39 research outputs found
Simulating brain resting-state activity: what matters?
No full textIn the field of computational neuroscience, large-scale biophysical modelling is a bottom-up approach to study the interaction between brain structure and function. In this thesis, we propose two models of varying biophysical plausibility as a mechanistic explanations for spontaneous brain activity, as measured with magnetoencephalography (MEG).
Mathematically, these models take the form of large systems of non-linear coupled delay-differential equations, and we implement software to numerically solve such systems efficiently. After analysing the empirical data and extracting key features of interest (relating to the temporal dynamics of measured signals), we use Bayesian optimisation to fit our models with two different parameterisation of increasing complexity: first assuming a spatially uniform brain in which the pattern of connections between cortical regions is the only source of temporal structure in the simulations; and second allowing smooth variations of intrinsic parameters across the cortical surface. Our results outperform those published in the scientific literature to date.
We contribute an original derivation of a conductance-based model, and an in-depth analysis of the effects of intrinsic model parameters; software to build and simulate large models of delay-networks efficiently; a new approach to the exploration of high-dimensional spaces in the context of Bayesian optimisation (using space-partitioning); an original parameterisation allowing smooth spatial variations of intrinsic parameters across the cortical surface (using spherical harmonics); a novel analysis of structural brain data (from tractography), and several original methods to analyse MEG data (e.g. exploiting the Hilbert phase and extending Riemannian metrics).</p
Modèle stochastique de compensation du mouvement cardiaque lors d'interventions percutanées des artères coronaires
No full textUne méthode stochastique de compensation du mouvement 3D des artères coronaires à partir d’une séquence angiographique sous un seul angle de vue est présentée. La ligne centrale 3D des artères coronaires est segmentée sur un examen tomodensitométrique (MSCT) préopératoire, puis déformée de manière non-rigide à l’aide d’un modèle de déformation comportant peu de paramètres. Les images angiographiques sont également segmentées à l’aide d’un filtre vasculaire multi-échelle, seuillées, puis amincies de manière à obtenir des images binaires 2D de la ligne centrale des artères coronaires à chaque instant.
Un modèle génératif est ensuite introduit pour modéliser le processus qui résulte en l’observation des images angiographiques de la séquence comme un processus de Markov caché. Ce dernier est utilisé conjointement avec un filtre de particules pour contraindre l’évolution temporelle des paramètres de déformation au cours de la séquence. Ces contraintes sont basées sur des scores de similarité cosinus impliquant une transformation de distance sur les images binaires de la ligne centrale des artères coronaires. Le recalage 3D est exécuté de manière itérative et projective.
La validation est tout d’abord effectuée au travers d’un ensemble de simulations utilisant des lignes centrales 3D réelles des artères coronaires, puis finalement avec une ligne centrale 3D et la séquence angiographique associée
'The Face as a Battlefield' by Ygal Bursztyn
No full textThis text transposes, in the form of an article, the main themes tackled by the director Ygal Bursztyn in his book Face, Battlefield (Tel Aviv, Hakibbutz Hameuhad, 1990). Daniel Dayan thanks the author and the translator Sonia Hadida for their collaboration on this adaptation, reproduced with the kind permission of the review Hermes.</jats:p
Intensity-based hierarchical clustering in CT-scans: application to interactive segmentation in cardiology
No full text
Bayesian optimization of large-scale biophysical networks
No full textThe relationship between structure and function in the human brain is well established, but not yet well characterised. Large-scale biophysical models allow us to investigate this relationship, by leveraging structural information (e.g. derived from diffusion tractography) in order to couple dynamical models of local neuronal activity into networks of interacting regions distributed across the cortex. In practice however, these models are difficult to parametrise, and their simulation is often delicate and computationally expensive. This undermines the experimental aspect of scientific modelling, and stands in the way of comparing different parametrisations, network architectures, or models in general, with confidence. Here, we advocate the use of Bayesian optimisation for assessing the capabilities of biophysical network models, given a set of desired properties (e.g. band-specific functional connectivity); and in turn the use of this assessment as a principled basis for incremental modelling and model comparison. We adapt an optimisation method designed to cope with costly, high-dimensional, non-convex problems, and demonstrate its use and effectiveness. Using five parameters controlling key aspects of our model, we find that this method is able to converge to regions of high functional similarity with real MEG data, with very few samples given the number of parameters, without getting stuck in local extrema, and while building and exploiting a map of uncertainty defined smoothly across the parameter space. We compare the results obtained using different methods of structural connectivity estimation from diffusion tractography, and find that one method leads to better simulations
Short-term memory advantage for brief durations in human APOE ε4 carriers
No full textThe Apolipoprotein-E (APOE) ε4 gene allele, the highest known genetic risk factor for Alzheimer's disease, has paradoxically been well preserved in the human population. One possible explanation offered by evolutionary biology for survival of deleterious genes is antagonistic pleiotropy. This theory proposes that such genetic variants might confer an advantage, even earlier in life when humans are also reproductively fit. The results of some small-cohort studies have raised the possibility of such a pleiotropic effect for the ε4 allele in short-term memory (STM) but the findings have been inconsistent. Here, we tested STM performance in a large cohort of individuals (N = 1277); nine hundred and fifty-nine of which included carrier and non-carriers of the APOE ε4 gene, those at highest risk of developing Alzheimer's disease. We first confirm that this task is sensitive to subtle deterioration in memory performance across ageing. Importantly, individuals carrying the APOE ε4 gene actually exhibited a significant memory advantage across all ages, specifically for brief retention periods but crucially not for longer durations. Together, these findings present the strongest evidence to date for a gene having an antagonistic pleiotropy effect on human cognitive function across a wide age range, and hence provide an explanation for the survival of the APOE ε4 allele in the gene pool.</p
Connectome and FC from real MEG data and from simulations.
No full textFirst column shows resting-state MEG static FC (orthogonalized alpha band envelope correlations, top), and the anatomical network or connectome (bottom). The second and third columns show the FCs obtained from simulations with homogeneous and heterogeneous ensembles, respectively. The top row shows the static FCs. The bottom row shows the histograms of time-resolved FC recurrence (trFC: correlation of static FCs for 15 sec. moving window with 12 sec. overlap) for simulation (blue) and MEG data (pink). Brain lobes are color-coded around the connectivity matrices–blue, temporal; orange, occipital; red, parietal; and green, frontal.</p
Global and local metastability of heterogeneous ensembles as a function of trFC and sFC similarity between 300-second simulations and MEG for the best 2000 simulations.
No full textEach dot represents one simulation similar to Fig 3. (Left panel) Dots are colored by their global metastability. (Right panel) Dots are colored by their local metastability.</p
Evolution of local couplings optimization sorted by fitness to MEG sFC.
No full textEach row inside the upper panel depicts the local couplings of one ensemble. The columns correspond to iterations of the respective optimizer (aDE and PSO). The brain lobe associated with each ensemble is color coded on the left (same as Fig 2). The fitness (Pearson correlation between MEG and simulated sFC) of each iteration is at the bottom of the panels. trFC was evaluated at the parameters to the right of the black vertical line. (TIF)</p
