Can psychological aspects ibe localized in the brain? – Comparison of PET data with psychological mapping

(1)

Can psychological aspects be localized in the brain?

-

Comparison of PET data with Psychological Mapping

Carl Forsmark

November 14, 2014

(2)

A Master of Science Thesis in Engineering Physics

Supervisors:

Lars Kai Hansen, DTU Claus Svarer, NRU Finn Årup Nielsen, DTU/NRU

Hans Bornefalk, KTH Examiner:

Mats Danielsson, KTH

Department of Informatics and Mathematical Modulation Technical University of Denmark (DTU), Lyngby

and

Neurobiological Research Unit

Rigshospitalet University Hospital, Copenhagen Denmark, 2005

TRITA-FYS 2014:74 ISSN 0280-316X ISRN KTH/FYS/–14:74-SE

(3)

Abstract

This thesis uses the NRU PET and volumetric MR data in combination with psychological test data from 83 healthy volunteers to calculate different methods to find correlations between cerebral regions and psychological traits. Predom- inantly a preprocessing method has been used, where age dependency in both data sets has been removed in different ways. The results after preprocessing were of much less relevance than those from the unprocessed data.

Analys över hjärnans funktionalitet -

Jämförelse mellan PET-data och psykologisk kartläggning

Sammanfattning

Detta examensarbete använder data framtagna vid Rigshospitalet ur PET och strukturell MR-avbildningar av hjärnan, samt data från psykologiska tester, hos 83 friska försökspersoner. Modelleringar görs för att söka finna korrelatio- ner mellan olika regioner i hjärnan och psykologiska karaktärsdrag. Framför allt förbehandlas data genom att söka ta bort åldersberoendet i bägge datagrup- perna. Resultaten efter dessaförbehandlingar visar sig vara mindre relevanta än resultat ur obehandlat data.

(4)

Who, Where, Why, What?

The programming and all work up until the end of 2005 has been done at DTU in Lyngby and at NRU, Rigshospitalet in Copenhagen. The PET imaging and the NEO PI-R tests were done at Righospitalet, the MRI imaging was done at Hvidovre hospital. Delayed parts of the thesis were written mainly at KTH, Stockholm.

Thank you

Lars Kai Hansen, Claus Svarer, Finn Årup Nielsen, Vibe Frøkjær for pushing me through. Erik Lykke Mortensen for the use of his original data set. Marie for coping to share apartments with me for four months. Mette for being my mentor and helping me get back on track after so many years of thesis beauty sleep. All you friends who really believed that I one day would graduate.

(5)

Background and theory

The 18th century pseudoscience phrenology, the art to determine a persons personality by feeling the lumps on his or her head has not been in use for a long time. But neuroscientists today are on the brink of deciding how ones personality works by viewing a different kind of lumps. In this case one looks at the amount of signal substances and neuronic cells in the different parts of the human brain.

Several investigations on the correlation between signal substances, regions of the brain and psychological traits are being undertaken at the moment. Know- ing the location of certain traits, and their specific signal substance dependencies, might let doctors locally deposit pharmaceuticals where they are needed.

This diminishes unwanted side effects that will occur if the pharmaceutical in question is for example deposited in the blood stream, leading to global cerebral effects. One might for example simultaneously drastically increase the amount of serotonin in the frontal cortex, while keeping the amount of the same substance in hippocampus to a minimum.

(8)

1.1. NEO PI-R Background and theory

1.1 NEO PI-R

The Neuroticism Extraversion Openness to experience Personality Inventory Revised questionnaire is a way of determining a persons psychological traits according to the five factor model.

1.1.1 Psychological taxonomy

Many researchers within human personality assessment have for several years used different scales to pinpoint the quintessence of their subjects identities.

However the number of scales has been severely large, and several personality scales were similarly named but measured disparate concepts, while differently named scales measured highly comparable concepts [1].

A taxonomy – a descriptive model – of the psychology subject matter was needed. One approach was to use the natural language of personality description, after the lexical hypothesis: “Those individual differences that are most salient and socially relevant in peoples lives will eventually become encoded in their language; the more likely is it to become ex- pressed as a single word” [2]. In 1936 Allport and Odbert listed all words in an unabridged English dictionary that were expressing distinctions between humans. After completing their list they had about 18,000 different terms, from which they extracted about 4500 adjectives that they thought described relatively permanent observable traits. This list was defined as “generalized and personalized determining tendencies-consistent and stable modes of an individual’s adjustment to his environment”. The other terms were placed in three other categories. One included temporary states, activities and moods; one spanned the subjective judgements of peoples behavior and reputation. The descriptive words in the last category were predominantly those of physical characteristics, capacities and talents, and terms that one could not place in the other categories.

