Division of Radiological Sciences

Link¨oping University Medical Dissertations, No. 1121

F ^UNCTIONAL M ^AGNETIC R ^ESONANCE I ^{MAGING FOR} C ^LINICAL D ^IAGNOSIS

-E XPLORING AND I MPROVING THE E XAMINATION C HAIN

M ^ATTIAS R ^AGNEHED

Division of Radiological Sciences

Department of Medical and Health Sciences

Center for Medical Image Science

and Visualization (CMIV)

Link¨oping University, Sweden, 2009

Functional Magnetic Resonance Imaging for Clinical Diagnosis -Exploring and Improving the Examination Chain

2009 Mattias Ragnehed c

Printed by LiU-tryck, Link¨oping 2009.

ISBN: 978-91-7393-645-3

ISSN 0345-0082

A BSTRACT

Functional Magnetic Resonance Imaging (fMRI) is a relatively new imaging tech- nique, first reported in 1992, which enables mapping of brain functions with high spatial resolution. Functionally active areas are distinguished by a small signal increase mediated by changes in local blood oxygenation in response to neural activity. The ability to non-invasively map brain function and the large number of MRI scanners quickly made the method very popular, and fMRI have had a huge impact on the study of brain function, both in healthy and diseased subjects.

The most common clinical application of fMRI is pre-surgical mapping of brain functions in order to optimise surgical interventions.

The clinical fMRI examination procedure can be divided into four integrated parts: (1) patient preparation, (2) image acquisition, (3) image analysis and (4) clinical decision. In this thesis, important aspects of all parts of the fMRI ex- amination procedure are explored with the aim to provide recommendations and methods for prosperous clinical usage of the technique.

The most important results of the thesis were: (I) administration of low doses

of diazepam to reduce anxiety did not invalidate fMRI mapping results of primary

motor and language areas, (II) the choice of visual stimuli equipment can have se-

vere impact on the mapping of visual areas, (III) three-dimensional fMRI imaging

sequences did not perform better than two-dimensional imaging sequences, (IV)

adaptive spatial filtering can improve the fMRI data analysis, (V) clinical deci-

sions should not be based on activation results from a single statistical threshold.

L IST OF P APERS

This thesis is based on the following papers, which will be referred to by their Roman numerals. The articles are reprinted with permission from the respective publisher.

P

I Influence of Diazepam on clinically designed fMRI

M. Ragnehed, I. H˚akansson, M. Nilsson, P. Lundberg, B. S¨oderfeldt, M. Engstr¨om. Journal of Neuropsychiatry and Clinical Neuroscience, 2007; 19(2):164-172. doi:10.1176/appi.neuropsych.19.2.164 P

II Projection screen or video goggles as stimulus modality in func-

tional magnetic resonance imaging

M. Engstr¨om, M. Ragnehed, P. Lundberg. Magnetic Resonance Imag- ing, 2005; 23:695-699. doi:10.1016/j.mri.2005.04.006 P

III Visual Grading of 2D and 3D fMRI compared to image based de-

scriptive measures

M. Ragnehed, O. Dahlqvist Leinhard, J. Pihlsg˚ard, S. Wirell, H. S¨okjer, P. F¨agerstam, B. Jiang, ¨ O. Smedby, M. Engstr¨om, P. Lundberg. Sub- mitted Manuscript, 2009

P

IV Restricted Canonical Correlation Analysis in Functional MRI - validation and a novel thresholding technique

M. Ragnehed, M. Engstr¨om, H. Knutsson, B. S¨oderfeldt, P. Lund- berg. Journal of Magnetic Resonance Imaging, 2009; 29(1):146-154.

doi:10.1002/jmri.21494

P

V Brain lateralisation assessed by fMRI and dichotic listening

H.M. Van Ettinger-Veenstra, M. Ragnehed, M. H¨allgren, T. Karlsson,

A-M. Landtblom, P. Lundberg, M. Engstr¨om. Manuscript

List of Papers vi

Contributions to Papers

P

I In this project I was responsible for the study design and the fMRI data acquisition. I also performed the fMRI data analysis and was the main author of the manuscript.

P

II I was involved in the planning and set-up of the study, participated in the design of the visual stimuli and did some of the analysis of the data. I also made important contributions to the interpretation of the results.

P

III I performed all fMRI data analysis and I was responsible for design and performance of the Visual Grading. In addition I performed all statistical evaluations and wrote the manuscript.

P

IV In this project I was responsible for data collection and all data anal- ysis. I came up with the idea for and developed the significance esti- mation method and wrote the manuscript.

P

V I was involved in the planning and design of the study and came up with the idea to include dichotic listening. I did some of the fMRI analysis and wrote part of the manuscript.

Other peer reviewed publications not included in the thesis Regular articles

• Paradigm design of sensory-motor and language tests in clinical fMRI.

M. Engstr¨om, M. Ragnehed, P. Lundberg, B. S¨oderfeldt. Clinical Neuro- physiology, 2004; 34(6):267-277

Abstracts

• Influence of performance-related language ability on cortical activa- tion.

H.M. Veenstra, J. Pettersson, C. Nelli, M. Ragnehed, A. McAllister, P.

Lundberg, M. Engstr¨om. Human Brain Mapping, Honolulu, 2009.

• Brain lateralization assessed by fMRI and dichotic listening.

H.M. Veenstra, M. Ragnehed, M. H¨allgren, P. Lundberg, M. Engstr¨om. Hu-

man Brain Mapping, Honolulu, 2009.

vii List of Papers

• Using Visual Grading Characteristics for the evaluation of different fMRI data acquisition methods

M. Ragnehed, O. Dahlqvist Leinhard, ¨ O. Smedby, M. Engstr¨om, P. Lund- berg. European Society for Magnetic Resonance in Medicine and Biology, Valencia, 2008.

• Does diazepam influence the BOLD response? M. Ragnehed, M. En- gstr¨om, P. Lundberg. International Society for Magnetic Resonance in Medicine, Berlin, 2007.

• Influence of diazepam on clinically-designed fMRI. B. S¨oderfeldt, M.

Ragnehed, I. H˚akansson, P. Lundberg, M. Nilsson, J. Ahlner, M. Engstr¨om.

Journal of Neuropsychiatry and Clinical Neuroscience, 18, 2006.

• LI and the effect of thresholding. M. Ragnehed, M. Engstr¨om, B. S¨oder- feldt. European Society for Magnetic Resonance in Medicine and Biology, Copenhagen, 2004.

• Comparison between fMRI and Wada test. L. B¨orjesson, J. Stockhaus, H. Gauffin, M. Ragnehed, P. Lundberg, B. S¨oderfeldt. EPILEPSIA, 45(S3), 2004.

• Quantitation of atherosclerosis in a minipig model with MRI and dy- namic contours. ¨ O. Smedby, M. Ragnehed, A. Knutsson, L. Jacobsson, X. Yuan, R. Andersson. Society of Cardiovascular Magnetic Resonance, Barcelona, 2004.

