Kyeongwon Park. (2025). Resting state of human brain measured by fMRI experiment is governed more dominantly by essential mode rather than default mode network. doi: 10.22677/THESIS.200000837763
인간의 뇌는 휴식 상태인 Resting-state에서도 특유의 활동 패턴을 보이는 것이 알려져 있다. Resting-state는 뇌의 reference 상태로 간주되기에 Resting-state에 대한 이해는 큰 중요성을 갖는다. 연구자들은 Task-state에서는 활동이 적은 반면, Resting-state에서는 활발한 활동을 보이는 뇌 영역들의 모임인 default mode network (DMN)을 발견하였고, DMN은 Resting-state를 특정하는 공간적 뇌 활동 패턴으로 알려져 Resting-state fMRI 분야에서 집중적인 연구의 대상이 되어 왔다. 본 연구에서는 과연 DMN이 Resting-state를 특정하는 유일한 공간적 패턴인지 탐구하기 위해, 166명의 Resting-state fMRI 데이터를 특이값 분해를 통해 독립적인 공간적 기저들과, 그 공간적 기저들의 변화를 설명하는 시간적 기저들로 분해한 뒤, 공간적 기저들의 clustering을 통해 166명에게서 공통적으로 발견되는 공간적 뇌 활동 패턴들을 탐지하였다. 그 결과, Resting-state BOLD signal에 가장 큰 영향력을 갖는 공간적 패턴은 우리가 essential mode (EM)라고 명명한, 뇌 전반을 어우르는 공간적 기저였고, 종래에 알려진 DMN은 두 번째로 영향력이 큰 공간적 기저로 밝혀졌다. 따라서 앞으로 Resting-state는 DMN 뿐만 아니라, EM 패턴의 시각에서도 분석되어야 함을 제안한다.|Even in the Resting-state, it is known that the human brain exhibits characteristic activation pattern. For the Resting-state is regarded as the reference state of the human brain, an understanding of the Resting-state human brain has significance in brain and cognitive science research. Researchers have identified a network of regions in the human brain which deactivates in Task-states, but activates in the Resting-state, named it default mode network (DMN). Subsequently, DMN has been known as the brain activation pattern that characterizes the Resting-state of the human brain, and studied extensively by researchers in fMRI field. In this study I wanted to see if there are any additional brain activation patterns which are also crucial for understanding the Resting- state. I have decomposed the Resting-state fMRI BOLD signal data of 166 people into spatial patterns and their corresponding temporal patterns through singular value decomposition, then conducted clustering on spatial patterns to identify common spatial activation pattern shared among 166 people. As a result, I have identified essential mode (EM) which has the most dominant influence in the Resting-state. DMN was identified as the second most dominant spatial activation pattern, following EM. It is therefore recommended that when studying the Resting-state of human brain, researchers should investigate not only DMN, but also EM.
Table Of Contents
Ⅰ. Introduction 1 1.1) Origin of magnetic resonance imaging 1 1.1.1) Nuclear magnetic resonance 1 1.1.2) Magnetic resonance imaging 2 1.2) functional MRI 3 1.2.1) Blood oxygen level dependent signal 3 1.2.2) Task fMRI 4 1.3) Methods to analyze fMRI data 6 1.3.1) fMRI data and preprocessing 6 1.3.2) Functional connectivity analysis 9 1.3.3) Decomposition based analysis 10 1.4) Resting-state fMRI 11 1.4.1) Task-negative region 11 1.4.2) Default mode network 11 1.5) Motivation and overview of study 12 1.5.1) Is DMN the most representative component characterizing the resting-state?12 1.5.2) Necessity of holistic approach 12 1.5.3) Overview of study 13
ⅠI. Theoretical Background 14 2.1) Eigenvalue and eigenvector 14 2.1.1) Eigen-decomposition 14 2.1.2) Eigen-basis representation of a matrix 15 2.2) Singular value decomposition 16 2.2.1) Singular value decomposition 16 2.2.2) Applications of singular value decomposition 17 2.3) Spectral graph theory 18 2.3.1) Matrix representation of a graph 18 2.3.2) Laplacian clustering 19
ⅠII. Methods 21 3.1) Materials 21 3.1.1) HCP dataset 21 3.1.2) HCP fMRI image acquisition protocol 21 3.2) Methods 23 3.2.1) Preprocessing 23 3.2.2) Singular value decomposition of fMRI BOLD signal matrix 25 3.2.3) Laplacian clustering with distillation 26 3.2.4) Finding the optimal threshold for Laplacian clustering 28 3.2.5) Fractional distillation of U vector clusters 30 3.2.6) Visualization of U vector clusters 31 3.2.7) Frequency analysis of V-vectors and its quadratic differential 32
ⅠV. Results 34 4.1) Singular value decomposition 34 4.1.1) Singular value decomposition result 36 4.2) Laplacian clustering and fractional distillation for resting-state fMRI 37 4.2.1) Laplacian clustering result overview 37 4.2.2) EM: Cluster of the most dominant U modes 39 4.2.3) DM: Cluster of the second most dominant U modes 41 4.2.4) Leftover groups 42 4.2.5) Time-robustness of our analysis 44 4.3) Laplacian clustering and fractional distillation for motor-task fMRI 46 4.3.1) Applying Laplacian clustering on motor-task fMRI 46 4.3.2) Findings from motor-task fMRI analysis 48 4.4) Frequency analysis 50 4.4.1) Quadratic differential of V-modes 51 4.4.2) Similarities between two leftover groups, shown in the power spectrum of quadratic differentials of V-modes 51 4.4.3) Time-evolution of EM and DM exhibit higher power under 0.2 Hz frequency region 52 4.4.4) Other characteristics of quadratic differential of V-modes 53
V. Discussion 56 5.1) Preprocessing 56 5.1.1) Global signal regression 56 5.2) EM is the most dominant mode of resting-state, and DM is the second most dominant mode 59 5.2.1) Discovery of common and dominant modes: EM and DM 59 5.2.2) EM: Characteristics 59 5.2.3) Why EM was more dominant than DM 60 5.2.4) Shortcomings of the motor-task result 60 5.3) The low-frequency oscillation in both EM and DM are distinct from 1/f noise · 61 5.3.1) Shadowing problem of 1/f noise in the low-frequency band 61 5.3.2) Quadratic differential reveals the genuine signal from the shadowed frequency band 61 5.3.3) EM and DM shows genuine oscillatory behavior in low frequency band 61 5.4) Summary and conclusion 63 5.4.1) The resting-state brain is a composition of EM and DM, modulating in time according to their genuine oscillation 63 5.4.2) EM shows the brain regions known to be responsible for task execution in the motor-task fMRI 63 5.4.3) EM and DM show the genuine time oscillatory behavior in the low-frequency band · 63 5.4.4) Advantages of Laplacian clustering with distillation method 64 5.4.5) Future work 64
