Дисертації з теми "Brain aging, MRI, machine learning"

Segmenting brain MR scans could be highly benecial for diagnosing, treating and evaluating the progress of specic diseases. Up to this point, manual segmentation,performed by experts, is the conventional method in hospitals and clinical environments. Although manual segmentation is accurate, it is time consuming, expensive and might not be reliable. Many non-automatic and semi automatic methods have been proposed in the literature in order to segment MR brain images, but the levelof accuracy is not comparable with manual segmentation. The aim of this project is to implement and make a preliminary evaluation of a method based on machine learning technique for segmenting gray matter (GM),white matter (WM) and cerebrospinal uid (CSF) of brain MR scans using images available within the open MICCAI grand challenge (MRBrainS13).The proposed method employs supervised articial neural network based autocontext algorithm, exploiting intensity-based, spatial-based and shape model-basedlevel set segmentation results as features of the network. The obtained average results based on Dice similarity index were 97.73%, 95.37%, 82.76%, 88.47% and 84.78% for intracranial volume, brain (WM + GM), CSF, WM and GM respectively. This method achieved competitive results with considerably shorter required training time in MRBrainsS13 challenge.

Over the past years early detection and intervention in schizophrenia have become a major objective in psychiatry. Early intervention strategies are intended to identify and treat psychosis prior to fulfilling diagnostic criteria for the disorder. To this aim, reliable early diagnostic biomarkers are needed in order to identify a high-risk state for psychosis and also predict transition to frank psychosis in those high-risk individuals destined to develop the disorder. Recently, machine learning methods have been successfully applied in the diagnostic classification of schizophrenia and in predicting transition to psychosis at an individual level based on magnetic resonance imaging (MRI) data and also neurocognitive variables. This work investigates the application of machine learning methods for the early identification of schizophrenia in subjects at high risk for developing the disorder. The dataset used in this work involves data from the Edinburgh High Risk Study (EHRS), which examined individuals at a heightened risk for developing schizophrenia for familial reasons, and the FePsy (Fruherkennung von Psychosen) study that was conducted in Basel and involves subjects at a clinical high-risk state for psychosis. The overriding aim of this thesis was to use machine learning, and specifically Support Vector Machine (SVM), in order to identify predictors of transition to psychosis in high-risk individuals, using baseline structural MRI data. There are three aims pertaining to this main one. (i) Firstly, our aim was to examine the feasibility of distinguishing at baseline those individuals who later developed schizophrenia from those who did not, yet had psychotic symptoms using SVM and baseline data from the EHRS study. (ii) Secondly, we intended to examine if our classification approach could generalize to clinical high-risk cohorts, using neuroanatomical data from the FePsy study. (iii) In a more exploratory context, we have also examined the diagnostic performance of our classifier by pooling the two datasets together. With regards to the first aim, our findings suggest that the early prediction of schizophrenia is feasible using a MRI-based linear SVM classifier operating at the single-subject level. Additionally, we have shown that the combination of baseline neuroanatomical data with measures of neurocognitive functioning and schizotypal cognition can improve predictive performance. The application of our pattern classification approach to baseline structural MRI data from the FePsy study highly replicated our previous findings. Our classification method identified spatially distributed networks that discriminate at baseline between subjects that later developed schizophrenia and other related psychoses and those that did not. Finally, a preliminary classification analysis using pooled datasets from the EHRS and the FePsy study supports the existence of a neuroanatomical pattern that differentiates between groups of high-risk subjects that develop psychosis against those who do not across research sites and despite any between-sites differences. Taken together, our findings suggest that machine learning is capable of distinguishing between cohorts of high risk subjects that later convert to psychosis and those that do not based on patterns of structural abnormalities that are present before disease onset. Our findings have some clinical implications in that machine learning-based approaches could advise or complement clinical decision-making in early intervention strategies in schizophrenia and related psychoses. Future work will be, however, required to tackle issues of reproducibility of early diagnostic biomarkers across research sites, where different assessment criteria and imaging equipment and protocols are used. In addition, future projects may also examine the diagnostic and prognostic value of multimodal neuroimaging data, possibly combined with other clinical, neurocognitive, genetic information.

