EECE 575 PROJECT REPORT 2005 Ashish Uthama



fMRI Data Characterization for the Study of Cortical Reorganization in Acute Optic Nerve Inflammation(Dec 2005)

Ashish Uthama, Student Number 6222054

Abstract—A fMRI study of patients recovering from optic neuritis could give an insight to the process of compensatory reorganization in the human central nervous system. Analyzing fMRI data requires complex image processing and use of statistical inference methods. Pilot data for a healthy person was collected using an fMRI experiment geared to stimulate the visual system. Activation regions were identified using the statistical parameter mapping using the SPM2 software toolbox.

Index Terms—fMRI, Optic Neuritis, SPM2

I.INTRODUCTION

ULTIPLE Sclerosis (MS) is an autoimmune disease that affects the Central Nervous System (CNS) of young adults. This disease is characterized by the failure of the immune system to recognize normal tissues, resulting in damage. The elements of the immune system target and destroy the myelin sheath that covers the axon of neurons. This demyelination of nerves results in varying degrees of physical disability, generally degenerating over time. It is observed that there is no direct correlation between demyelination and clinical disability. One explanation put forth is that the brain is able to adapt itself to limited degeneration by cortical re-organization [1].

Functionally related parts of the brain, normally not actively involved in a particular task might reorganize themselves to compensate for damaged tissue elsewhere. There exist methods (for example, using Magnetic Resonance Imaging, MRI) to detect the presence of MS, the complete pathology of this disease is still largely unknown. Current studies are aimed at gaining more knowledge of the disease progression and examining the response of the CNS. It is hoped that this knowledge will help scientists devise ways to slow down the degeneration in physical ability.

Optic Neuritis is a condition arising due to acute inflammation of the optic nerve and is often an initial presentation of MS. Patients with ON experience a sharp decline in visual acuity at the onset of the disease. Most patients recover their vision to near normal levels within one to two weeks of the attack. However, studies have shown that patients who recover from ON activate larger areas of the visual cortex than normal people [1]. This is hypothesised to be an example of the compensatory reorganization at work within the CNS. Optic neuritis can infect either a single eye or both; the focus in this study is on mono ocular ON.

The visual system of the human brain is extensively studied and there exist methods to quantify the performance of the optic nerves (Visually evoked potentials, VEP). Hence, a study of optic neuritis could be easily conducted with sufficient data available. An extensive study of the progress of ON in patients might give researchers and neurologists a better understanding of the adaptation mechanism of the CNS. This, in turn will be invaluable in evaluating treatment procedures for CNS diseases. Since in most cases physical disability is not a direct indication and there are few other means of quantifying the response of the CNS.

To identify areas of the brain involved in the processing of a particular stimulus, a MR technique called functional MRI (fMRI) has been invented. fMRI techniques can be used on patients with ON to identify areas of the brain involved in vision and quantify the difference from normal people. Tracking the changes in vision activated brain regions over the course of the disease will give a clear picture of the reorganization mechanism.

The aim of this study is to understand and develop techniques to process the fMRI data to help neurologists draw meaningful conclusions. Extracting information from fMRI data requires a wide range of image processing algorithms involving registration, normalization, smoothening etc. Statistical methods are used to quantify the results obtained from processing the image data. This report presents the various aspects of obtaining the required data for this study, the required processing steps and finally the results of such a processing.

II.fmri basics

fMRI is a non invasive technique used to spatially localize areas of activation in the brain when presented with a particular stimulus. It has been used extensively since its inception in the 1990s to map cognitive tasks to the physical structure of the brain. Due to rapid advances in Magnetic Resonance (MR) technologies and cognitive psychology, complex mental tasks have been designed and studied using fMRI.

A.Physics of Magnetic Resonance Imaging

fMRI is a method of scanning the brain using the basic MR methodology. MR technology is based on the nature of response of the hydrogen nuclei to radio frequency (RF) pulses in the presence of strong magnetic fields. Combining complex gradients of magnetic fields with varying strengths of the radio frequency excitation, it is possible to selectively excite parts of the organ under study. Due to the abundance of water molecules in the body, a comprehensive image of the same can be obtained using this method. The most common organ scanned using this technique is the human brain. Various methods are available depending on the combination of the strengths and phase of the RF energy, the magnetic gradients and most importantly, the acquisition time of the induced MR signal. The contrast of the different parts of the brain depends on the extent of excitation from the RF signals and the nature of its decay. Two main categories of image acquisition in MR are termed T1 and T2. T1 is an exponential time constant which represents the time taken for the excited nuclei to return to equilibrium after the RF pulse has been turned off. T2 is an exponential time constant describing the time it takes for the excited nuclei to lose signal, mainlydue to dephasing in the transverse plane [2]. There is another category called the T2* images which are also dependent on the amount of oxygen in the blood.