The trait subset mentioned above was later used by Catell in 1943 to create a multidimensional model of personality structure [3]. Due to the limitations of data analysis at the time Catell clustered the 4500 traits into 35 variables. Had he not, the costs and complexity of the factor analysis made would have been too great. The analyses made identified twelve personality factors, to which Catell added four factors he felt were missing. On basis of these sixteen factors Catell constructed the 16PF Personality Questionnaire, a test that is still being used in businesses on potential employees and in university level studies. The version utilized today, however, is very different from the original, as more re- cent research has not been able to duplicate neither the number nor the nature of factors (e.g. Tupes and Christal [4]).

1.1.2 The Big Five

Using simplified descriptions of 22 of Catells variables Fiske managed in 1949 to see factor structures that were quite similar to what would later be known as the Big Five [5]. Tupes and Christals large correlation analysis of eight different samples in 1961 gave the find of “five relatively strong and reccurrent factors and nothing more of any consequence” [4]. Several researchers replicated the five-

(9)

1.1. NEO PI-R Background and theory

factor structure, e.g. Norman [6] in 1963, from whom the factors were initially labeled ([1], p.6):

(I) Extraversion or Surgency (talkative, assertive, energetic) (II) Agreeableness (good-natured, cooperative, trustful) (III) Conscientiousness (orderly, responsible, dependable)

(IV) Emotional Stability vs Neuroticism (calm, not neurotic, not easily upset) (V) Culture (intellectual, polished, independent-minded)

In 1981 the above factors were first called “the Big Five”, named so for their broadness rather than greatness (Goldberg 1981, [7]).

1.1.3 Big Five in NEO PI-R

Costa and MacCrae had begun their work with a questionnaire concerning only the dimensions Neuroticism, Extraversion and Openness to experience by clus- ter analysis of Catells 16PF. In 1983 they realized that they would get a five- factor questionnaire highly resembling the Big Five factors by adding the dimensions Agreeableness and Conscientiousness. Several studies during the eighties demonstrated that their scales were converging quite well with the Big Five, with the exception of Openness to experience which was broader than the culture/intellect factor (V above). The remodeled questionnaire was named the NEO Personality Inventory Revised in 1985, with each major personality trait being spanned by six subtraits - facets - as follows:

• Neuroticism (Anxiety, Angry Hostility, Depression, Self-Consciousness, Impulsiveness, Vulnerability)

• Extraversion (Warmth, Gregariousness, Assertiveness, Activity, Excitement- Seeking, Positive Emotions)

• Openness to experience (Fantasy, Aesthetics, Feelings, Actions, Ideas, Values)

• Agreeableness (Trust, Straightforwardness, Altruism, Compliance, Mod- esty, Tender-mindedness)

• Conscientiousness (Competence, Order, Dutifulness, Achievement-Striving, Self-Discipline, Deliberation)

The NEO PI-R traits are sometimes referred to as “OCEAN”, or “CANOE” for mnemonic reasons.

1.1.4 NEO PI-R in Danish

The Costa-McCrae test was translated to Danish by Skovdal Hansen, Lykke Mortensen and correspondents with knowledge of the English test ([11] p. 17).

The test was then retranslated to English by an English psychology PhD who had lived in Denmark for several years. The translation was emphasized on the

(10)

1.2. Positron emission tomography Background and theory

Danish versions of the items - the “questions” - were meaningful to the superior traits or the facet that the item aims to describe. Items were formulated to maximize the probability that the mean scoring of the item would be approximately like the mean scoring of the American original item. Items which posed problems had several different translations with consideration of the empirical testing of the translation. For the Danish NEO PI-R manual a large number of Danes had done the scoring test. Out of these approximately 600 individuals were twins; the individual scoring of these twins is used in this thesis project, and that data set is for convenience referred to as D600 further on.