• Localization of Signed and Heard Episodic and Semantic Memory Tasks using fMRI. P. Nystr¨om, M. Ragnehed, O. Friman, M. Engstr¨om, P. Lundberg, H. Knutsson, B. S¨oderfeldt. The 9th International Conference on Functional Mapping of the Human Brain, June 19-22, 2003, New York, NY. Neuroimage, 19(2).

• Comparing CCA and SPM99. M. Ragnehed, O. Friman, P. Lundberg, B.

S¨oderfeldt, H. Knutsson. The 9th International Conference on Functional

Mapping of the Human Brain, June 19-22, 2003, New York, NY. Neuroim-

age, 19(2).

L IST OF A BBREVIATIONS

2D Two-Dimensional

3D Three-Dimensional

#voxels number of activated voxels

ADP Adenosine Di-Phosphate

ANOVA Analysis of Variance

ATP Adenosine Tri-Phosphate

auROC area under ROC curve

BOLD Blood Oxygenation Level Dependent

CBF Cerebral Blood Flow

CBV Cerebral Blood Volume

CCA Canonical Correlation Analysis

CMR Cerebral Metabolic Rate

CMRGlc Cerebral Metabolic Rate of Glucose CMRO

Cerebral Metabolic Rate of Oxygen

CT Computed Tomography

EEG ElectroEncephaloGraphy

Abbreviations x

EPSP Excitatory Post-Synaptic Potential FWHM Full Width at Half Maximum

GABA Gamma-AminoButyric Acid

GLM General Linear Model

hr Haemodynamic Response

hrf Haemodynamic Response Function

IPSP Inhibitory Post-Synaptic Potential

LI Lateralisation Index

MEG MagnetoEncephaloGraphy

MRI Magnetic Resonance Imaging

mROC modified ROC

PET Positron Emission Tomography

PRESTO PRinciples of Echo-Shifting with a Train of Observations ROC Receiver Operating Characteristic

SENSE SENSitivity Encoding

SNR Signal to Noise Ratio

SPECT Single Photon Emission Computed Tomography

T Tesla

TCA Tri-Carboxylic Acid

TMS Transcranial Magnetic Stimulation

TR Repetition Time

tSNR temporal Signal to Noise Ratio

VG Visual Grading

VGC Visual Grading Characteristic

T ABLE OF C ONTENTS

Abstract iii

List of Papers v

Abbreviations x

I Introduction 1

C

1 Introduction 3

1.1. Medical Imaging . . . . 3

1.2. fMRI History . . . . 4

C

2 Brain Activity 7 2.1. Neuronal Activity . . . . 7

2.2. Neuronal Energy Requirements . . . . 9

2.3. Physiological Effects of Neuronal Activity . . . . 10

C

3 Functional MRI 13 3.1. Detecting Brain Activity . . . . 13

3.1.1. Paradigm Design . . . . 13

3.1.2. Pre-processing . . . . 14

3.1.3. The BOLD Response . . . . 15

3.1.4. Analysis . . . . 17

Table of Contents xii

C

4 Clinical Applications of Functional MRI 19

4.1. Special Requirements . . . . 20

4.2. Tumour Resection . . . . 20

4.2.1. Task Selection . . . . 21

4.3. Surgical Treatment of Epilepsy . . . . 22

4.3.1. Source Localisation . . . . 23

4.4. Brain Plasticity . . . . 23

C

5 Aims 25 II Methods, Results and Discussions 27 C

6 Patient Preparation 31 6.1. Anxiolytics and fMRI . . . . 31

6.1.1. Background . . . . 31

6.1.2. Methods . . . . 32

6.1.3. Results . . . . 33

6.1.4. Discussion . . . . 33

C

7 Data Acquisition 35 7.1. Visual Stimulus Delivery . . . . 35

7.1.1. Background . . . . 35

7.1.2. Methods . . . . 36

7.1.3. Results . . . . 36

7.1.4. Discussion . . . . 37

7.2. Imaging Sequences . . . . 37

7.2.1. Background . . . . 38

7.2.2. Methods . . . . 39

7.2.3. Results . . . . 39

7.2.4. Discussion . . . . 39

C

8 Analysis 41 8.1. Statistical Analysis . . . . 41

8.1.1. Background . . . . 41

8.1.2. Methods . . . . 42

8.1.3. Results . . . . 42

8.1.4. Discussion . . . . 43

C

9 Interpretation of fMRI Results 45

9.1. Objective and Subjective Measures . . . . 45

xiii

9.1.1. Background . . . . 45

9.1.2. Methods . . . . 46

9.1.3. Results . . . . 46

9.1.4. Discussion . . . . 47

9.2. Significance Testing of RCCA Results . . . . 48

9.2.1. Background . . . . 48

9.2.2. Methods and Results . . . . 48

9.2.3. Discussion . . . . 51

9.3. Thresholding Issues for Clinical fMRI . . . . 52

9.3.1. Background . . . . 52

9.3.2. Lateralisation of Brain Function . . . . 52

9.3.3. Subjective Threshold Selection . . . . 53

9.3.4. Presurgical fMRI . . . . 54

9.3.5. Discussion . . . . 55

C

10 Summary 57

Bibliography 59

Acknowledgements 73

III Papers 75

Part I

Introduction

1

I NTRODUCTION

This thesis is divided into three different parts. In Part I, Introduction, material necessary to understand the concept of fMRI is presented. Then some clinical applications of fMRI that highlights the importance of proper examination proce- dures are introduced. Finally the specific aims of this thesis are listed. In Part II, Methods, Results and Discussions, the research conducted to answer the specific research questions raised in Chapter 5, Aims, is reviewed, followed by a discus- sion of the achieved results. In Part III, Papers, the articles that provide the basis of the thesis are reprinted.

1.1 Medical Imaging

Medical imaging, including technologies such as microscopy, ultrasound, X-ray, Computed Tomography (CT), Magnetic Resonance Imaging (MRI) etc., have be- come an immensely important tool in many medical disciplines. The examination chain for medical imaging involves several important procedures,

1. patient preparation 2. image acquisition 3. image analysis

4. visualisation and diagnosis

All procedures of the examination chain have to be performed in a professional

and standardised manner in order to provide proper information to the clinician

who is responsible for making the diagnose.

Chapter 1. Introduction 4

fMRI is a perfect example to demonstrate the importance of all procedures involved in the medical imaging examination chain:

1. the patient (or test subject) must be prepared for the examination and receive proper instructions on how to perform the behavioural task presented during scanning. If the task is not executed as intended the result of the examination will be misleading, other areas then expected might be highlighted and/or some desired functional areas will not be detected at all

2. it is important that the selected data collection sequence has good functional contrast, limited signal drifts and provide enough tissue contrast to allow proper registration

3. robust and efficient statistical data analysis methods are required to extract the functional information in the quite noisy fMRI data

4. the statistical results need to be properly presented to ensure correct inter- pretation of the results

If these steps are performed in a standardised and professional manner, fMRI can be a useful clinical tool. This chain of events will be referred to throughout the thesis, especially Part II, Methods, Results and Discussions will closely follow the flow of the clinical imaging examination chain.