B.Blood Oxygen Level Dependent (BOLD) Response

The most wide spread technique in fMRI imaging, termed the BOLD fMRI, is based on the oxygen level of intra cranial blood [3]. Hemoglobin is a blood component which is responsible for the transportation of oxygen in the human body. Oxygen binds with hemoglobin molecules forming oxyhemoglobin which is then transported by the blood. Tissues extract oxygen from these molecules leaving behind deoxyhemoglobin. It was found experimentally that oxyhemoglobin and deoxyhemoglobin have different magnetic properties. This is the fundamental principle behind BOLD imaging.

The patient is subjected to a predetermined stimulus pattern while continuous brain volumes are being acquired. The stimulus is designed so as to activate only certain parts of the brain. Also, all stimulus paradigms have baseline and active phases, which provides for the contrast in the brain region scanned when the task is being carried out. The brain is in the resting state during the baseline condition. During the active phase of the stimulus, a certain part of the brain receives and begins to processes the signals. Processing needs energy and the brain tissue uses the oxygen in the cranial blood vessels during this process. As shown in Figure 1, at the onset of the stimulus there is a dip in the level of oxygen in the blood (Shown as a corresponding drop in the detected MR signal). The brain then proceeds to replenish this with an increased level of oxygenated blood flow into the active region, shown as the positive BOLD response in the figure. After the stimulus ceases, the oxygen level returns to the normal baseline level. The spatial resolution of the fMRI technique in identifying the region of activation is thus dependent on the change in blood flow rather than direct neurological activity. The temporal resolution of fMRI depends on the capacity of the scanning hardware to acquire volumes in short duration of time. An MR technique called EPI (Echo Planar Imaging) is used to speed up the volume acquisition time. EPI uses spatial encoding techniques which allow for complete brain volume scans to be acquired within 40 to 150 milliseconds [4].

Fig. 1. The time course of the BOLD response, also called the Hemodynamic Response Function (HRF). The dark bar below the time axis represents the duration of the stimulus (the active phase).

Fig. 2. The optic pathway in the human brain. The visual cortex is the anterior part of the brain. The two arcs shown represent the left and the right visual hemispheres.

III.the human optic system

The human brain is known to have different areas for processing specific tasks. The visual cortex, a region in the anterior part of the brain is the centre for vision processing. The optic nerves from both eyes are connected to this region, as shown in Figure 2. The two semicircular arcs represent the two visual hemispheres. The optic chiasm is a cross over point for the optic nerves. Signal from each eye for the overlapping region of the hemispheres are transmitted to both ends of the visual cortex. Whereas the signal from the non overlapping parts are only transmitted to the laterally inverted part (Left eye to the right part of the visual cortex, Right eye to the left part).

IV.fmri experiment setup and data collection

A.Stimulus used

The stimulus was designed keeping this structure in mind to activate all paths and regions of the visual cortex. A projector and a screen were setup on the MR machine to cover the field of view of the patient. The Presentation® software tool was used as the stimulus delivery program [5]. This software also allowed for the stimulus to be synchronized with the scan acquisition time. This information is important in the analysis of the fMRI data later. A visual stimulus in the form of a reversing (8 Hz) 8x8 black and white checker board pattern for the active phase and a centered red dot for the baseline phase was used. The stimulus starts with a 30 second fixation (baseline condition) followed by 30 seconds of active stimulus. This cycle was repeated for a duration of 8 minutes. 5 sub experiments (sessions) were carried out with the same stimulus using the setup described below:

Stimulus present in both visual hemispheres and both eyes are open
Stimulus present in both visual hemispheres, left eye is open and right eye is masked
Stimulus present in both visual hemispheres, right eye is open and left eye is masked
Stimulus present only in the left hemisphere and both eyes are open
Stimulus present only in the right hemisphere and both eyes are open

B.Data collected

A pilot scan was performed with a healthy 24 year old male volunteer. The 3T Philips MRI machine at the High Field MR Imaging Centre, UBC hospital was used. This initial data was collected with the intention of verifying the various parameters for the MRI scanning. It also served as a test data for the validation of the fMRI analyses approach. Some of the settings and parameters used are summarized in Table I. These settings were chosen to give an optimal balance between volume acquisition time (temporal resolution) and spatial resolution. Four sets of data were collected for four of the sub experiments in the bulleted list. The right hemisphere stimulus did not work as expected; hence data was not collected for the same.