1.2 Positron emission tomography

By marking a bioactive molecule with a radioactive atom, and observing the decay of that atom, one can see the density of the bioactive molecule in living tissue. Particularly within neurophysiology this has been used to study functions in the human brain. Most commonly used radioligands are¹¹C,¹³N ,¹⁵O and

18F , with half-lives between two minutes and two hours. After the isotopes are attached to the bioactive molecules, the solution is added to the patients body – mainly through a venous infusion. The patient is placed in a cylinder of detectors. When the radioligand decays the emitted β⁺ travels shortly before hitting an electron, thus annihilating both particle and antiparticle resulting int two 511 keV photons, directed in perpendicular angles relative the annihilated particles trajectories, and going in opposite directions. When two detectors more or less on opposite sides both get readings within 6-12 ns, the system detects this incident event as a so called line of response. More modern PET systems have a higher time resolution (less than 0.5 ns) which use the time difference between the detections to determine from which segment (down to 10 cm) in the line of response the annihilation event occurred. Through image reconstruction algorithms a layer by layer view of the patient is composed, where the density of the radioligand within the tissue is displayed by color gradients.

[¹⁸F ] → [¹⁸O] + β⁺+ ν_e β⁺+ e⁻ → 2γ

Depending on what molecule the isotope is attached to, one can study different occurrences in the body. The fludeoxyglucose[¹⁸F ]-FDG molecule is commonly used when screening for cancer tumors, as cancer cells have a very strong uptake of the sugar molecule. Radioligands are a group of molecules used for research on the neurons and their receptors. In this study [¹⁸F ] altanserin was used.

Altanserin binds to the 5HT2A receptor – one of approximately 15 receptor types normally binding to the neurotransmitter serotonin.

1.3 Magnetic resonance imaging

Nuclear magnetic resonance imaging is used to get a clear structure of the body soft tissue – muscles, body fat, tendons, nerves and blood vessels et. alt. The principle is that atomic nuclei with an odd number of protons or neutrons (approximately 1/3 of all stable isotopes) have a spin, which causes a magnetic

(11)

1.4. Statistics Background and theory

dipole moment. As hydrogen [¹H] is by far the most abundant detectable ele- ment in the body (water, organic molecules et. alt.), MRI apparati are mainly detecting hydrogen excitations. In the scanner a strong magnetic field is in- duced in the patient, causing the hydrogen nuclei to align their spins in the direction of the field, but not entirely parallell. The wobble along the cone shaped path is called the precession motion, and the speed of the precession is called the Larmor frequency, ω. Radio pulses then add energy to the nuclei, causing the precession angles to increase and the precession motion of neighbor- ing nuclei to align in phase. The nuclei will return to their original energy levels in two steps. The spin-lattice relaxation time T₁is the time window after which approximately 63% of the nuclei have returned to their lower energy state, re- turning the precession angles to unexcited levels. The spin-spin relaxation time T2 is the time window after which 63% of the nuclei precession motions have de-aligned. T2 is never larger than T1, and they both differ depending on the tissue. The relaxations are detected and these signals are used to project an image of the examined body.

1.4 Statistics

1.4.1 Data correlation

The main aim of the project was to determine correlations between the serotonin levels in different part of the brain to personality traits. Comparing all individual results gives correlations. However, we don not know if this is merely a causality. To determine this we need to test our results.

1.4.2 Independent component analysis

When receiving a signal made out of overlapping individual, and independent signals, ICA is a computational method to separate the components. This is for example of good use when getting volumetric MRI data. In our project the individual MRI brain volumes have been fitted to the PET scans.

1.4.3 Statistical Parametric Mapping

In functional neuroimaging experiments, with fMRI or (in our case) PET scans, the SPM is a computational method to better compare individual brains. Al- though all human brains in general are very similar, there are always individual differences in topography and morphology. With SPM, one can warp the individual results into a common brain atlas, from which statistical comparisons can be made.

1.4.4 Permutation test

With data that has a normal distribution, there are fairly easy ways to control the correlation between two data sets. In our case however, we do not know the distribution of our data sets (density of cerebral serotonin receptor binding and psychological mapping). For such non-parametric data, one uses a permutation

(12)

1.4. Statistics Background and theory

test to see the validity of the correlations, by comparing a number of permutations of the given data sets. Ideally one looks at all possible permutations – n! for a data set of n elements – but for larger sets one is usually satisfied by watching the results after 10 000 permutations. The test goes as follows:

1. Do parallell tests and group the datasets in groups A and B.

2. iI there seems to be a correlation between A and B then assume null hypothesis H0: there is no correlation.

3. Compute the difference between means (or in our case the correlations) between the datasets. Set this to the zeroth level.

4. Repeat R times; 10 000 ≤ R ≤ n!

• Randomize the order of the data in A and check for correlation between A^∗ and B.