1.2 fMRI History

In 1990 Ogawa et al. [1990] performed some MRI experiments on rats at high field strengths. By manipulating the blood oxygenation level they found that de-oxygenated blood caused distortions on gradient-echo images. They specu- lated that this effect, which later was termed Blood Oxygenation Level Dependent (BOLD) contrast, would make it possible to measure changes in brain activity.

In 1992 the first reports of functional mapping using the BOLD contrast were reported. Using long blocks of sustained visual stimulation followed by rest, Kwong et al. [1992] reported a sharp increase of the MRI signal in relevant brain regions that remained for the whole stimulation period. The result was replicated in a study published shortly afterwards by Ogawa et al. [1992]. Bandettini et al.

[1992] used a motor task to induce brain activation and obtained similar results.

Later that year, Blamire and colleagues reported that even short duration stimuli

gave rise to the same kind of MR signal increase [Blamire et al., 1992]. How-

ever, they also noted that there was a small delay, of approximately 3.5 seconds,

between stimulus onset and the observable signal increase.

5 1.2 fMRI History

Shortly after the introduction of BOLD fMRI in 1992 [Bandettini et al., 1992;

Blamire et al., 1992; Kwong et al., 1992; Ogawa et al., 1992], several research groups repeated and extended their experiments. Ever since, fMRI has been used to answer research question about the functional organisation of the human brain.

Nowadays, MRI scanners capable of performing functional imaging are available in most hospitals. The high availability of MR scanners has resulted in an ex- plosion of fMRI studies, see Figure 1.1, and fMRI is now the standard tool for functional neuroimaging. As the fMRI technique has matured and been refined the acceptance to use fMRI for clinical purposes is increasing. The most common clinical application of fMRI is perhaps pre-surgical mapping of eloquent areas and evaluation of hemispheric language dominance prior to temporal lobectomy of certain epilepsy patients.

1994 1996 1998 2000 2002 2004 2006 2008

0 400 800 1200 1600 2000

YEAR

#PUBLICATIONS

fMRI PUBLICATIONS IN PUBMED

F

1.1. Publications on fMRI by year. The number of fMRI publications in

major scientific journals is has increased rapidly since its inception in 1992.

2

B RAIN A CTIVITY

Understanding the functional organisation of the intriguing human brain has been the ultimate goal for scientists for a long time. A multitude of methods have been used in the search for knowledge about the inner functioning of the brain. Im- portant information has been obtained by studying functional deficits caused by brain lesions. More recently, the electric and magnetic signals originating from neural activity have been studied using Electroencephalography (EEG) and Mag- netoencephalography (MEG). Since the electrodes have to be placed outside of the skull, the ability to identify the location of the neuronal signals is limited. Using techniques like positron emission tomography (PET) and single photon emission computed tomography (SPECT), it became possible to acquire images of brain function. The main drawback of these imaging techniques is that they rely on the use of radioactive tracers, which has limited their use. With the advent of fMRI, a completely non-invasive alternative for functional imaging became available. In this chapter the principles of BOLD fMRI will be reviewed. Most of the material has been extracted from the textbooks by Aguirre and D’Esposito [2001],Brodal [2003], Buxton [2002], Huettel et al. [2004], and Jezzard et al. [2001].

2.1 Neuronal Activity

The human brain contains about 10

neurons. About 20% of the neurons are lo-

cated in the cerebral cortex. Just like any other cell, the neuron has a cell body

containing cytoplasm, organelles and a nucleus. A typical neuron, see Figure 2.1,

has branches, called dendrites, forming the dendritic tree, which is the main in-

formation receiving network of the neuron. The number of dendrites and the size

and shape of the dendritic tree shows great variability. The neurons also have one

axon. The axon is used to transmit information to other neurons through connec-

Chapter 2. Brain Activity 8

tions, called synapses, located on the dendritic tree or directly on the cell body itself. Even though each neuron only has one single axon, the neuron can com- municate with several other neurons due to the fact that the axon usually undergo extensive branching. It has been estimated that there is about 10

synapses in the human brain, which means that on average each neuron has a few thousand synapses [Drachman, 2005].

F

2.1. A detailed view of a typical neuron with a dendritic tree and a branching myelinated axon. [Picture from Wikipedia Commons.]

The chemical environment in the brain is maintained at a state which is far away from chemical equilibrium. For instance, inside the neuron the concentra- tion of K

is larger than on the outside, and on the outer side of the neuronal membrane there is a greater concentration of Na

, Ca

and Cl

than on the in- side. Thus there is an electrical potential between the inside and outside of the cell membrane (negative inside). This resting membrane potential is about −40 mV.

Since the neuronal membrane prevents unrestricted diffusion of ions along their

concentration gradients, they are unable to reach chemical equilibrium. However,

the membrane contains selective ion channels where specific ions are allowed to

pass. These ion channels have gating mechanisms, which open or close for ion

transport, controlled by the action of specific molecules. The gates may also be

9 2.2 Neuronal Energy Requirements

voltage dependent, opening when the electrical potential across the membrane reaches a threshold value. Apart from passive diffusion along concentration gra- dients through ion channels, there are mechanisms for transporting ions against the concentration gradients. These so called ion pumps require energy, which is released by adenosine triphosphate (ATP), for their action.

The activity of neurons may be divided into integrative or signalling activity.

Integrative activity refers to the collection and integration of inputs from many synapses on the dendritic tree, while signalling activity refers to transmission of the result of the integration of inputs to other neurons. The exchange of informa- tion at the synapses is mediated by different neurotransmitters released by the pre- synaptic process of the axon. The neurotransmitters influence the environment at the post-synaptic membrane inducing either an excitatory post-synaptic potential (EPSP) or an inhibitory post-synaptic potential (IPSP). The most common neu- rotransmitters are glutamate and gamma-Aminobutyric acid (GABA). Glutamate opens normally closed Na

channels resulting in an influx of Na

into the cell thereby reducing the electrical potential across the membrane. The local depolar- isation of the post-synaptic membrane is an example of an EPSP. Thus, glutamate is an excitatory neurotransmitter [Meldrum, 2000]. GABA on the other hand en- ables an active influx of Cl

and efflux of K

, thus increasing the membrane potential, a hyperpolarisation. This is known as an IPSP and accordingly, GABA is known as an inhibitory neurotransmitter [Li and Xu, 2008].

The neuron integrates the effects of the EPSPs and IPSPs from all of its synapses. If the net effect, over a brief time interval, is to depolarise the mem- brane below a threshold voltage a large number of voltage gated sodium channels will open. There will be a large influx of Na

into the cell and it will be fur- ther depolarised and even more ion channels will open. This depolarisation wave will propagate down the axon and it is known as an action potential or ’nerve im- pulse’. When the action potential reaches the pre-synaptic axon terminal a num- ber of events occur that leads to the release of neurotransmitters into the synaptic cleft. The transmitter interacts with receptors that control ion channels on the post-synaptic membrane initiating either an EPSP or IPSP.

2.2 Neuronal Energy Requirements

The human brain has no internal energy stores and thus requires a continuous sup- ply of energy and oxygen. Constituting a mere 2–3% of the body mass, the brain still receives about 15% of the total cardiac output of blood [Siegel et al., 1999].