Background: Volumetric analysis of brain structures from structural Mag- netic Resonance (MR) images advances the understanding of the brain by providing means to study brain morphometric changes quantitatively along aging, development, and disease status. Due to the recent increased emphasis on large-scale multicenter brain MR study design, the demand for an automated brain MRI processing tool has increased as well. This dissertation describes an automatic segmentation framework for subcortical structures of brain MRI that is robust for a wide variety of MR data. Method: The proposed segmentation framework, BRAINSCut, is an inte- gration of robust data standardization techniques and machine-learning approaches. First, a robust multi-modal pre-processing tool for automated registration, bias cor- rection, and tissue classification, has been implemented for large-scale heterogeneous multi-site longitudinal MR data analysis. The segmentation framework was then constructed to achieve robustness for large-scale data via the following comparative experiments: 1) Find the best machine-learning algorithm among several available approaches in the field. 2) Find an efficient intensity normalization technique for the proposed region-specific localized normalization with a choice of robust statistics. 3) Find high quality features that best characterize the MR brain subcortical structures. Our tool is built upon 32 handpicked multi-modal muticenter MR images with man- ual traces of six subcortical structures (nucleus accumben, caudate nucleus, globus pallidum, putamen, thalamus, and hippocampus) from three experts. A fundamental task associated with brain MR image segmentation for re- search and clinical trials is the validation of segmentation accuracy. This dissertation evaluated the proposed segmentation framework in terms of validity and reliability. Three groups of data were employed for the various evaluation aspects: 1) traveling human phantom data for the multicenter reliability, 2) a set of repeated scans for the measurement stability across various disease statuses, and 3) a large-scale data from Huntington's disease (HD) study for software robustness as well as segmentation accuracy. Result: Segmentation accuracy of six subcortical structures was improved with 1) the bias-corrected inputs, 2) the two region-specific intensity normalization strategies and 3) the random forest machine-learning algorithm with the selected feature-enhanced image. The analysis of traveling human phantom data showed no center-specific bias in volume measurements from BRAINSCut. The repeated mea- sure reliability of the most of structures also displayed no specific association to disease progression except for caudate nucleus from the group of high risk for HD. The constructed segmentation framework was successfully applied on multicenter MR data from PREDICT-HD [133] study ( < 10% failure rate over 3000 scan sessions pro- cessed). Conclusion: Random-forest based segmentation method is effective and robust to large-scale multicenter data variation, especially with a proper choice of the intensity normalization techniques. Benefits of proper normalization approaches are more apparent compared to the custom set of feature-enhanced images for the ccuracy and robustness of the segmentation tool. BRAINSCut effectively produced subcortical volumetric measurements that are robust to center and disease status with validity confirmed by human experts and low failure rate from large-scale multicenter MR data. Sample size estimation, which is crutial for designing efficient clinical and research trials, is provided based on our experiments for six subcortical structures.

L'imagerie par résonance magnétique du processus de diffusion (IRMd) de l'eau dans le cerveau a connu un succès fulgurant au cours de la décennie passée pour cartographier les connexions cérébrales. C'est toujours aujourd'hui la seule technique d'investigation de la connectivité anatomique du cerveau humain in vivo. Mais depuis quelques années, il a été démontré que l'IRMd est également un outil unique de biopsie virtuelle in vivo en permettant de sonder la composition du parenchyme cérébral également in vivo. Toutefois, les modèles développés à l'heure actuelle (AxCaliber, ActiveAx, CHARMED) reposent uniquement sur la modélisation des membranes axonales à l'aide de géométries cylindriques, et restent trop simplistes pour rendre compte précisément de l'ultrastructure de la substance blanche et du processus de diffusion dans l’espace extra-axonal. Dans un premier temps, un modèle analytique plus réaliste de la substance blanche cérébrale tenant compte notamment de la dépendance temporelle du processus de diffusion dans le milieu extra-axonal a été développé. Un outil de décodage complexe permettant de résoudre le problème inverse visant à estimer les divers paramètres de la cytoarchitecture de la substance blanche à partir du signal IRMd a été mis en place en choisissant un schéma d'optimisation robuste pour l'estimation des paramètres. Dans un second temps, une approche Big Data a été conduite pour améliorer le décodage de la microstructure cérébrale. Un outil de création de tissus synthétiques réalistes de la matière blanche a été développé, permettant de générer très rapidement un grand nombre de voxels virtuels. Un outil de simulation ultra-rapide du processus de diffusion des particules d'eau dans ces voxels virtuels a ensuite été mis en place, permettant la génération de signaux IRMd synthétiques associés à chaque voxel du dictionnaire. Un dictionnaire de voxels virtuels contenant un grand nombre de configurations géométriques rencontrées dans la matière blanche cérébrale a ainsi été construit, faisant en particulier varier le degré de gonflement de la membrane axonale qui peut survenir comme conséquence de pathologies neurologiques telles que l’accident vasculaire cérébral. L'ensemble des signaux simulés associés aux configurations géométriques des voxels virtuels dont ils sont issus a ensuite été utilisé comme un jeu de données permettant l'entraînement d'un algorithme de machine learning pour décoder la microstructure de la matière blanche cérébrale à partir du signal IRMd et estimer le degré de gonflement axonal. Ce décodeur a montré des résultats de régression encourageants sur des données simulées inconnues, montrant le potentiel de l’approche computationnelle présentée pour cartographier la microstructure de tissus cérébraux sains et pathologiques in vivo. Les outils de simulation développés durant cette thèse permettront, en utilisant un algorithme de recalage difféomorphe de propagateurs de diffusion d’ensemble également développé dans le cadre de cette thèse, de construire un atlas probabiliste des paramètres microstructuraux des faisceaux de matière blanche
Diffusion Magnetic Resonance Imaging of water in the brain has proven very useful to establish a cartography of brain connections. It is the only in vivo modality to study anatomical connectivity. A few years ago, it has been shown that diffusion MRI is also a unique tool to perform virtual biopsy of cerebral tissues. However, most of current analytical models (AxCaliber, ActiveAx, CHARMED) employed for the estimation of white matter microstructure rely upon a basic modeling of white matter, with axons represented by simple cylinders and extra-axonal diffusion assumed to be Gaussian. First, a more physically plausible analytical model of the human brain white matter accounting for the time-dependence of the diffusion process in the extra-axonal space was developed for Oscillating Gradient Spin Echo (OGSE) sequence signals. A decoding tool enabling to solve the inverse problem of estimating the parameters of the white matter microstructure from the OGSE-weighted diffusion MRI signal was designed using a robust optimization scheme for parameter estimation. Second, a Big Data approach was designed to further improve the brain microstructure decoding. All the simulation tools necessary to construct computational models of brain tissues were developed in the frame of this thesis. An algorithm creating realistic white matter tissue numerical phantoms based on a spherical meshing of cell shapes was designed, enabling to generate a massive amount of virtual voxels in a computationally efficient way thanks to a GPU-based implementation. An ultra-fast simulation tool of the water molecules diffusion process in those virtual voxels was designed, enabling to generate synthetic diffusion MRI signal for each virtual voxel. A dictionary of virtual voxels containing a huge set of geometrical configurations present in white matter was built. This dictionary contained virtual voxels with varying degrees of axonal beading, a swelling of the axonal membrane which occurs after strokes and other pathologies. The set of synthetic signals and associated geometrical configurations of the corresponding voxels was used as a training data set for a machine learning algorithm designed to decode white matter microstructure from the diffusion MRI signal and estimate the degree of axonal beading. This decoder showed encouraging regression results on unknown simulated data, showing the potential of the presented approach to characterize the microstructure of healthy and injured brain tissues in vivo. The microstructure decoding tools developed during this thesis will in particular be used to characterize white matter tissue microstructural parameters (axonal density, mean axonal diameter, glial density, mean glial cells diameter, microvascular density ) in short and long bundles. The simulation tools developed in the frame of this thesis will enable the construction of a probabilistic atlas of the white matter bundles microstructural parameters, using a mean propagator based diffeomorphic registration tool also designed in the frame of this thesis to register each individual

