Functional connectivity analysis in the human brain using ultra-high field MRI

Abstract

Introduction: Functional magnetic resonance imaging (MRI) is a non-invasive method which uses a combination of a strong magnetic field and radio frequency pulses to image magnetic difference between oxygenated and deoxygenated blood in the human brain. This contrast differences can be used to identify areas in the brain when subjects performing an active task in the MRI scanner. It is also possible to measure spontaneous BOLD oscillation in absence of an external stimuli, a method called resting-state fMRI (rsfMRI). However, it is necessary to estimated and remove physiological noise, like head movements or heartbeat, as well as MRI scanner noise. Those estimated signals are called confounds. Therefore, an accurate preprocessing of the time signals is necessary. However, available preprocessing pipelines are not well established yet for rsfMRI data from ultra-high field MRI scanners. The goal of this study was to compare two slightly different rsfMRI preprocessing pipelines on the same dataset. Further, to investigate the influence of these differences on the robustness and functional connectivity of specific resting-state networks (RSN).

Methods: A rsfMRI dataset from ten healthy subjects was acquired on an ultra-high field seven Tesla MRI scanner and preprocessed with both pipelines, CPAC and fMRIprep. A group-wise independent component analysis (ICA) was performed to measure the functional and spatial connectivity between and within RSN. Additionally, we performed a detailed comparison of the confounds between the pipelines.

Results: We identified six different RSN. Subjects preprocessed with fMRIprep showed a strong temporal correlation within the visual, sensory motor as well as between the left and right memory function network. However, there were no significant spatial differences between the pipelines. Although head motion confounds were similar, confounds using brain masks to extract the signal differ.

Discussion: The stronger positive and negative correlation is in line with the literature although the study lack in statistical power. The major impact of the pipeline differences could be addressed to varying brain masks from the estimated confounds. This detailed comparison may help to further investigate the influence of different preprocessing steps to functional connectivity.

Popular Abstract

The active brain at rest

What are you doing when you are at rest? When you take a short break from your daily life without looking on your phone or thinking about something special. And what is your brain doing? The neurons in your brain are constantly communicating, even when you are doing nothing. It is possible to measure the neuronal activity indirectly over the higher energy consumption and blood flow using functional magnetic resonance imaging (fMRI). A MRI scanner uses a combination of a strong magnetic field (140.000 times stronger than the earth’s magnetic field) and radio frequency pulses to investigate the human brain at rest, a method which is called resting-state fMRI (rsfMRI).

During a rsfMRI examination the whole brain is parcellated into small voxels, which are 3-dimensional cubes of 1-3 mm3. During 5-7 minutes of rest, a time signal can be measured for each voxel. Voxels which are “functional connected” have a similar time signal are likely to be connected and processing the same information. Those voxels can be summarized together into a network, so called resting-state networks (RSN). Examples of such networks can be seen in the image below. The most important and robust network is the default mode network (DMN). It is only active when the brain is at rest and it is hypnotized to represent an introspective phase of sorting and processing past experience.

The measured time signal consists of the neuronal signal but also of physiological and artificial noise. This noise can come from the heartbeat, respiration, small head movements or even from the MRI scanner. To remove those unwanted noise from the signal, it is necessary to carefully preprocess the data. In this work two slightly different preprocessing programs, fMRIprep and CPAC, were compared to investigate the robustness and differences in the functional connectivity of six different brain networks.

The voxels with are involved in the different networks were the same for subjects preprocessed with fMRIprep and CPAC. However, the functional connectivity between and within the networks was higher in subjects preprocessed with fMRIprep. Those differences could be attributed to the fact that the programs used slightly different ways to estimate the head motion as well as the physiological noise e.g., by using a different mask of vessels in the brain. This comparison may help to better understand the influence and interaction of different preprocessing steps to the functional connectivity inside the human brain at rest.

Master’s Degree Project in Bioinformatics 60 credits 2021
Department of Biology, Lund University
Advisor: Peter Mannfolk and Olof Strandberg
Department of Clinical Science/Diagnostic Radiology, MR Physics group, Faculty of Medicine, Lund University

• author: Rumetshofer, Theodor
• supervisor: Peter Mannfolk; Olof Strandberg
• organization: Degree Projects in Bioinformatics
• course: BINP52 20201
• year: 2021
• type: H2 - Master's Degree (Two Years)
• subject: Biology and Life Sciences
• language: English
• id: 9059301
• date added to LUP: 2021-06-28 11:52:12
• date last changed: 2021-06-28 11:52:12