The distribution of blood within the brain is heterogeneous, gray matter receives

several times more blood per gram of tissue than white matter. The amount of

blood delivered per gram of tissue to gray matter is comparable to that in heart tis-

Chapter 2. Brain Activity 10

sue. When a group of neurons become active their metabolic demands increase, which results in regional changes in cerebral blood flow.

The production of IPSPs, EPSPs and action potentials does not consume any energy. They simply use the fact that the system is in a non equilibrium state, and their actions moves the system closer to equilibrium. However, energy is required in order to restore the resting potential and concentration gradients of the different ions after neural activity. To regenerate the resting potential and ion concentration gradients so called ion pumps are utilised. Since these pumps act to restore a non equilibrium state they require a source of energy to function.

The primary source of energy in the human brain is ATP. Conversion of ATP to adenosine diphosphate (ADP) releases a large amount of energy. In order to utilise the energy stored in the ATP/ADP system the conversion of ATP to ADP is directly coupled to energy consuming processes. For instance, the Na/K-pump, which moves three Na

ions out of the cell and two K

into the cell, consumes one ATP molecule. The Na/K-pump is important for the recovery of the neuron following an action potential and for maintaining the resting potential. At the synapse there are a number of other mechanisms involved in the recovery after an action potential. For instance, the neurotransmitters have to be brought back into the pre-synaptic terminal. This process is governed by mechanisms that consumes ATP. In summary, the neural activity itself does not require any energy but the restoration of the chemical gradients and the resting potential does.

After ATP has been consumed, the ADP must be converted back into ATP.

This is mainly done through glycolysis, in which the consumption of one glucose molecule results in two ATP and a rest product called pyruvate. The pyruvate is either further reduced into lactate (anaerobic glycolysis) or, if oxygen is present, the pyruvate enters a complex chain of reactions called aerobic glycolysis, where an additional 34 ATP is produced. Anaerobic metabolism, generating only two ATP per consumed glucose molecule, is very inefficient when compared to aero- bic metabolism, generating 36 additional ATP. However, anaerobic metabolism is about 100 times faster than aerobic metabolism.

2.3 Physiological Effects of Neuronal Activity

As early as 1890 the influential psychologist William James wrote ([James, 1890], as quoted by Buxton [2002]):

We must suppose a very delicate adjustment whereby the circulation

follows the needs of the cerebral activity. Blood very likely may rush

to each region of the cortex according as it is most active, but of this

we know nothing.

11 2.3 Physiological Effects of Neuronal Activity

With the introduction of PET it was possible to test James’s assumption on humans. Using PET one can measure cerebral blood volume (CBV), cerebral blood flow (CBF), cerebral metabolic rate of glucose consumption (CMRGlc), and cerebral metabolic rate of oxygen consumption (CMRO

). Several studies on animals and humans have established that there is a close relationship be- tween local functional activity and local glucose metabolism. Activation studies have revealed increased CMRGlc in activated brain regions. In addition, CMR- Glc shows a response which correlates with different levels of functional activity [Schwartz et al., 1979; Sokoloff, 1977; Kadekaro et al., 1985]. A PET study with a somatosensory task showed a task induced increase of CBF of 26% and CMR- Glc increase of 17% [Ginsberg et al., 1988]. In another study of the sensorimotor cortex with PET an 50% increase of both CBF and CMRGlc was observed during activity [Fox et al., 1988].

While glucose consumption and blood flow increases substantially and by ap- proximately the same amount during activation the oxygen consumption rate does not. Large discrepancies between CBF and CMRO

changes have been observed, while CBF increases 25–50% CMRO

increases only a few percent [Fox et al., 1988]. Using PET, the ratio of percent change in CBF to percent change in CMRO

have been found to be 3–6 [Fox and Raichle, 1986]. This imbalance results in a substantial drop in deoxy-haemoglobin in venous blood.

In summary, during brain activation there is a large increase in CBF and CMR-

Glc localised to the activated brain region. The large increase of blood flow makes

PET and fMRI possible. It is also important to realise that the blood flow change

also depend on the intensity of the stimuli such that the flow changes reflects the

amount of neural activity. The increase in oxygen consumption is, however, much

smaller than the flow increase during brain activity. This leads to a lower oxygen

extraction fraction during activity and thus an increased deoxy-haemoglobin con-

tent in venous blood. The drop in deoxy-haemoglobin content forms the basis of

the BOLD response used in most fMRI studies, see Chapter 3, Functional MRI.

3

F UNCTIONAL MRI

3.1 Detecting Brain Activity

When performing ordinary brain imaging, one image is often sufficient to classify different tissue types. The fundamental difference with functional brain imaging is that one is interested in evaluating signal changes over time, thus images are collected for an extended time interval, say 5–10 minutes. During the fMRI ex- periment image volumes are continuously collected with a repetition time (TR) of 2–4 seconds, resulting in a total of 200–300 images for the whole time-series.

During the image acquisition, the subject performs a specific task interleaved with a control task (rest). There are two main types of task sequence designs (paradigms), event-related designs [Dale and Buckner, 1997] and block designs.

In event-related designs the subject is confronted with short duration (usually <

1 second) stimuli interleaved with a base-line condition of longer duration (usu- ally 5–30 seconds). In blocked designs the neural activity is sustained for longer periods, usually 15–30 seconds, inter-spaced with periods of baseline activity.

To identify voxels activated by the task the collected voxel time-series are compared to a model of the expected BOLD signal of an activated voxel. Voxels whose time-series resemble the model close enough are then labelled as active, which results in a map of brain structures activated by the behavioural paradigm used.

3.1.1 Paradigm Design

When designing a fMRI paradigm it is most important to determine exactly which

function(s) to map. Then a task which involves the desired function(s) and a con-

trol task that. Ideally, the control task should elicit the same cognitive processes

Chapter 3. Functional MRI 14

as the activation task except for the function(s) to map. Designed this way the experiment, relying on the concept of pure insertion [Price and Friston, 1997;

Zarahn et al., 1997a], will highlight the desired function(s).

Early fMRI studies did not use any explicit base-line task, instead the subject was simply told to relax between the task periods. It has been shown that during passive rest significant cognitive processes are present. For instance the brain may be engaged in planning and problem solving, episodic memory encoding and free thoughts. These are all processes which one cannot control. It is therefore very important to design a control task that keeps the subject in a desired cognitive state [Binder et al., 1999; Stark and Squire, 2001; Luca et al., 2006].

As stated earlier there are two main categories of paradigms, blocked de- signs and event-related designs, which have different strengths and weaknesses.

Blocked designs have high detection power, are quite insensitive to small varia- tions of the individual BOLD response, and are conceptually easy to understand.

Event-related designs are appropriate for estimation of the BOLD response, to find timing differences between activated regions and they are more flexible than blocked designs [Aguirre and D’Esposito, 2001; Huettel et al., 2004]. In most clinical applications high detection power is desirable, thus blocked designs are far more common than event-related paradigms in clinical settings.