Decision Support Systems (DSS) for assisted medical diagnosis are computer-based systems designed to assist clinicians with decision-making tasks by automatically determining diagnosis or improving diagnostic confidence. This could allow to perform early and differential diagnosis of neurological diseases, such as Alzheimer’s Disease (AD) and Parkinson’s Disease (PD), for which definite diagnosis still remains a crucial issue. Multivariate Machine Learning (ML) methods are gaining popularity within the neuroimaging community. Among these, supervised ML methods are able to automatically extract multiple information from image sets without requiring prior knowledge of where information may be coded. These methods have been proposed as a revolutionary approach for identifying sensitive biomarkers allowing for automatic classification of individual subjects. The aim of this thesis was to implement, optimize and validate a ML method able to perform automatic diagnosis of medical images by structural Magnetic Resonance Imaging data (sMRI). This method consists of 3 phases: 1) image preprocessing, mainly devoted to the co-registration of data from different patients to a common reference system; 2) feature extraction and selection, performed through Principal Components Analysis and Fisher’s Discriminant Ratio, with the aim of extracting and selecting the most discriminative features; 3) classification, performed by Support Vector Machine, with the aim of computing a predictive model for the diagnosis of new subjects. Moreover, I implemented a method for the generation of pattern distribution maps of brain structural differences, reflecting the importance of each voxel for classification. These maps could allow to identify new MR-related biomarkers for the diagnosis of neurological diseases. In order to test the feasibility of the implemented method, I applied it to the diagnosis of 3 pathologies: AD, PD and Eating Disorders (ED). Regarding PD, we acquired T1-weighted brain sMRI of 28 PD, 28 PSP (Progressive Supranuclear Palsy) and 28 healthy controls (CN). Classification performance in terms of accuracy (specificity/sensitivity) (%) was 94(91/97) for PD vs CN, 92(93/92) for PSP vs CN, 92(91/94) for PSP vs PD. Voxels influencing differential diagnosis of PD were localized in midbrain, pons, corpus callosum and thalamus, four critical regions involved in the pathophysiological mechanisms of PD. Regarding AD, I enrolled 509 subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, obtaining T1-weighted brain sMRI of 137 AD, 76 Mild Cognitive Impairment (MCI) patients who converted to AD (MCIc), 134 MCI who did not convert to AD (MCInc) within 18 months, and 162 CN. Classification performance (%) was 76±11 for AD vs CN, 72±12 for MCIc vs CN, 66±16 for MCIc vs MCInc. Voxels influencing the classification of AD vs CN were localized in the temporal pole, hippocampus, entorhinal cortex, amygdala, thalamus, putamen, caudate, insula, gyrus rectus, frontal and orbitofrontal cortices, anterior cingulate cortex, precuneus, posterior cerebellar lobule. Voxels influencing the classification of MCIc vs CN and MCIc vs MCInc were similar to those found for AD. Regarding ED, we acquired T1-weighted brain sMRI of 17 ED and 17 CN. The classifier allowed ED vs CN diagnosis with accuracy (specificity/sensitivity) of 85(73/93)%. Pattern distribution maps showed that voxels influencing ED vs CN discrimination were localized in the occipital cortex, posterior cerebellar lobule, precuneus, sensorimotor and premotor cortices, anterior cingulate and orbitofrontal cortices, all brain regions involved in the regulation of appetite and emotional processing. Results of this work were published in 7 ISI international papers, 3 indexed international papers, 1 international book chapter, 5 international conference proceedings and 1 national conference proceedings.