• Save found correlations

5. Correlation data should be evenly distributed around first found correlation value, ie ±0. Count the number of instances Ifwhere the “false”

correlations are greater than in the observed data.

6. The ratio IF/R gives the p-value – the likelihood of H0 being correct, similar to the 1 − 2α confidence interval. A p-value of 0.05 or lower gives a significant falsification of the assumption H0 .

7. if a larger correlation found in randomized value then H0stands.

(13)

Chapter 2

Experimental setup

Eighty-three Danish citizens have undergone the three part examination of their brains. The volunteers were all healthy men and women between the ages of 18 and 79. Each volunteer undertook the NEO PI-R questionnaire on the same day (in general just the hour before) they were PET scanned. The MRI scan was taken at the Hvidovre hospital, within two weeks after the PET scan.

2.1 The psychological questionnaires

The NEO PI-R self-report personality questionnaire consists of 240 questions divided in groups of eight, where each group concerns a specific personality trait, as listed in 1.1.2 (p. 2) above.

The Danish version of NEO PI-R was completed by all subjects on the same day as they were PET-scanned, and the scoring of of the questionnaire then was made according to the Costa-McRae manual [11]. All subjects also completed the Symptom Check-List revised (SCL-90-R) questionnaire to determine possible symptoms of distress and psychopathology with the subject [8].

2.2 The PET scan

On the same day as the subjects took the questionnaire they underwent the PET scanning. Before any radioligand was injected into the patients system blood samples were taken to view trombocytes and plasma. For some of the volunteers DNA sampling was also done. A venous bolus infusion¹ of [¹⁸F ] altanserin solution was injected in the subjects arm.Two hours after the initial burst infusion the amount of [¹⁸F ] altanserin having passed through the blood- brain-barrier had stabilized, and five scans with ten minutes intervals were done.

During that time three plasma controls (HPNC) were made to determine any necessary compensations due to metabolization and decay rate. Each scan was

1A bolus infusion indicates that a strong concentration of solution is injected at first (the bolus from the latin word ball ; followed by a continuous infusion of solution with a lower concentration. This is made to faster reach the requested concentration of the solution in the system.

(14)

2.3. The MRI Experimental setup

35 slices in grids of 128×128. Voxel volume 4.25×2x2mm. The flouride in the altanserin solution has a half-life of approximately two hours, turning into¹⁸O and a β⁺. The [¹⁸F ] altanserin triggers the serotonin receptors in the brain, targeting one of ca fifteen proteins, the 5HT2A.

3.7 Mbq dosage was injected per kg body weight in each patient.

Figure 2.1: An ideal graphic description of the bolus injection and the amount of [¹⁸F ] altanserin in the bloodstream of a 60 kg person.

Blood samples taken during scanning is purified to get an as clean [¹⁸F ] altanserin rest product as possible, the ’de-hemofied’ sample is then injected in a high density chromatograph to get the spectral image of the leftovers of the ligand and its derivatives.

2.3 The MRI

The brain structures were imagined through magnetization prepared rapid gradi- ent echo (MPRAGE) sequences aquired on either a 1.5-T Siemens Vision scanner (for 68 of the subjects) or a 3-T Siemens Trio scanner (for the remaining 15 subjects).

• T1 with voxel volume 1.2x1.2x1.1 mm gives a perfect structural scan

• T2 with voxel volumes 12.2x1.2x3 mm

• PD (proton density) with voxel volumes 12.2x1.2x3 mm

(15)

2.4. Quantification of receptor binding Experimental setup

Figure 2.2: The Rigshospitalet PET about to scan.

2.4 Quantification of receptor binding

In steady state, the binding potential of specific tracer binding (BP 1) is defined as:

BP 1 = CROI− Cref

Cplasma

= f1∗ Bmax

Kd

(mLmL⁻¹) (2.1)

where CROIand Cref are mean counts in the regions of interest and the reference volume (cerebellum); while the Cplasmarefers to the radio emission from plasma parent compounds; f1 stands for the free fraction of radio tracer; Bmax refers to the density of receptor sites that bind with the tracer; Kd is the tracer-to- receptor affinity constant.