If possible, behavioural data should always be recorded. Otherwise it is im- possible to know for sure that the subject performed the task as instructed and no conclusions can be drawn from the results of the experiment. It is crucial that the task instructions are very clear so that the subject performs the task correctly. It is also beneficial to train the subject using a realistic training version of the paradigm prior to the functional scanning session to reduce the risk of faulty execution of the task [Stippich, 2007].

3.1.2 Pre-processing

Before the statistical analysis of the data is performed a number of pre-processing steps needs to be done. Even when different devices such as bite bars, vacuum cushions or rubber bands, are used to restrict involuntary head movements, head motion during image acquisition is inevitable. Thus one has to re-orient the im- ages to make sure that any voxel contain the same brain tissue in all image volumes [Friston et al., 1996; Ardekani et al., 2001; Morgan et al., 2007].

In addition, the image slices within a volume are usually collected at different

times. This means that the BOLD response to a stimuli is sampled at different

times for different slices. This will reduce the possibility to detect activated vox-

els, especially for long TR, event-related designs. Thus, in some cases it is impor-

tant to apply a temporal interpolation such that all data appear to be sampled at

the same time [Aguirre et al., 1998].

15 3.1 Detecting Brain Activity

It is also common practice to smooth, i.e.spatially low-pass filter, the images with a Gaussian kernel of pre-specified size before the statistical analysis. The Gaussian filter is often recommended to be about 2.5 × voxel-size full width at half maximum (FWHM). Low-pass filtering is utilised to improve the signal to noise ratio (SNR) in the images. Since the images are contaminated with random noise, a weighted average of neighbouring voxels will reduce the noise compo- nent. Smoothing, however, reduces the spatial resolution and after smoothing fine scale details of the images are lost. The optimal size of the smoothing filter is dependent on image data properties as well as the specific aims of the study [Tri- antafyllou et al., 2006; Weibull et al., 2008b]. Hence, it is important to adjust the FWHM of the smoothing filter to match the requirements of the specific study.

In addition, the voxel time-courses are usually contaminated by low-frequency noise [Zarahn et al., 1997b]. Such signal drifts during the experiment can be caused by scanner instability, head motion and breathing. If left unmodelled the low frequency noise will reduce the detection power of activated voxels. Thus it is important to remove the signal drifts, for example by an ordinary high-pass filter.

3.1.3 The BOLD Response

Brain activity causes a slight change of the MR signal known as the haemody- namic response (hr). The hr depends, apart from physiological factors, primarily on the duration of the stimuli. The basic shape of the hr is the same over subjects and brain regions, making it is possible to generate models of the hr. Several ac- curate haemodynamic response function (hrf) have been suggested [Friston et al., 1995; Buxton et al., 2004]. Since the hr is (nearly) linear [Boynton et al., 1996;

Dale and Buckner, 1997], the hrf can be used to accurately model the predicted BOLD response to any task given a reasonable model of the task induced neu- ral activity. The predicted BOLD response is given by convolving the hrf with the model of the neural activity. However, it is important to note that there are regional as well as individual differences in the shape of the BOLD response [Aguirre et al., 1998]. To capture these differences in the BOLD response, it is common to use a model that consist of a set of basis function that include derivatives of the canon- ical hrf. By forming linear combinations of the basis functions, subtle changes in shape and latency of the hr are captured. In Figure 3.1 signal changes after short duration sequential finger tapping is shown. In the left panel several individual responses are shown and in the right panel the mean of those responses is shown.

The mean response closely matches the shape of the hrf shown in the left panel of Figure 3.2.

The large variability of the hr is due to the presence of noise and the fact

that the amplitude of the BOLD signal change is quite small, 0.5–3% for 1.5

T MR scanners. This is about the same order of magnitude as the noise in the

Chapter 3. Functional MRI 16

F

3.1. Measured haemodynamic responses (hr) in the motor cortex after from a finger tapping task. Left panel shows the individual haemodynamic response (hr) to several separate stimuli of an activated area. Right panel shows the mean of those responses.

images. Thus, in order to correctly identify BOLD signal changes that are due to neural activity one has to repeat the stimuli several times to allow averaging of the response.

As shown in Figure 3.1, the first measurable changes in the BOLD signal oc- curs after about 3.5 seconds, it then rises to its peak value at about six seconds after the onset of brief neural activity. For long duration activity, see Figure 3.2, the peak extends into a plateau. After reaching its peak, the BOLD signal de- creases to below baseline and remain there for several seconds before returning to

F

3.2. Left panel shows the modelled response to a short duration stimuli,

indicated by a spike at 0 s. Right panel shows the modelled response after a

long duration activity indicated by a boxcar starting at 0 s.

17 3.1 Detecting Brain Activity

F

3.3. An example of an experimental block paradigm and the BOLD model constructed by convolution of the paradigm and the haemodynamic response function (hrf).

baseline.

3.1.4 Analysis

To detect voxels activated by the task, a model of the expected signal changes in- duced by the paradigm is compared statistically with the collected BOLD signal time-series using correlation or linear regression. This analysis is usually per- formed within the General Linear Mode (GLM) framework [Friston et al., 1995;

Worsley and Friston, 1995]. Activations are assessed by calculating the statistical significance of the coefficients, most often using a t-test or F-test of the correlation coefficients or linear regression coefficients. The signal model is constructed by convolution of the paradigm time-course with the hrf, see Figure 3.3.

The GLM is formulated as

Y = Xβ + , (3.1)

where Y is a matrix of observed voxel time-series, X is a matrix of basis functions,

 is an error term with variance σ

and β a vector of parameters to be estimated.

The least squares solution to Eq. (3.1) can be written as β = X ˆ

X

X

Y. (3.2)

The estimated parameters ˆ β describe the amplitude of the task induced BOLD

response. If ˆ β 6= 0 then there is a task induced effect in the acquired images.

Chapter 3. Functional MRI 18

However, since the data are contaminated with noise, there is always the possibil- ity that the ˆ β become non-zero by chance alone. To control the effect of random noise, the significance of the estimated parameters can be tested by calculating a test statistic T and then compare the T -values with a known distribution. To test if the linear combination c

β 6= 0, the T -statistic of the linear combination c ˆ

β is ˆ given by

T = c

β ˆ

pσ

c

(X

X)

c) , (3.3) which is compared to a Students t-distribution. If the statistic value exceeds the 99th percentile, the effect is said to be statistically significant at p < 0.01. The statistical analysis results in an image of the statistical value for each voxel. An ac- tivation map is created by considering all voxels where the statistic value exceeds a specified threshold value, for instance the T -value corresponding to p < 0.001, to be active. The significance level is chosen to ensure that the number of false positives is kept low enough. If p < 0.001 and an image volume of 64 × 64 × 30 voxels is analysed, more than 100 voxels being active due to chance alone is ex- pected. The resulting activation map is usually colour coded and overlaid on an anatomical reference image, see Figure 3.4.

F

3.4. An example of fMRI results overlaid on an anatomical reference

image. Data is from a left hand finger tapping task. The subject has a

tumour on the right hand side of the brain.