Magnetic resonance imaging (MRI) is a valuable tool for non-invasively investigating human brain anatomy and functions. The features extracted from MRI data can be used as biomarkers for neurodegenerative diseases, like Dementia. To deeply understand the mechanisms driving the brain changes it is crucial to extract reliable measures from the brain MRI scans and to increase the statistical power by harmonizing different datasets, such as in the ENIGMA studies. Here we applied the ENIGMA-SULCI pipeline to estimate the reliability of the sulcal descriptors extracted across the whole brain and to investigate their correlation with CDR (Clinical Dementia Rating) in the open access dataset OASIS. The OASIS dataset includes T1-weighted acquired from 416 right-handed subjects, for 227 of whose we know CDR. The measurement reliability has been estimated through technical replicates of a subgroup of patients MRI scans. The correlation of each sulcal shape descriptor with the degree of Dementia has been tested through linear regressions between each feature and the CDR series. We have trained linear (regression) and nonlinear (Neural Networks) Machine Learning models in order to classify the subjects in two classes (Dementia and healthy subjects). We got models able to correctly classify more than the 70% of the dataset, starting from sulcal measures.

This thesis deals with a classification of glioma grade in high and low aggressive tumours and overall survival prediction based on magnetic resonance imaging. Data used in this work is from BRATS challenge 2019 and each set contains information from 4 weighting sequences of MRI. Thesis is implemented in PYTHON programming language and Jupyter Notebooks environment. Software PyRadiomics is used for calculation of image features. Goal of this work is to determine best tumour region and weighting sequence for calculation of image features and consequently select set of features that are the best ones for classification of tumour grade and survival prediction. Part of thesis is dedicated to survival prediction using set of statistical tests, specifically Cox regression

The brain is a complex system of several interconnected components which can be categorized at different Spatio-temporal levels, evaluate the physical connections and the corresponding functionalities. To study brain connectivity at the macroscale, Magnetic Resonance Imaging (MRI) technique in all the different modalities has been exemplified to be an important tool. In particular, functional MRI (fMRI) enables to record the brain activity either at rest or in different conditions of cognitive task and assist in mapping the functional connectivity of the brain. The information of brain functional connectivity extracted from fMRI images can be defined using a graph representation, i.e. a mathematical object consisting of nodes, the brain regions, and edges, the link between regions. With this representation, novel insights have emerged about understanding brain connectivity and providing evidence that the brain networks are not randomly linked. Indeed, the brain network represents a small-world structure, with several different properties of segregation and integration that are accountable for specific functions and mental conditions. Moreover, network analysis enables to recognize and analyze patterns of brain functional connectivity characterizing a group of subjects. In recent decades, many developments have been made to understand the functioning of the human brain and many issues, related to the biological and the methodological perspective, are still need to be addressed. For example, sub-modular brain organization is still under debate, since it is necessary to understand how the brain is functionally organized. At the same time a comprehensive organization of functional connectivity is mostly unknown and also the dynamical reorganization of functional connectivity is appearing as a new frontier for analyzing brain dynamics. Moreover, the recognition of functional connectivity patterns in patients affected by mental disorders is still a challenging task, making plausible the development of new tools to solve them. Indeed, in this dissertation, we proposed novel methodological approaches to answer some of these biological and neuroscientific questions. We have investigated methods for analyzing and detecting heritability in twin's task-induced functional connectivity profiles. in this approach we are proposing a geodesic metric-based method for the estimation of similarity between functional connectivity, taking into account the manifold related properties of symmetric and positive definite matrices. Moreover, we also proposed a computational framework for classification and discrimination of brain connectivity graphs between healthy and pathological subjects affected by mental disorder, using geodesic metric-based clustering of brain graphs on manifold space. Within the same framework, we also propose an approach based on the dictionary learning method to encode the high dimensional connectivity data into a vectorial representation which is useful for classification and determining regions of brain graphs responsible for this segregation. We also propose an effective way to analyze the dynamical functional connectivity, building a similarity representation of fMRI dynamic functional connectivity states, exploiting modular properties of graph laplacians, geodesic clustering, and manifold learning.