2.5 Co-registration

The PET and MRI data are aligned according to [12], to make it possible to determine how much [¹⁸F ] altanserin is deposited throughout the different parts of the brain, an example of this can be seen in figure 2.3. The data was extracted in regions of interest – ROI – and normalized with the amount of [¹⁸F ] altanserin distributed in the cerebellum of each patient, as the cerebellum basically only handles motor functions of the body, thus representing the average non-specific binding of the brain. The only radiation emitted from the cerebellum comes entirely from still venous [¹⁸F ] altanserin . The regions in which the brain

(16)

2.5. Co-registration Experimental setup

is separated can be seen in appendix A, p.30. A .csv-file is created where each subjects PET and NEO PI-R data are placed according to cerebral region resp. psychological trait ². Apart from the regions of interest the data sheet also contains the [¹⁸F ] altanserin deposition in cerebrospinal fluid (CSF), white matter binding, non-specific binding calculated from the cerebellar binding and mean count density in the gray matter that does not belong to any region.

Figure 2.3: A 2D chemical view on the [¹⁸F ] altanserin molecule, and a 3D illus- tration of the distribution of [¹⁸F ] altanserin receptors in the brain, displayed by co-registration of MRI and PET images, as described in 2.5.

2In the .csv-file the SCL-data was also submitted, but I am not using it in my research.

(17)

Chapter 3

Methods

3.1 What seems to be the problem?

When looking for cross correlations between data sets that are not intrinsically related, the strength of the significance is of great importance. In this analysis the age and sex of the test subjects are nuisances that might very well cause errors in the results. Hoping for better results in the way of correlations between cerebral regions, serotonin receptor density and psychological traits, one might attempt to remove the above mentioned nuisances. One must then check the validity of the new approximations.

3.2 How will you fix it?

3.2.1 Control for dependency on sex (gender)

A control to see whether the difference between male and female subjects is merely by chance or a fact to recon with in further calculations and experiments was made. The NRU subject data was separated into a male and a female group, and the separated mean values over each individual region and trait were calculated.

r =| 1 −M₁ M2

| (3.1)

Small differences were found (see table 4.1, p.15), but there was no common fact like one sex having less serotonin than the other, but variations in distinct parts of the brain. Nor were any gargantuan differences found in the psychological data, all in accordance with several papers, e.g. Costa-Terraciano-McRae [15].

Viewing the correlation coefficient calculations with only age, resp. age and sex as nuisance showed merely small differences between the two. This is all according to prior examinations done by e.g. Adams et.al [13]. As the differences relative to the subjects sex vary, it is difficult to assume some underlying mathematical function to make approximations to remove this nuisance, which is why I will not make any such attempts.

(18)

3.2. How will you fix it? Methods

3.2.2 Preprocessing with respect to age dependency

An attempt to decorrelate all data with reference to age before calculating the region-trait correlations was also done. This was made in two fashions: one by removing the time dependent variable by division, the other by subtraction.

For the PET data I initially used information from Fazio/Perani [14], whereas I tried calculating the decrease in 5HT2A as 20% per decade starting at 30 years. A remarkable difference from non-decorrelated data was seen. However, according to Adams one can only see such a large decrease in the age group of 35-65 [13], while in general the decrease is not larger than 6% per decade.

By viewing the brain region-by-region I’ve tried to see whether a difference in regional decrease can be found. The goal was to find decent exponential approximations of the data. Due to the small signal-to-noise ratio that proved to be a cumbersome project. Especially to find an approximate age where decrease can be assumed to begin. By stepping through all ages between 18 and 40 one can receive a specific age for each ROI at which the decrease can be approximated to start. The calculations gives regional decrease-initiations at ages ranging between 24 and 40, with a mean age of 29.3 and a median age of 27. For some of the simulated start ages of serotonin decrease, for certain ROIs the approximation functions yielded are increasing rather than decreasing. As all exponents, positive or negative, are small, the largest 0.03, I have chosen to neglect this increase, and instead treat these regions as age independent. I have chosen to include the cerebrospinal fluid and the white matter counts in my calculations, as they both show up as “regions” highly correlating with traits, before any age decorrelation has been made. Since it would be absurd to assume the decrease of the serotonin levels to start at different ages in different parts of one persons brain, I simply chose to, in further calculations, assume the starting age to be 24.

BP 1(t) = A exp(−λt)

⇒

BP 1^∗_subtr ≈ BP 1(t) − ˜A exp(˜λt) [t ≤ 24] (3.2) BP 1^∗_div ≈ BP 1(t) ∗ exp(˜λt) [t ≤ 24] (3.3)

Table 3.1: Assuming the time dependence of the tracer binding potential BP 1, we use given data to calculate approximated values ˜A and ˜λ, which we then use when we calculate our time independent values BP 1^∗.