4

C LINICAL A PPLICATIONS OF

F UNCTIONAL MRI

fMRI has had a huge impact on understanding normal brain function. How- ever, its use in clinical practice has been relatively limited. One important rea- son is that the fMRI procedure, including patient information, scanning param- eters, paradigm design and data analysis procedures, are not sufficiently stan- dardised. Thus recommendations and improvements to all parts of the imaging chain presented in Chapter 1, Introduction, are needed. Despite these limitations, applications are emerging in both clinical practice and in clinical neurosciences [Matthews et al., 2006]. Some clinical application of fMRI will be described in this chapter, namely:

I Localisation of normal brain function in relation to brain lesions, see Sec- tions 4.2 and 4.3.

I Demonstration of abnormal brain function, see Section 4.3.1.

I Elucidation of functional development and recovery mechanism, see Sec- tion 4.4.

Some of the problems encountered in these applications are the main motive for the research aims to be addressed in this thesis, see Chapter 5, Aims.

A fundamental question when fMRI is used in the clinic is whether the activa-

tion pattern obtained in a single session truly represents the functional anatomy of

the subject. Several studies have focused on the issue of reproducibility of fMRI

results. In general, it has been shown that there is good inter-session reproducibil-

ity [Otzenberger et al., 2005; Smith et al., 2005]. However, it is important to

ascertain that the activation tasks are well designed and that the task is restrictive

Chapter 4. Clinical Applications of Functional MRI 20

enough to force the subject to perform the task in a consistent way. In addition, during the functional task, it is important to monitor behaviour to make sure that the subject is really performing the task as expected. An additional difficulty is the fact that patients often suffer from impaired functions, which may render it difficult to perform certain tasks. Therefore it is most important to carefully adapt the tasks to match the abilities of the patient [Sunaert, 2006; Stippich, 2007].

4.1 Special Requirements

In general, a fMRI examination of a patient require more resources than exam- inations of healthy volunteers. Many fMRI patients have neurological deficits that, depending on the location of the lesion, influence their behaviour in different ways. Most prevalent among the fMRI patients are disturbed motor functions, lan- guage and memory deficits and reduced mental capacity. These conditions make it more difficult to perform successful functional imaging, mainly because the pa- tients find it difficult to perform the cognitive or sensory-motor task. However, in most cases it is still possible to obtain satisfactory results if sufficient training is performed with patient prior to the fMRI examination.

In many cases it is important not only to practice the task beforehand, but also to adapt the task to the individual patient. In case of partial paresis, the patient may not be able to perform some movements, it is then necessary to adapt the motor task to match the ability of the the individual patient. If complete paresis of the targeted limb is present, knowledge about the functional organisation may be achieved by manually moving the affected limb. Some information may also be gained by examining the contralateral limb [Stippich, 2007].

4.2 Tumour Resection

Localisation of eloquent brain functions prior to tumour resection is the most well established clinical application of fMRI [Sunaert, 2006; Stippich, 2007]. The ul- timate goal of surgical treatment of brain tumours is to completely remove the pathological tissue, while minimising damage to healthy tissue in order to reduce the risk of inducing new (possibly permanent) neurological deficits. It is therefore important that the resection margin does not extend into functionally intact areas.

Mapping of eloquent functional areas has traditionally been performed using

invasive methods, for instance using intra-operative cortical stimulation (patient

awake) or extra-operative mapping by implantation of a subdural grid. These

methods are accurate, but there are some important disadvantages. They require

extra surgery time, or even a separate surgical procedure for the mapping of brain

21 4.2 Tumour Resection

function. In addition, a larger craniotomy than required for the tumour removal may be needed for the mapping. As both morbidity and cost of surgery is related to the time required for the procedure, there is clear benefit both for the patient and the clinic if prior knowledge about functional organisation is available.

FMRI allows non-invasive mapping of eloquent areas to be performed prior to surgery. To date, functional mapping of motor, sensory and language functions are considered to be appropriate for pre-surgical mapping [Bookheimer, 2007;

Stippich, 2007; Berntsen et al., 2008; Pujol et al., 2008]. By combining fMRI results with anatomical reference images, the relationship between lesion mar- gin and functional areas can be established. This information can be used for pre-operative risk assessment, allowing the patient and physician to make an in- formed decision about treatment options. If functionally intact tissue is too close to the desired resection margin, the fMRI results indicate that intra-operative map- ping is required. As the distance between resection margin and functional activity increases, the risk for postoperative neurologic deficits decreases [Yetkin et al., 1998b; H˚aberg et al., 2004].

Some groups have proposed that there is a safe distance between resection margin and eloquent cortex, where surgery can be considered safe [Yetkin et al., 1998b; H˚aberg et al., 2004]. However, defining such a safe distance between ac- tivity and resection margins is problematic for several reasons. For instance, the distance will be influenced by several parameters including the imaging sequence used, analysis options (smoothing) and the statistical threshold used to define ac- tivations. In addition, the safe distance is influenced by the local anatomy; it is likely that a larger distance is needed along the cortex, while a shorter distance may be considered safe if moving across a sulcus.

In conclusion, fMRI can contribute to surgical planning in three ways:

I Risk assessment of post-surgical neurologic deficits I Selecting patients for intra-operative mapping I Planning and guiding the neurosurgical procedure

4.2.1 Task Selection

When using fMRI for pre-surgical mapping, it is natural to choose the activation task on the basis of lesion location and behavioural symptoms. However, since there is variability between anatomy and function even in the healthy brain, it may be beneficial to map the anatomy of various functions [Hirsch et al., 2000].

In addition, the normal relationships between function and anatomy may be dis-

turbed by mass lesions and functional areas may also be relocated in response to

pathology.

Chapter 4. Clinical Applications of Functional MRI 22

Sensory-motor functions have been mapped using a variety of different tasks such as finger tapping, hand clenching, tongue movement, lip movement, foot movement and toe movement. Sensory functions have been examined by rub- bing, stroking or brushing the relevant body part. For mapping of Broca’s area, different forms of word generation tasks have been utilised. Since speaking out loud induces extensive head motion, the patients are in most cases instructed to perform silent word generation. Even so, involuntary tongue movements may be a problem. The word generation tasks include verb generation tasks, picture nam- ing tasks, verbal fluency task or recital tasks [Engstr¨om et al., 2005]. Activity in Wernicke’s area has been induced by tasks that require language comprehension such as semantic or grammatical judgment tasks. Another way of activating Wer- nicke’s area is to let the patient listen to spoken language or read written language.

Mapping of visual areas has been performed by having the patient view flickering checkerboards, flashing lights or light emitting diodes.

4.3 Surgical Treatment of Epilepsy

Especially for patients suffering from pharmaco-resistant temporal lobe epilepsy, surgical treatment offers the possibility of improved seizure control or even cure.