La surface du cortex cérébral est très convoluée, avec un grand nombre de plis, les sillons corticaux. Ces plis sont extrêmement variables d'un individu à l'autre. Cette grande variabilité constitue un problème pour de nombreuses applications en neurosciences et en imagerie cérébrale. Un problème central est que les sillons cérébraux ne sont pas la bonne unité pour décrire les plis sur la surface corticale. En particulier, leur géométrie (forme) et leur topologie (branches, nombre de pièces) sont très variables. Les "Plis de passages" (PPs) peuvent expliquer une partie de cette variabilité. Le concept de PPs a été introduit pour la première fois par Gratiolet (1854) pour décrire les gyri transversaux qui interconnectent les deux côtés d'un sillon, sont fréquemment enfouis dans la profondeur de ces sillons, et sont parfois apparents sur la surface corticale. En tant que caractéristique intéressante du processus de plissement cortical, la connectivité structurelle sous-jacente des PP a également suscité beaucoup d'intérêt.Cependant, la difficulté d'identifier les PPs et le manque de méthodes systématiques pour les détecter automatiquement ont limité leur utilisation. Cette thèse vise à détecter et à caractériser les PPs sur la surface corticale tant du point de vue de la morphologie que de la connectivité. Elle s'articule autour de deux axes de recherche principaux : 1. Définition d'un processus de détection des PPs basé sur l'apprentissage automatique et utilisant leurs caractéristiques géométriques (ou morphologiques).2. Étudier les relations entre les PP et leur connectivité structurelle sous-jacente, et poursuivre le développement de modèles d'apprentissage automatique multimodaux. Dans la première partie, nous présentons une méthode de détection automatique des PP sur le cortex en fonction des caractéristiques morphologiques locales proposées dans (Bodin et al., 2021). Pour enregistrer les caractéristiques morphologiques locales de chaque sommet de la surface corticale, nous avons utilisé la méthode de profilage de la surface corticale (Li et al., 2010). Ensuite, le problème de reconnaissance tridimensionnelle des PP est converti en un problème de classification d'image bidimensionnelle avec un déséquilibre de classe où plus de points dans le STS sont des non-PP que des PP. Pour résoudre ce cas, nous proposons un modèle “Ensemble SVM” (EnsSVM) avec une stratégie de rééquilibrage. Les résultats expérimentaux et les analyses statistiques quantitatives montrent l'efficacité et la robustesse de notre méthode. Dans la deuxième partie, nous étudions la connectivité structurelle, en particulier les fibres U à courte portée, qui sous-tend la localisation des PPs, et proposons une nouvelle approche pour étudier la densité des terminaisons des fibres U sur la surface corticale. Nous émettons l'hypothèse que les PPs sont situés dans des régions de haute densité de terminaisons de fibres U croisées. En effet, nos analyses statistiques montrent une corrélation de robustesse entre les PPs et la densité de terminaisons des fibres U. De plus, nous discutons de l'impact de l'hétérogénéité de la connectivité dans le STS sur les résultats de l'apprentissage automatique. Enfin, nous investiguons l'utilisation de cartes de myéline comme un complément à la connectivité structurelle
The surface of the cerebral cortex is very convoluted, with a large number of folds, the cortical sulci. Moreover, these folds are extremely variable from one individual to another. This great variability is a problem for many applications in neuroscience and brain imaging. One central problem is that cerebral sulci are not the good unit to describe folding over the cortical surface. In particular, their geometry (shape) and topology (branches, number of pieces) are very variable. “Plis de passages” (PPs) or “annectant gyri” can explain part of the variability. The concept of PPs was first introduced by Gratiolet (1854) to describe transverse gyri that interconnect both sides of a sulcus, are frequently buried in the depth of these sulci, and are sometimes apparent on the cortical surface. As an interesting feature of the cortical folding process, the underlying structural connectivity of PPs also generated a lot of interest. However, the difficulty of identifying PPs and the lack of systematic methods to automatically detecting them limited their use. This thesis aims to detect and characterise the PPs on the cortical surface from both morphology and connectivity aspects. It was structured around two main research axes: 1. Definition of a machine learning-based PPs detection process using their geometrical (or morphological) characteristics. 2. Investigate the relationships between PPs and their un- derlying structural connectivity, and further development of multi-modal machine learning models. In the first part, we present a method to detect the PPs on the cortex automatically according to the local morphological characteristics proposed in (Bodin et al., 2021), To record the local morphological patterns for each vertex on the cortical surface, we used the cortical surface profiling method (Li et al., 2010). After that, the three-dimensional PP recognition problem is converted to a two-dimensional image classification problem of class-imbalance where more points in the STS are non-PPs than PPs. To solve this case, we propose an ensemble SVM model (EnsSVM) with a rebalancing strategy. Experimental results and quantitative statistics analyses show the effectiveness and robustness of our method. In the second part, we study the structural connectivity, particularly short-range U-fibers, underlying the location of PPs, and propose a new approach to study the density of U-fiber terminations on the cortical surface. We hypothesize that the PPs are located in regions of high density of intercrossing U-fibers termination. Indeed, our statistical analyses show a robustness correlation between PPs and U-fibers termination density. Moreover, we discuss the impact of connectivity heterogeneity in the STS on the machine learning results, and the myelin map is then used as a supplement to the structural connectivity

