Analysis of connectivity between local multi-variate patterns of functional MRI data [Elektronische Ressource] / vorgelegt von Volker Fischer

Sujets

• Psychologie
• Physik

Informations

• Publié par: universitat_regensburg
• Publié le: 01 janvier 2011
• Nombre de lectures: 17
• Langue: English
• Poids de l'ouvrage: 11 Mo

Extrait

ANALYSIS OF CONNECTIVITY
BETWEEN LOCAL MULTI-VARIATE
PATTERNS OF FUNCTIONAL MRI
DATA
Dissertation
zur Erlangung des Doktorgrades
der Naturwissenschaften (Dr. rer. nat.)
der Fakult¨at fur¨ Physik
der Universit¨at Regensburg
vorgelegt von
Volker Fischer
aus Regensburg
Januar 2011Promotionsgesuch eingereicht am: 17.01.2011
Die Arbeit wurde durchgefuhrt¨ am Institut fur¨ Biophysik und physikalische Bio-
chemie unter Anleitung von Prof. Dr. Elmar W. Lang in Zusammenarbeit mit
dem Institut fur¨ Experimentelle Psychologie unter Betreuung von Prof. Dr. Mark
W. Greenlee.
Prufungsaussc¨ huss:
Vorsitzender: Prof. Dr. Andreas Sch¨afer
1. Gutachter: Prof. Dr. Elmar W. Lang
2.hter: Prof. Dr. Mark W. Greenlee
3. Gutachter: Prof. Dr. Josef ZweckAcknowledgments
This dissertation would not have been possible without the guidance and the
help of several individuals contributing their assistance in the preparation and
completion of this thesis. First and foremost, my utmost gratitude to my super-
visors Prof. Dr. Mark W. Greenlee and Prof. Dr. Elmar W. Lang. I will never
forget their continuous support and encouragement to make this thesis possible.
Especially I thank Prof. Dr. Mark W. Greenlee for offering me the opportunity to
work with functional magnetic resonance imaging, his guidance concerning neuro-
scientific issues, and the financial support with which he and the Federal Ministry
of Education and Research supported this project.
I owe my deepest gratitude to Dr. Ingo Keck and Dr. Anton Beer who, during
countless conversations, inspired many aspects of this thesis with their ideas and
knowledge.
My colleagues, Dipl.-Psych. Markus Raabe and Dr. Ference Acs, I would like
to thank for their patient guidance during my first steps with functional magnetic
resonance imaging and Markus Raabe especially for his help with data aquisition
and providing paradigm stimuli.
I would like to thank Mariella Strehl, Maressa McConkey, and Helmut Nebl for
reviewing this script and searching for the last missing comma.
I am also indebted to the numerous contributors to the Open Source pro-
gramming community for providing the countless toolboxes and systems I have
used to produce both my results and this thesis.
Last but not the least, I thank my family and friends for their moral support
throughout all my studies.
Volker Fischer
??Contents
1 Introduction 1
2 Methods 5
2.1 Functional magnetic resonance imaging . . . . . . . . . . . . . . . . 5
2.1.1 MRI physics . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 Functional MRI and the BOLD signal . . . . . . . . . . . . 8
2.1.3 Measurement protocol and preprocessing of fMRI data . . . 12
2.2 FMRI correlates of neuronal connectivity . . . . . . . . . . . . . . . 15
2.2.1 Neuronal connectivity . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Dynamic causal modeling . . . . . . . . . . . . . . . . . . . 17
2.3 Multi-variate pattern analysis and matrix decompositions . . . . . . 21
2.3.1 Multi-variate patterns . . . . . . . . . . . . . . . . . . . . . 21
2.3.2 Matrix decompositions of fMRI data . . . . . . . . . . . . . 24
2.3.3 The general linear model . . . . . . . . . . . . . . . . . . . . 24
2.3.4 Principal component analysis . . . . . . . . . . . . . . . . . 26
2.3.5 Independent component analysis . . . . . . . . . . . . . . . 27
3 Pattern Connectivity and its applications 30
3.1 Pattern connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1.2 Assumptions. . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.3 Pattern selection . . . . . . . . . . . . . . . . . . . . . . . . 34
3.1.4 Statisticalconnectivitybaselinecausedbycomponentselection 40
3.2 I. Paradigm: Validation of pattern connectivity . . . . . . . . . . . 44
3.2.1 Experimental setup and hypothesis . . . . . . . . . . . . . . 44
3.2.2 Region of interest definition . . . . . . . . . . . . . . . . . . 47
3.2.3 Component selection . . . . . . . . . . . . . . . . . . . . . . 50
3.2.4 Comparing connectivity of PCA and ICA components . . . . 54
3.2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.3 II. Paradigm: Potentials of pattern connectivity . . . . . . . . . . . 66
3.3.1 Experimental setup and hypothesis . . . . . . . . . . . . . . 66
3.3.2 Behavioral data . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.3.3 Synthetic data. . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.3.4 Region of interest definition . . . . . . . . . . . . . . . . . . 74