However, surgical planning requires an understanding of the language lateralisa- tion (hemispheric dominance) of the individual patient. The standard method for evaluation of language laterality is the Wada procedure [Wada and Rasmussen, 1958]. In this procedure, an anaesthetic (barbiturate) is delivered to into one of the internal carotid arteries, which effectively knocks out one of the brain hemi- spheres. While one of the hemispheres is anaesthetised, the patient is tested on some cognitive tasks. Usually language and memory functions are evaluated. The patients response to the cognitive tasks is graded and after a period of rest the other hemisphere is tested in the same way. Based on the left and right hemisphere test scores, a laterality index (LI) can be calculated as

LI = Score

− Score

Score

+ Score

. (4.1) The main drawback of the Wada procedure is its invasiveness, associated with significant risks for the patient. Furthermore, the procedure gives no information on the functional organisation within the hemispheres.

It has been suggested that fMRI could provide a non-invasive method for the

evaluation of language lateralisation. The most common procedure when deter-

mining hemispheric language dominance with fMRI uses a combination of differ-

ent language tasks to improve the detect-ability of language related areas [Ram-

sey et al., 2001]. Then the laterality index is calculated as the difference of the

23 4.4 Brain Plasticity

number of activated voxels in the left and right hemisphere respectively divided by the total number of activated voxels, which is equivalent to LI in Eq. (4.1). To further increase the robustness of the laterality estimation, the analysis is usually limited to predefined regions of interest.

Several studies have been performed that compare the outcome of Wada test- ing and fMRI [Desmond et al., 1995; Binder et al., 1996; Yetkin et al., 1998a;

Springer et al., 1999; Rutten et al., 2002b,a; Sabbah et al., 2003; B¨orjesson et al., 2004; Baciu et al., 2005]. In general, good agreement between results from Wada testing and fMRI based LIs was observed. In addition, fMRI provides more details on the functional organisation of the functions tested. This information may be useful for determining the boundaries of resection. These studies have confirmed that fMRI is a useful complement, or even a possible replacement of the Wada test. The reproducibility of the results has been shown to be satisfactory, making it feasible to make clinical decisions based on the fMRI results.

4.3.1 Source Localisation

EEG of epilepsy patients often exhibits spikes or spike wave bursts that are not accompanied by clinical manifestations. Theses spike bursts are nevertheless im- portant for diagnostic purposes. They are caused by abnormal spiking activity in populations of hyper-synchronous neurons. These regions of hyper-synchronous hyperactivity are probably causing the abnormal neuronal activity causing the epileptic seizures. These epileptic sources can be reliably localised by simulta- neously collecting EEG and fMRI data [B`enar et al., 2006; Zijlmans et al., 2007].

By correlating the fMRI data with the EEG time-course, it is possible to iden- tify the source of the ictal sources and thereby provide information for improved surgical planning and procedure.

4.4 Brain Plasticity

The location of various brain functions may change with disease or injury. It

has been thought that the young developing brain has a large potential for plastic

changes. However, an fMRI study of children suffering from hemiplegia chal-

lenges this view, concluding that factors other than age dominate the potential for

adaptive changes in the functional organisation [Holloway et al., 2000]. Several

studies in patients after stroke have confirmed that areas of intact brain tissue are

recruited in the motor cortex ipsilateral to the hand moved. Johansen-Berg and

colleagues combined transcranial magnetic stimulation (TMS) and fMRI to as-

sess whether the greater activity represents functionally important changes or not

[Johansen-Berg et al., 2002]. For patients, TMS over ipsilateral motor cortex sig-

Chapter 4. Clinical Applications of Functional MRI 24

nificantly slowed the movement reaction time. For healthy controls there was no significant effect, thus increased functional activity must contribute to the function of the affected limb.

A very exciting extension of this concept is to study functional re-organisation in response to rehabilitation. Specific brain regions may show altered functional activity correlated with improved clinical results after treatment. The initial pat- tern of functional activity after stroke, and the fMRI activation alterations induced by initial therapy may be used as markers for the potential of functional recovery [Ward et al., 2003b,a; Dobkin et al., 2004]. Thus, fMRI results could be used as a predictor for how much the patient will benefit from rehabilitation training.

Based on the results of fMRI examinations, it may also be possible to tailor the

rehabilitation strategy to be optimal for the individual patient.

5

A IMS

As noted in Chapter 1, Introduction, fMRI research includes a wide range of pro- cedures, as exemplified by the fMRI examination chain. All parts of the fMRI examination chain are important for the outcome of the experiment. The goal of this work has been to increase the success rate of clinical fMRI examinations by providing insights to important issues arising in all parts of the clinical fMRI ex- amination chain. This broad aim has been divided into a few specific questions related to a specific problem or question in the examination chain.

The first question was whether diazepam, a common sedative, influences the outcome of fMRI examinations. This question originates from the observation that, due to claustrophobia or anxiety, some fMRI patients need to be pre-medicated with diazepam to be able to complete the examination. Since the sedation alters the behaviour of the patient, it seems plausible that the results of an fMRI exam- ination may potentially be severely affected, even invalidated, by the administra- tion of a sedative. This issue was investigated in P

I.

The second aim was related to the presentation of visual stimuli. Two different techniques for presentation of visual stimuli are used in fMRI, projection screen and video goggles. The question then arose whether the results of fMRI exami- nations depend on the technique used to present the visual stimuli. This was the topic of P

II.

The third question was related to the acquisition of fMRI data. Since its in-

ception in 1992, two-dimensional (2D) gradient echo echo-planar-imaging (GRE-

EPI) has been the number one imaging sequence for fMRI. With the advent of

parallel imaging it became possible to collect 3D BOLD weighted data at a high

enough rate for fMRI. The question was then if a 3D parallel imaging technique

(PRESTO-SENSE) could offer better fMRI properties than the traditionally used

2D imaging sequences (GRE-EPI). This topic is explored in P

III.

Chapter 5. Aims 26

The fourth research aims was to investigate the analysis methods used for fMRI data. A specific question arose whether spatially adaptive filtering improves the analysis of fMRI data, a subject explored in P

IV.

The fifth aim related to the interpretation of fMRI examination results. Per- formance of different acquisition methods and analysis are usually investigated using some signal processing measures such as image SNR, activation volume or explained variance. It is then of importance to know whether these measures relate to the subjective data quality perceived by the clinician. This subject is investigated in P

III

In summary, the specific aims of this thesis were:

Aim 1 To investigate whether diazepam administration prior to fMRI examina- tions influences the outcome of the fMRI examination.

Aim 2 To investigate whether the results of fMRI examination of visual cortex are affected by the visual stimulus modality.

Aim 3 To investigate whether fMRI using the PRESTO-SENSE imaging sequence can provide better results than GRE-EPI.

Aim 4 To investigate whether spatially adaptive filtering improves the results of the fMRI analysis.

Aim 5 To investigate the conformity of subjective assessment of diagnostic qual-

ity of fMRI results and objective measures of fMRI data quality.

Part II

Methods, Results and Discussions

29

The content of this part of the thesis will closely follow the imaging pipeline described in Chapter 1, Introduction, while addressing the specific research ques- tions stated in Chapter 5, Aims.

The research questions that were declared in Chapter 5, Aims, will be covered

sequentially. For each question, more detailed background to the specific research

question is provided, followed by a short section describing the research methods

and the most important results. Then a discussion based on the research results

and the literature will follow. The backgrounds and discussions presented here

will partially overlap with the presentations in the included papers. For more

detailed descriptions on materials, methods and results, the reader is directed to

the included papers.