The NEO PI-R data set is also fluctuating by age. A number of studies suggest that neuroticism, extraversion and openness traits and facets tend to decline with age, while agreeableness and conscientiousness tend to increase, especially up to around 40 years of age ([16], [17]). By using the prior data acquired by Skovdal Hansen and Lykke Mortensen (D600) to calculate the approximate age dep. functions and comparing with the same calculated from our data we decorrelate our data set for age. I have compared approximations made with the D600 data and our own data (NRU) and see that for most of the traits the difference between the two is below 10%. The largest difference is 18% for vulnerability (n_vulner). Since the D600 data set only goes up to 67 years, with three subjects aged 61 and older, while our set has nineteen subjects

(19)

3.2. How will you fix it? Methods

between the ages 61 and 79, I made comparisons between approximations of the D600 and NRU data sets, and found that for the latter set, three of the five largest differences decreased, while two increased. The smallest difference grew immensely, a reason for which we henceforth will view data from our full age span, ie the D600 and our own 83 subjects together. See figure 4.2, p. 17.

In the D600 data approximations of linear, quadratic and exponential kind are all made for each facet and the one that best fits the data is chosen as the age decorrelator function. In these calculations no assumptions of when any changes start taking place, so the starting age is set to birth. The time dependent functions are then used to decorrelate all individual data as seen below. As each trait, dimension, is in fact the sum of its facets, I have chosen to make the age decorrelations on these sums, rather than summing the decorrelated values.

This to minimize the error progress.

N EO(t) = kt + m

N EO_subtr^∗ ≈ N EO(t) − ˜kt (3.4)

N EO^∗_div ≈ N EO(t) ∗ ˜m

˜kt + ˜m (3.5)

N EO(t) = at²+ bt + c

N EO_subtr^∗ ≈ N EO(t) − ˜at²− ˜bt (3.6) N EO^∗_div ≈ N EO(t) ∗ ˜c

˜

at²+ ˜bt + ˜c (3.7)

N EO(t) = B exp(−ωt)

N EO_subtr^∗ ≈ N EO(t) − ˜B exp(˜ωt) (3.8) N EO^∗_div ≈ N EO(t) ∗ exp(˜ωt) (3.9)

Table 3.2: Assuming different time dependencies for our traits, we calculate approximations of those functions using our data sets (D600, NRU or a combination of both). We then use these approximated constants to calculate time independent values N EO^∗.

For the subtraction approximations calculations are made using all three approximation types. For the division approximations however focus is set on calculations based on age decorrelations by exponential function alone. Athough each individual trait probably has a rather intricate age dependency, I would assume some combination of a high frequency sinusoidal and a sigmoid forma- tion rather than linear, quadratic or exponential, I’ve chosen to focus on the exponential function because it from a biological point of view makes more sense that something has such a dependency, rather than linear or quadratic. I choose this even though the square polynomial function for most facets seemingly fitted better against the square function (where the sum of squared errors were smallest).

(20)

3.3. Calculating significance values Methods

3.3 Calculating significance values

By using the PET and NEO PI-R approximations made above we get a new data set that can be seen as age independent; which in turns will give us a new correlation coefficient matrix with new p-values as follows in the ensuing chapter. The region-factor cross-correlation values calculated all need to be put in a context, namely that of how relevant they are. The stronger significance levels a correlation displays, the more likely is the correlation to exist as more than a statistical flimsy. The significance value, the P -value, should be no larger than 0.05 (meaning that there is a mere 5% chance that the correlation is faulty).

The uncorrected significance value, Puc, is calculated directly by a student t distribution calculation; assuming the correlation coefficients viewed have a Gaus- sian distribution. The lower tail area derived from the student t distribution is then transformed to a two-tailed area, which is our uncorrected p-value.

The corrected significance value, Pcor, is obtained through a permutation test.

In these permutations, ten thousand runs, the individual data from the PET scans were shuffled between test subjects, while the psychological data were left untouched.

P_cor= 1 n

n

X

k=1

(l_k) where lk = 0 if r^∗(k) ≤ r 1 if r^∗(k) > r

From the false correlation coefficients received, the largest one of each run was extracted. After all runs were done, one could compare each “real” correlation coefficient with these false. The corrected P -value that is extracted is the fraction of the number of false corr.coefs. that has a larger absolute value than the viewed real corr.coef. The resulting data is found in tables 4.9 and 4.10, pages 22-23.