L’hallucination est une expérience subjective vécue en pleine conscience consistant en une perception impossible à distinguer d’une perception réelle, mais survenant en l’absence de tout stimulus en provenance de l’environnement externe. Les symptômes hallucinatoires, qui peuvent concerner toutes les modalités sensorielles, sont retrouvés dans divers troubles neurologiques et psychiatriques mais également chez certains sujets indemnes de toute pathologie. Dans le champ de la psychiatrie, la pathologie la plus fréquemment associée aux hallucinations reste la schizophrénie et la modalité auditive est la plus représentée, puisque 60 à 80% des patients souffrant de ce trouble sont concernés. Le retentissement fonctionnel des hallucinations auditives peut être important, altérant significativement la qualité de vie des patients.Dans ce contexte, la prise en charge de ce type de symptômes s’avère un enjeu considérable pour les personnes souffrant de schizophrénie. Pourtant, les moyens thérapeutiques actuellement disponibles (traitements médicamenteux antipsychotiques notamment) ne permettent pas toujours une rémission complète de la symptomatologie hallucinatoire et l’on considère que 25 à 30% des hallucinations auditives sont « pharmaco-résistantes ». C’est à partir de ce constat que, ces dernières années, ont émergé, pour le traitement des hallucinations auditives, des techniques de neuromodulation comme la stimulation magnétique transcrânienne répétée ou la stimulation électrique transcrânienne par courant continu. Toutefois, les résultats de ces nouvelles thérapies sur les hallucinations auditives résistantes restent modérés et le développement de stratégies alternatives demeure un enjeu de recherche majeur.Actuellement, les travaux en imagerie fonctionnelle permettent d'affiner les modèles physiopathologiques des hallucinations auditives, mais leur intérêt pourrait aller au-delà de la recherche fondamentale, avec possiblement des applications cliniques telles que l'assistance thérapeutique. Ce travail de thèse s’inscrit précisément dans le développement de l’imagerie cérébrale de « capture » des hallucinations auditives, c’est-à-dire l’identification des patterns d’activation fonctionnels associés à la survenue des hallucinations auditives.La première partie de ce travail est consacrée à la détection automatisée des hallucinations auditives en IRM fonctionnelle. L’identification des périodes hallucinatoires survenues au cours d’une session d’IRM fonctionnelle est actuellement possible par une méthode de capture semi-automatisée validée. Celle-ci permet une labellisation des données acquises au cours d’une session de repos en périodes « hallucinatoires » et « non-hallucinatoires ». Toutefois, le caractère long et fastidieux de cette méthode limite largement son emploi. Nous avons donc souhaité montrer comment les stratégies d’apprentissage machine (support vector machine ou SVM, notamment) permettent l’automatisation de cette technique par le développement de classificateurs performants, généralisables et associés à un faible coût de calcul (indispensable en vue d’une utilisation en temps réel). Nous proposons également le développement d’algorithmes de reconnaissance de la période « pré-hallucinatoire », en mettant en évidence que ce type de classificateur présente aussi des performances largement significatives. Enfin, nous avons pu montrer que l’utilisation de stratégies d’apprentissage-machine alternatives au SVM (e.g, le TV-Elastic-net), obtient des performances significativement supérieures au SVM [...]
Hallucination is a transient subjective experience perceived as real, but occurring in the absence of an appropriate stimulation coming from the external environment. Hallucinatory events, which can occur across every sensory modality, are observed in various neurological and psychiatric disorders but also among “non-clinical” populations. The most frequent disorder associated with hallucinations in the field of psychiatry is schizophrenia. Auditory-verbal experiences are particularly frequent, with a lifetime-prevalence of 60 to 80% in patients suffering from schizophrenia. Hallucinations may cause long-term disability and poorer quality of life.In this context, the management of auditory-verbal hallucinations in patients with schizophrenia constitutes a major challenge. However, despite the increasing sophistication of biological and psychosocial research methods in the field, no significant therapeutic breakthrough has occurred in the last decade and a consensus exists that a significant proportion of patients with schizophrenia (i.e., around 25 %), exhibit drug-resistant auditory-verbal hallucinations. Non-pharmacological treatments, such as repetitive transcranial magnetic stimulation (rTMS) or transcranial direct current stimulation (tDCS) have been proposed as an option for addressing the unmet medical needs described above. However, these neuromodulation techniques show a moderate effect in alleviating drug-resistant auditory-verbal hallucinations and the development of innovative therapeutic strategies remains a major challenge.In recent years, the number of brain imaging studies in the field of auditory-verbal hallucinations has grown substantially, leading to a better pathophysiological understanding of this subjective phenomenon. Recent progress in deciphering the neural underpinnings of AVHs has strengthened transdiagnostic neurocognitive models that characterize auditory-verbal hallucinations, but more specifically these findings built the bases for new therapeutic strategies. In this regards the development of auditory hallucinations “capture" brain-imaging studies (i.e. the identification of functional patterns associated with the occurrence of auditory hallucinations), was the main topic of this thesis.The first part of this work is devoted to the automatized detection of auditory-verbal hallucinations using functional MRI (fMRI). The identification of hallucinatory periods occurring during a fMRI session is now possible using a semi-automatized procedure based on an independent component analysis applied to resting fMRI data combined with a post-fMRI interview (i.e. the patient is asked to report auditory-verbal hallucinations immediately after acquisition). This “two-steps method” allows for the identification of hallucination periods (ON) and non-hallucination ones (OFF). However, the time-consuming nature of this a posteriori labelling procedure considerably limits its use. In these regards, we show how machine-learning, especially support vector machine (SVM), allows the automation of hallucinations capture. We present new results of accurate and generalizable classifiers which could be used in real-time because of their low computational-cost. We also highlight that algorithms able to identify the "pre-hallucinatory" period exhibit significant performances. Finally, we propose the use of an alternative learning-machine strategy, based on TV-Elastic-net, which achieves slightly better performances and more interpretable discriminative maps than SVM [...]