3.3.5 ICA parameters and component selection . . . . . . . . . . . 78
3.3.6 Connectivity between VA and FM COIs . . . . . . . . . . . 80
3.3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4 Discussion, summary, and future prospects 92
4.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.3 Future Prospects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5 Appendix 98
5.1 GLM contrasts for BMS and ORIENT paradigms . . . . . . . . . . 98List of gures 105
List of tables 107
Index 109
Bibliography 1111 INTRODUCTION 1
1 Introduction
“Brain: an apparatus with which I think I think.”
Ambrose Bierce (1842 - 1914), The Devil’s Dictionary
The human brain, with its approximately 100 billion neurons and 10:000 synap-
tic connections per neuron [119], is known to be the central controlling organ of
the human body. Studying the brain has led to numerous findings concerning
important concepts, such as free will (e.g. [63] [145] [165]) and consciousness (for
example [92] [61] among many more).
Many studies in the fields of medicine, biology, and psychology have demon-
strated that our brain can be divided into functionally segregated regions with
different specialization. In the visual domain, for example, signals originating
from the retina of the eye accumulate in the primary visual cortex located at the
posterior end of the brain. Information is then passed to higher areas concerned
with integration and evaluation. Depending on its content the original signal can
cause responses and actions governed e.g. by the motor cortex.
Information processing in different brain areas heavily relies on their intercon-
nections and the properties of these connections. Understanding how brain areas
are connected does not only lead to a deeper knowledge of how such complex
human behavior arise, but can also be used to diagnose neuronal disorders and
injuries [3] [149]. Investigating neuronal connectivity between brain areas has
therefore become a very important interdisciplinary field of neuroscience.
Aim of this thesis is to introduce, illustrate, and discuss a new way of investi-
gating connections between brain areas using non-invasive imaging techniques.
During the last century, many innovations in medicine, chemistry, and physics
allowed a deeper insight into organization and functionality of the brain. One
of the newest of these methods is functional magnetic resonance imaging (fMRI),
which, compared to other techniques, achieves a high spatial resolution of about
31mm . Fig. 1.1 shows two images of the human brain obtained with MR imaging
techniques.
Figure 1.1: Different views of the human brain showing a) an anatomical T -1
?weighted MR image, b) a functional T -weighted MR image (for details about T -12
?and T -weighted images please see section 2.1.1).2
Due to its physical nature, fMRI does not measure neuronal activity directly,
but only indirectly changes in the local oxygen concentration also called blood-2 1 INTRODUCTION
oxygen-level-dependent response or short the BOLD response, which is caused by
neuronal activation. In a typical fMRI experiment, neural correlates of a given
task are evaluated by comparing associated activations and/or deactivations with
those elicited by a control task.
FMRIhasbeenusedtostudyconnectivitybetweenbrainareas. Accordinglythe
temporal behavior of a brain area is represented by the first principal component
analysis (PCA) component of this region. This component is the time course
that explains most variance of the region’s behavior. At present, this approach
is the standard way of extracting temporal information from a brain region [46]
[49] [144] [172]. During the last few years, some approaches have been developed
to study connectivity between brain areas on the basis of these time series and
yielded important insights, for example, into pathological patterns of prodromal
psychosis [3], schizophrenia [149], and aging [134].
ItshighspatialresolutionmakesfMRIdataalsoaccessibletoadvancedstatistical
approaches to data analysis, such as support vector machines (SVM), which were
developedinotherresearchfields. Applicationofthesetechniquesledtoimportant
insights, such as, for example, about conscious and unconscious perception of
stimuli (e.g. [60] [61]). These methods take advantage of the fact that during
different tasks and stimuli, voxels (3D equivalent of a pixel) inside an area related
to the task or stimulus, form different spatial activation distributions (so-called
multi-variate or multi-voxel patterns (MVP)).
The key idea of this thesis is to re