(21)

Chapter 4

Results

4.1 Differences between the sexes

Comparisons between sexes, rest of mean values.

rbrain Region rNEO T rait

0.0083 r_orbfrc 0.0019 a_agreea 0.0104 l_medinf 0.0039 a_altrui 0.0126 l_suptem 0.0055 a_tender 0.0127 l_entorhin 0.0065 o_values 0.0282 l_senmot 0.0083 a_trustp 0.0304 r_postci 0.0153 c_dutifu 0.0325 l_occ 0.0167 o_openne 0.0433 l_parc 0.0172 c_orderp 0.0474 l_orbfrc 0.0196 a_modest

0.0568 l_th 0.0197 e_assert

0.1195 r_antcin 0.0925 c_delibe 0.1403 r_ins 0.0953 n_self_c 0.1476 l_entorh 0.0980 o_feelin

0.1556 r_th 0.1145 o_aesthe

0.1608 r_cau 0.1250 n_anxiet 0.1699 l_ins 0.1414 n_neurot 0.4569 l_cau 0.1465 e_excite 0.5072 midbrain 0.1805 n_angryh 0.5325 l_put 0.1851 n_depres 0.8667 l_hippoc 0.2218 n_vulner

Table 4.1: By the ascending order we can see what brain regions (based on the serotonin triggers) and psychological traits seem to behave differently between the sexes.The rest values r_brain and, r_NEO are calculated from r =| 1 − (M1/M2) |. Only the ten smallest and ten largest means are displayed.

(22)

4.2. Preprocessing data Results

4.2 Preprocessing data

To decorrelate our data from age dependency, exponential function approximations are made for both the PET regional data, (table 4.2) and for the NEO PI-R psychological trait data [11]. The latter is tried with three different approaches:

calculations from the D600 data alone (approximation from 600 data points);

from the NRU data alone (approximation from 83 data points, both tables on page 18); and from the sum of both data sets (table 4.7, approximation from all 683 data points).

I have made controls for some NEO PI-R traits to see the difference in the approximation functions from using our own data, the NEO PI-R Danish Manual data and a combination of the two in calculating the exponential approximation functions (table 4.4).

PET approximation functions BP 1 = A ∗ exp(−λ ∗ t)

A λ region A λ region

3.0981 0.0067 l_orbfrc 3.0708 0.0063 r_orbfrc 3.2573 0.0023 l_medinf 3.3266 0.0021 r_medinf 2.6063 0.0048 l_antcin 2.7027 0.0044 r_antcin 0.7423 0.0025 l_th 0.6356 -0.0011 r_th 2.5366 0.0064 l_ins 2.3362 0.0047 r_ins 0.5681 -0.0047 l_cau 0.4491 -0.0073 r_cau 0.6017 -0.0048 l_put 0.5197 -0.0067 r_put 2.8182 0.0021 l_suptem 2.9495 0.0024 r_suptem 3.3489 0.0009 l_parc 3.3886 0.0011 r_parc 3.0064 0.0049 l_medin2 3.1591 0.0051 r_medin2 3.5497 0.0042 l_supfc 3.5919 0.0042 r_supfc 3.0050 0.0009 l_occ 2.9141 0.0012 r_occ 2.4796 -0.0024 l_senmot 2.5702 -0.0020 r_senmot 2.8652 0.0044 l_postci 2.8136 0.0035 r_postci 2.1963 0.0217 l_entorh 1.6894 0.0177 r_entorh 1.5554 0.0281 l_hippoc 1.1517 0.0204 r_hippoc 0.9339 0.0185 midbrain -0.8066 -0.0000 wm -2.1242 -0.0000 csf 1.2491 -0.0121 non_spec

2.3354 0.0049 gm_no_re

Table 4.2: When assuming that the amount of serotonin sensitive neuronic cells are starting to change at the age of 24, an approximative function of the age dependency is here shown. Nota bene that some exponent values are negative, yielding in an augmenting function. These are emphasized above, and those regions are in the age decorrelation seen as age independent. Note that the cerebro-spinal fluid and the white matter binding both get negative constants in their approximations, which of course is impossible in real life.

(23)

Figure 4.1: This plot shows the difference between the found deposition of [¹⁸F ] altanserin (equivalent to the binding potential) in subjects and the theoretical one calculated by decorrelation of age. Data from all of brain in each subject has been summarized

Figure 4.2: Scatterplots of results from (exponential) approximations, D600 (on the x axes, called “manual” and “original”) vs NRU (on the y axes, called “own”

and “Vibes”), comparing a set using all 83 individuals and only those 67 years old and younger.