This thesis is focused on developing novel and fully automated methods for the detection of new multiple sclerosis (MS) lesions in longitudinal brain magnetic resonance imaging (MRI). First, we proposed a fully automated logistic regression-based framework for the detection and segmentation of new T2-w lesions. The framework was based on intensity subtraction and deformation field (DF). Second, we proposed a fully convolutional neural network (FCNN) approach to detect new T2-w lesions in longitudinal brain MR images. The model was trained end-to-end and simultaneously learned both the DFs and the new T2-w lesions. Finally, we proposed a deep learning-based approach for MS lesion synthesis to improve the lesion detection and segmentation performance in both cross-sectional and longitudinal analysis
Esta tesis se centra en el desarrollo de métodos novedosos y totalmente automatizados para la detección de nuevas lesiones de esclerosis múltiple en la resonancia magnética longitudinal del cerebro. Primero, propusimos un marco totalmente automatizado basado en la regresión logística para la detección y segmentación de nuevas lesiones T2-w. El marco se basaba en la sustracción de intensidad y el campo de deformación (DF). En segundo lugar, propusimos un enfoque de red neuronal totalmente convolucional para detectar nuevas lesiones T2-w en imágenes de resonancia magnética del cerebro longitudinal. El modelo se entrenó de extremo a extremo y aprendió simultáneamente tanto los DF como las nuevas lesiones T2-w. Por último, propusimos un enfoque basado en el aprendizaje profundo para la síntesis de las lesiones de la EM, a fin de mejorar el rendimiento de la detección y la segmentación de las lesiones tanto en el análisis transversal como en el longitudinal

碩士
國立陽明大學
生醫光電研究所
107
Background Computational Analysis of MRI has been developed for a long time. With the purpose of automation, rapidity, precision and low inter-subject variability, it offers feature quantification for scientific research, and clinically assists to diagnosis and medical treatment. The algorithms also include machine learning (ML) and deep learning, and they can apply on image segmentation and classification. Migraine is a primary headache that the etiology is still unclear, and its pathology is diverse also complex. Therefore, in this study, we aimed to analyze clinical migraine data, using different algorithms to make an integrated investigation to complete three missions: 1. Applying computational analysis to T1W image and rsfMRI, and compare binary sub-groups in four groups: migraine without depression (MwoD) versus with depression (MwD), migraine without aura (MwoA) versus with aura (MwA), control versus migraine with severe headache above 5 days per month, and control versus migraine occurred over 10 years 2. To perform ML and deep learning binary classification by using structural and functional features and the knowledge based on previous studies. Materials and methods In this study, we collected 46 normal control MRI and 251 migraine MRI with clinical data, and MRI sequences contain structural T1W image and rsfMRI. T1W analysis were using both VBM and SBM method. VBM calculates the spatial standardized GM distribution, and SBM calculates the numeric features including volume, thickness and surface area in each GM parcellation, and volume of WM and ventricles. In rsfMRI, we calculated fALFF map, ReHo map and functional connectivity in each subject. We use permutation test for functional connectivity statistics (correlation coefficient and dynamic variance), and two sample T-test for others. Finally, we apply ML by using numeric or volumetric feature to perform binary classification between two sub-groups. Results We found from structural MRI that migraine occurring above 5 years would decrease GM volume in right parietal and frontal lobe, and migraine with headache above 10 days per month would have similar pattern in frontal and temporal lobe. Aura would not affect GM volume, while depression would decrease GM volume in several regions. In fALFF and ReHo results, headache, aura and depression had distinct activation and deactivation regions. In the group of control versus migraine occurring above 5 years, and control versus migraine with headache above 10 days per month, the differences appeared in many places and both had similar patterns in occipital lobe. Among these 4 groups, the group MwoA versus MwA was found with the strongest and highest amount of difference in functional connectivity correlation coefficient and variance. In depression group, the differences were mainly in bilateral temporal lobe, and in two headache groups, the differences were more fragmentary. Performances for classifying migraine and migraine subgroup were not ideal. The best result was classifying aura in migraine using ML and functional connectivity features, and the testing accuracy is 73.3%. Conclusion The depression, aura and headache symptoms would affect in brain microstructure and functions respectively. Therefore, the structural and functional MRI may find various individual differences, and the individual differences affects directly the ML performance.