C.Data formats, conversion and terminology

Data for each sub experiment is obtained from the scanner in the Philips native format. The scanner software does the required conversion from the K space (the nature of the data collected by the hardware) into the spatial domain. The native format consists of two files, a file with .PAR extension (called the data description file) contains information about the scan parameters and other custom data fed into the scanner. The .REC file contains the scan data, which in this experiment consists of 163 complete brain volumes collected over the course of one sub experiment.

To convert this proprietary data format to the more widely used Analyze file format, the free software MRIcro [6] was used. The multivolume PAR and REC files are converted to the Analyze format equivalents, the IMG and HDR files. The HDR file (the header file) contains the data from the PAR file, whereas the IMG (the image file) file contains the raw voxel values. MRIcro is then used to split these multivolume Analyze files into 163 individual IMG-HDR combinations.

The complete data set for the four sub experiments is termed one Run. Each of the four individual sub experiments (also called sessions) consists of 163 scans or volumes. Each of these volumes consist of 45 slices with an in plane resolution of 128x128 pixels. The time course (also referred to as the time series) of a voxel refers to the single dimensional signal obtained by concatenating the signal values of a particular voxel from each of the 163 volumes from a session. Since the stimulus used was active for a period of time it is termed block stimulus (as opposed to event stimulus, where stimulus is applied for a brief period of time only).

V.fmri data analysis

The goal of fMRI data analysis is to detect the activation changes due to the BOLD response. The detection scheme has to be robust and sensitive given that the signal levels could be as low as .5% of the mean value in a given region. Various artifacts like head motion, scanner field drift and field inhomogeneity contribute to the noise in the fMRI data.

Most approaches analyze the 4 dimensional data (the time series of the 3 dimensional volumes collected) by looking at the time courses of individual voxels. The most common approach is the Statistical Parameter Mapping (SPM), which involves the use of a GLM (Generalized Linear Model) to estimate the amount of signal present in the time course. Other approaches like the GBAM (Generalized Brain Activation Map) [7] use a similar approach, but use sinusoids to model the activation. Various other methods using the Fourier analysis and the Wavelet approach have been proposed. This study used the SPM approach using the SPM2 [8] toolbox by the Wellcome Department of Imaging Neuroscience. The use of SPM2 (and its predecessor versions) is widespread and the results obtained have been well received by the fMRI community.

VI.introduction to spm2

SPM2 is a Matlab® toolbox which provides for a suite of methods to analyze fMRI data. It provides for an optimal blend of pedagogy and processing efficiency by providing both the Matlab source files for the algorithms and the pre-compiled .MEX files for computing efficiency. SPM2 provides for an easy interface to perform all the data processing steps required to analyze fMRI data. It also provides for helpful graphs and figures to give the user a clearer picture of the various parameters available and noise artifact levels present.

The Matlab toolbox structure of this software makes it highly customizable. Researchers can use part of the SPM2 software modules and part of their own custom implementation to try out new approaches in dealing with the data. This initial study used the default data processing pipeline recommended for use with SPM. The various steps and the algorithmic details are provided in the sections below.

VII.importing data into spm2

SPM2 handles images in the Analyze file format. The data for each session has to be provided in the form of individual single volume files. SPM2 assumes a default orientation of the brain with respect to the coordinate system. Data exported by MRIcro was flipped to ensure that this requirement was met. Validity of the data was checked by manual inspection of the volumes using SPM2s display option. Orientation in this step refers to the general direction of the transverse and vertical axis of the brain with respect to the volume coordinates. Figure 3 shows the correct orientation of the image. The anterior part of the brain has to appear at the upper part of the lower left pane, while it has to appear to the left in the upper right pane.

Fig. 3. The Display option in SPM2. One of the volumes of the scan was used to check for the correct orientation of the brain within the acquired volume.

VIII.reorientation of volumes

The exact position of the subject inside the MRI machine varies with different runs and different subjects. It is beneficial in the data analysis procedure if all images are oriented to the same attitude before the processing starts. Reorientation is applied to all images in a session using rigid body transforms.