel insights into the temporal dynamics that underlie resting state activity. In particular, we focus on the application of non- stationary analysis methods, which are able to capture fast temporal changes in activity. We first develop a framework for preprocessing MEG data and measuring interactions within different RSNs (Chapter 3). We then extend this framework to assess temporal variability in resting state functional connectivity by applying time- varying measures of interactions and show that within-network functional connectivity is underpinned by non-stationary temporal dynamics (Chapter 4). Finally we develop a data driven approach based on a hidden Markov model for inferring short lived connectivity states from resting state and task data (Chapter 5). By applying this approach to data from multiple subjects we reveal transient states that capture short lived patterns of neuronal activity (Chapter 6).

L'activité cérébrale, reliée spécifiquement à la rétention d'information en mémoire à court-terme tactile, a été investiguée à l'aide de l'enregistrement des champs magnétiques produits par l'activité neuronale générée durant la période de rétention par une tâche de mémoire tactile. Une, deux, trois ou quatre positions, sur une possibilité de huit sur les phalangines et les phalangettes, de la main gauche ou droite, lors de blocs d'essai différents, ont été stimulées simultanément. Le patron de stimulation tactile devait être retenu pendant 1800 ms avant d'être comparé avec un patron test qui était, soit identique, soit différent par une seule position. Nos analyses se sont concentrées sur les régions du cerveau qui montraient une augmentation monotone du niveau d'activité soutenu durant la période de rétention pour un nombre croissant de positions à retenir dans le patron de stimulation. Ces régions ont plus de chance de participer à la rétention active de l'information à maintenir en mémoire à court-terme tactile. Le gyrus cingulaire (BA32), le gyrus frontal supérieur droit (BA 8), le precuneus gauche (BA 7, 19), le gyrus postcentral gauche (BA 7), le gyrus precentral droit (BA 6), le gyrus frontal supérieur gauche (BA 6) et le lobule pariétal inférieur droit (BA 40) semblent tous impliqués dans un réseau mnésique qui maintient les informations sensorielles tactiles dans un système de mémoire à court-terme spécialisé pour l'information tactile. / Brain activity specifically related to the maintenance of information held in tactile short-term memory was investigated, using recordings of magnetic fields from a whole-head magnetometer. This neuronal activity was measured during the retention interval of a tactile memory task. One, two, three, or four locations on distal and intermediate phalanges, out of eight positions, were simultaneously stimulated on the left or right hand in different blocks of trials. The tactile stimulation pattern was held in memory for 1800 ms before being compared with a test pattern that was either the same or different by one location. Our analyses focused on regions in the brain that showed a monotonic increase of the sustained activity levels during the retention interval with an increasing number of stimulated locations in the to-be-remembered pattern. These regions are the most likely to participate in the active retention of the information to be held in tactile sensory memory. The right cingular gyrus (BA 32), the right superior frontal gyrus (BA 8), the left precuneus (BA 7, 19), the left postcentral gyrus (BA 7), the right precentral gyrus (BA 6), the left superior frontal gyrus (BA 6) and the right inferior parietal lobule (BA 40) all appear to be involved in a memory system that maintains tactile sensory input in a short-term memory system specialized for tactile information.

Le temps peut être à la fois une dimension structurant l’organisation de la perception et un contenu de la conscience, ce qui en fait un objet d’étude crucial pour comprendre la cognition humaine. Le but de cette thèse est de déterminer dans quelle mesure l’ordonnancement temporel de l’information est lié à l’ordre perçu (explicite ou implicite) des évènements. Le premier axe d’étude s’intéressant au temps comme contenu conscient montre que l’ordre temporel est un biais psychologique indépendant de l’attention. Une étude complémentaire en magnétoencéphalographie a révélé que les participants ayant les plus larges biais (c.à.d. que leur perception de la simultanéité correspond à une large désynchronisation physique) montrent aussi les plus larges fluctuations de la puissance des oscillations alpha spontanées lorsque la perception de l’ordre est contrastée pour un même délai physique. Ces oscillations sont donc proposées comme moyen pour compenser un biais individuel. Le deuxième axe de recherche explore comment le temps façonne implicitement la perception d’une séquence visuelle, par l’étude de la postdiction, c.à.d. quand des informations plus tardives influencent la représentation d’informations antérieures. Dans l’illusion du Lapin, un flash intermédiaire est mal localisé à cause de la régularité temporelle de la séquence. Nous avons montré que percevoir l’illusion active des régions parieto-centrales après la fin de la séquence. Ces résultats suggèrent que des régions de haut niveau pourraient contribuer à la reconstruction postdictive d’une séquence visuelle, en accord avec l’hypothèse que l’illusion dépend d’une connaissance préalable sur la vitesse d’un stimulus. / Time in the brain can be considered both as a dimension structuring the organization of perception and as a conscious content: this confers to time a particular flavor that makes it a paramount object of study to understand the basis of human cognition. The present thesis aims at empirically determining to which extent the temporal ordering of information relates to the perceived order of events, both implicitly and explicitly. First, we investigated how time may become an explicit conscious content by showing that temporal order is a psychological bias, independent of attention. A follow-up magnetoencephalography study revealed that participants having a large order bias (i.e. their perceived synchrony corresponds to a large physical asynchrony) also showed the largest fluctuations in ongoing alpha power when perceived orders were contrasted for a given physical asynchrony. Alpha oscillations are herein argued to be a means to compensate for an individual internal bias. Second, we investigated how time implicitly shaped the perception of visual sequences by studying postdiction, i.e. when late inputs strikingly influence the representation of earlier information. In the Rabbit illusion, the intermediate flash of a visual sequence is spatially mislocalized due to the temporal regularity of the sequence. We showed that parieto-frontal regions were more activated following the presentation of the full sequence when the illusion was perceived. These results suggest that high-order regions may contribute to the postdictive reconstruction of a visual sequence, consistently with the hypothesis that the illusion is shaped by prior knowledge on a stimulus speed.

Visual working memory (WM) is the ability to hold information in mind for a short time before acting on it. The capacity of WM is strikingly limited. To make the most of this precious resource, humans exhibit a high degree of cognitive flexibility: We can prioritize information that is relevant to behavior, and inhibit unnecessary distractions. This thesis examines some behavioral and neural correlates of flexibility in WM. When information is of particular importance, anticipatory attention can be directed to where it will likely appear. Oscillations in visual cortex, in the 10-Hz range, play an important role in regulating excitability of such prioritized locations. Chapter 4 describes how even spontaneous fluctuations in 10-Hz synchronization (measured by electroencephalography, EEG) before encoding influence WM. Chapters 2 and 3 describe how attention can be directed retrospectively to items even if they are already stored in WM. Chapter 3 discusses how retrospective cues change neural synchronization similarly to anticipatory cues. Behavioral and neural measures additionally indicate that the boosting of an item through retrospective cues does not require prolonged deployment of attention: rather, it may be a transient process. The second half of this thesis additionally examines how items are represented in visual WM. Chapter 5 summarizes a study using pattern analysis of magnetoencephalographic (MEG) and EEG data to decode features of visual templates stored in WM. Decoding appears transiently around the time when potential target stimuli are expected, in line with a flexible reactivation mechanism. Chapter 6 further examines separate cortical networks involved in protecting vs. updating items in WM, and tests whether task relevance changes how well WM contents can be decoded. Finally, Chapter 7 summarizes the thesis and discusses how attentional flexibility can merge WM with a wider range of sources of behavioral control.