碩士
國立臺灣科技大學
電機工程系
105
In this study, we reproduced the investigation of Tustison, which evaluated cortical thickness calculation algorithms by machine-learning approaches. First, we used FreeSurfer to measure cortical thickness from structural MR TI images. We applied machine-learning algorithms to a large-scale database to get effective MR biomarkers of cortical thickness. Subsequently, we used linear regression algorithms to predict genders and ages from the T1 data sets, and then analyze the accuracy of the prediction. We compared the performances of these three neuroanatomical parcellation schemes by using the area under the curve of receiver operating characteristic curve of gender and the root-mean-square error of age. We then applied the established machine learning procedures to the task of discriminating ADHD patients from normal subjects. In addition, we performed feature selection by using the weights produced during recursive feature elimination with cross validation to potentially provide localization information for brain regions related to ADHD. In summary, the results obtained using the Desikan–Killiany parcellation scheme generally outperformed the other schemes in all tasks, predicting the gender and age of each participant and discriminating ADHD types.

Volumetric segmentation of brain sub-cortical structures within the basal ganglia and thalamus from Magnetic Resonance Image (MRI) is necessary for non-invasive diagnosis and neurosurgery planning. This is a challenging problem due in part to limited boundary information between structures, similar intensity profiles across the different structures, and low contrast data. With recent advances in ultrahigh-field MR technology, direct identification and clear visualization of such brain sub-cortical structures are facilitated. This dissertation first presents a semi-automatic segmentation system exploiting the visual benefits of ultrahigh-field MRI. The proposed approach utilizes the complementary edge information in the multiple structural MRI modalities. It combines optimally selected two modalities from susceptibility-weighted, T2-weighted, and diffusion MRI, and introduces a tailored new edge indicator function. In addition to this, prior shape and configuration knowledge of the sub-cortical structures are employed in order to guide the evolution of geometric active surfaces. Neighboring structures are segmented iteratively, constraining over-segmentation at their borders with a non-overlapping penalty. Experiments with data acquired on a 7 Tesla (T) MRI scanner demonstrate the feasibility and power of the approach for the segmentation of basal ganglia components critical for neurosurgery applications such as Deep Brain Stimulation (DBS) surgery.

DBS surgery on brain sub-cortical regions within the Basal ganglia and thalamus is an effective treatment to alleviate symptoms of neuro-degenerative diseases. Particularly, the DBS of subthalamic nucleus (STN) has shown important clinical efficacy for Parkinson’s disease (PD). While accurate localization of the STN and its substructures is critical for precise DBS electrode placement, direct visualization of the STN in current standard clinical MR imaging (e.g., 1.5-3T) is still elusive. Therefore, to locate the target, DBS surgeons today often rely on consensus coordinates, lengthy and risky micro-electrode recording (MER), and patient’s behavioral feedback. Recently, ultrahigh-field MR imaging allows direct visualization of brain sub-cortical structures. However, such high fields are not clinically available in practice. This dissertation also introduces a non-invasive automatic localization method of the STN which is one of the critical targets for DBS surgery in a standard clinical scenario (1.5T MRI). The spatial dependency between the STN and potential predictor structures from 7T MR training data is first learned using the regression models in a bagging way. Then, given automatically detected such predictors on the clinical patient data, the complete region of the STN is predicted as a probability map using learned high quality information from 7T. Furthermore, a robust framework is proposed to properly weight different training subsets, estimating their influence in the prediction accuracy. The STN prediction on the clinical 1.5T MR datasets from 15 PD patients is performed within the proposed approach. Experimental results demonstrate that the developed framework enables accurate prediction of the STN, closely matching the 7T ground truth.

Dissertation