6

P ATIENT P REPARATION

This chapter will address a specific problem regarding patient preparation for fMRI examinations, namely the issue stated as Aim 1: are fMRI results valid when an anxiolytic is administered prior to scanning? The text is is based on the background, results and conclusions in P

I. Some extensions to the back- ground material and conclusions in the paper are made, which to some extent are based on related publications not included in the thesis [S¨oderfeldt et al., 2006;

Ragnehed et al., 2007].

6.1 Anxiolytics and fMRI

6.1.1 Background

Many common substances affect the BOLD response in various ways. For in- stance caffeine is well known for increasing the magnitude of the BOLD response [Mulderink et al., 2002], thus offering the patient a cup of coffee prior to scanning is considered to be beneficial. However, others have noticed that the effect of caf- feine on the BOLD signal depend on the level of chronic caffeine intake. Increased BOLD signal has been observed for high users whereas a reduced BOLD magni- tude was observed for low users [Laurienti et al., 2002]. Several studies have tried to explore this effect in more detail, and it has been shown that the BOLD tem- poral dynamics are altered by caffeine administration [Liu et al., 2004; Behzadi and Liu, 2006; Liau et al., 2008]. Other studies have shown that for instance alco- hol and heroin reduce the extent of activation to visual stimuli [Sell et al., 1997;

Levin et al., 1998].

Benzodiazepines, such as diazepam, are a class of psychoactive drugs with

varying hypnotic, sedative, anxiolytic, muscle relaxant and amnesic properties,

Chapter 6. Patient Preparation 32

which are slowing down the central nervous system. Benzodiazepines are useful in treating anxiety, insomnia, agitation, seizures, and muscle spasms. Benzo- diazepines bind to the GABA receptors and increases the affinity of the recep- tor to the GABA neurotransmitter, effectively increasing the inhibitory effect of the available GABA, producing sedatory and anxiolytic effects [McKernan et al., 2000].

In clinical practice, fMRI patients sometimes feel worried or anxious about the examination, sometimes to the extent that they find it difficult or even impos- sible to undergo the examination. In such cases it is common to give the patient a small dose of diazepam to alleviate the patients discomfort. Even though it has not been well known how the fMRI results are affected, diazepam has been routinely administered to worried patients. In a study by Kleinschmidt et al.the effect of diazepam on the baseline BOLD signal was studied [Kleinschmidt et al., 1999]. They found that diazepam administration did not alter the magnitude of the baseline BOLD signal. However, more important is to know what happens to the activation induced BOLD response when diazepam is administered. Some hints may be obtained from studies on other benzodiazepines. For instance it was shown that lorazepam administration attenuated amygdalar and insular activity in emotion processing, whereas activity in the pre-frontal cortex was unaffected [Paulus et al., 2005]. Further, the benzodiazepine midazolam reduced the activity for anticipation of painful stimuli, primarily in the insula [Wise et al., 2007].

Given the anxiolytic and sedative effects of benzodiazepines the down regula- tion of limbic activity seem appropriate. Hence, it appears as if benzodiazepines do not have a significant influence on the BOLD signal in cortical areas. On the other hand, it also known that single doses of diazepam are capable of caus- ing significant decreases in performance, such as longer reaction time, decreased eye-hand coordination, impaired information retrieval and reduced cognitive skills [Gier et al., 1981; Friedman et al., 1992; Kozena et al., 1995; Drummer, 2002].

These are all effects that could influence the outcome of an fMRI examination, es- pecially since clinical fMRI examinations most often consider motor and language retrieval functions [Sunaert, 2006; Stippich, 2007]. If the results of a clinical fMRI examination do not represent the patient’s normal brain function, decisions based on the fMRI results may be erroneous, resulting in sub-optimal treatment. In worst case this could lead to loss of cognitive functions which may severely limit the patient’s quality of life.

6.1.2 Methods

To answer the question whether the results of a clinical fMRI examination would

suffer from administration of diazepam, twenty volunteers were recruited for a

fMRI study. The participants underwent two fMRI examinations and received ei-

33 6.1 Anxiolytics and fMRI

ther a placebo or the active substance (diazepam, 5 mg) about 30 minutes prior to the examinations. The administration of the substances was random, counter- balanced and double-blind. The fMRI sessions included a motor task, a language task and a working memory task. The fMRI data were evaluated to assess sub- stance related differences in activation patterns and activation volumes as well as differences in BOLD dynamics in primary functional areas using ANOVA.

6.1.3 Results

The main findings of the study were

I Significant BOLD activity was found in relevant areas for all tasks. There were no significant differences in the activation patterns between the phar- macological conditions for any of the tasks.

I There were no significant difference in activation volume or lateralisation between the pharmacological conditions for any of the tasks.

I No significant differences in BOLD dynamics between the pharmacological conditions were found.

6.1.4 Discussion

Other fMRI studies on the effects of diazepam on human behaviour have found down-regulation of the functional activity of subcortical areas [Paulus et al., 2005;

Wise et al., 2007]. In the present study, administration of diazepam induced no significant effects on the fMRI results or the BOLD dynamics. However, it should be realised that the BOLD dynamics analysis was restricted to primary functional areas (motor cortex and Broca’s area only), which means that there could be dif- ferences in secondary and supporting areas.

In conclusion, it was demonstrated that a clinically relevant low dose of di- azepam administered prior to fMRI examination had no statistically significant effects on the activation results. It was also shown that there was no effect on the temporal dynamics of the BOLD response [Ragnehed et al., 2007] in pri- mary functional areas. In combination with previous fMRI results, demonstrat- ing that Benzodiazepines primarily affect subcortical areas [Paulus et al., 2005;

Wise et al., 2007], there is considerable evidence supporting that mapping of pri-

mary motor and language functional areas is feasible after the administration of

low doses of diazepam.

7

D ATA A CQUISITION

Two different aspects of the data acquisition will be covered. First the impact of stimulus modality on the outcome of the fMRI experiment is explored. Next a comparison of some important characteristics of different fMRI data acquisition methods is presented. This section is primarily based on the background, results and conclusions in P

II and P

III and specifically concern Aim 2: is mapping of visual cortex affected by stimulus modality, and Aim 3: is PRESTO- SENSE able to improve on EPI results.

7.1 Visual Stimulus Delivery

This section is primarily related to Aim 2 and P

II.

7.1.1 Background

For clinical usage of fMRI, accurate and reproducible results are utterly impor- tant. Decisions based on invalid results could have a huge negative impact on the individual patient. Also, to study rehabilitation processes, brain plasticity and functional changes due to neurological disturbances, the results need to be repro- ducible over time. One limiting factor in this aspect is the stimulus modality. If the applied stimuli varies in an uncontrolled manner between fMRI session it should be expected that the results also vary in unpredictable ways, rendering it difficult or even impossible to make informed interpretations of the results.

Regardless of the specific fMRI task being used, some instructions and/or

stimuli have to be delivered to the patient or subject. The stimuli and instruc-

tions can be of different kinds, instructions can be auditory or visual in the form