(24)

Exponential approximation data, assuming N EO(t) = B exp(−ω ∗ t)

D600 data NRU data

Positive exponent value

B ω trait B ω trait

82.4497 0.0017 n_neurot 79.1901 0.0027 n_neurot 14.1527 0.0010 n_anxiet 11.3195 0.0009 n_anxiet 12.4553 0.0033 n_angryh 12.9100 0.0054 n_angryh 13.4084 0.0017 n_depres 21.6439 0.0065 n_impuls 14.5743 0.0009 n_self_c 10.1123 0.0031 n_vulner 18.7521 0.0041 n_impuls 66.3810 0.0022 N_adj(ang.h) 70.0212 0.0015 N_adj(ang.h) 44.9415 0.0005 N_adj(ah_imp) 51.3402 0.0006 N_adj(ah_imp) 142.4906 0.0042 e_extrav 131.8886 0.0035 e_extrav 24.3532 0.0010 e_warmth

22.8966 0.0003 e_warmth 23.7685 0.0035 e_gregar 22.8965 0.0040 e_gregar 18.9087 0.0029 e_assert 17.1231 0.0030 e_assert 23.0880 0.0033 e_activi 22.6115 0.0016 e_activi 25.7458 0.0096 e_excite 24.2767 0.0113 e_excite 27.6684 0.0060 e_positi 23.2919 0.0028 e_positi 135.7820 0.0041 o_openne 116.7676 0.0029 o_openne 25.1538 0.0098 o_fantas

22.1241 0.0096 o_fantas 18.9991 0.0013 o_aesthe 21.9425 0.0027 o_feelin 23.8291 0.0039 o_feelin 17.7965 0.0019 o_action 21.9535 0.0040 o_action 20.2207 0.0050 o_ideasp 24.0617 0.0052 o_ideasp 20.7741 0.0007 o_values 22.8611 0.0018 o_values 23.2196 0.0005 a_altrui 22.9974 0.0002 a_altrui 21.7425 0.0002 c_compet 22.5165 0.0002 c_compet 19.7503 0.0015 c_achiev

Negative exponent value

9.2508 -0.0017 n_vulner 10.2838 -0.0011 n_depres 14.8731 -0.0020 o_aesthe 13.3120 -0.0002 n_self_c 111.4888 -0.0030 a_agreea 112.0597 -0.0024 a_agreea 20.3431 -0.0023 a_trustp 22.8016 -0.0003 a_trustp 18.1109 -0.0048 a_straig 18.3581 -0.0033 a_straig 15.4468 -0.0040 a_compli 14.3684 -0.0041 a_compli 15.3572 -0.0062 a_modest 14.5923 -0.0060 a_modest 19.4430 -0.0016 a_tender 19.3994 -0.0017 a_tender 113.0756 -0.0016 c_consci 106.4692 -0.0024 c_consci

18.0803 -0.0004 c_orderp 14.2878 -0.0042 c_orderp 20.1435 -0.0040 c_dutifu 19.6458 -0.0036 c_dutifu 19.1902 -0.0025 c_self_d 17.8198 -0.0006 c_achiev 14.5801 -0.0044 c_delibe 18.8899 -0.0023 c_self_d 13.5709 -0.0041 c_delibe

Table 4.3: The NEO PI-R exponential approximation functions evaluated from the D600 (left) and NRU (right) data sets.

(25)

Some comparisons of NEO PI-R constants and exponents n_neurot c_compet n_depres o_aesthe c_orderp n_vulner BD600 82.4497 21.7425 13.4084 14.8731 18.0803 9.2508 ωD600 0.0017 0.0002 0.0017 -0.0020 -0.0004 -0.0017 BN RU 79.1901 22.5165 10.2838 18.9991 14.2878 10.1123 ωN RU 0.0027 0.0002 -0.0011 0.0013 -0.0042 0.0031 B_combo 82.8905 21.7459 13.1077 15.3728 17.4986 9.5948

ω_combo 0.0021 0.0001 0.0015 -0.0015 -0.0011 -0.0004

Table 4.4: Comparison of NEO PI-R approximations. The first group is derived from the Danish Manual data, the second from our own experimental data and the third from the combination of both data sets. Traits are chosen from the relative size of the constants, see table 4.5.

The mean percentage difference between approximations

100∗ | 1 − (d_D600/d_{N RU}) |

value trait value trait

0.1887 c_compet 3.0773 c_consci 0.2350 e_assert 3.8175 n_neurot 0.4774 a_compli 4.6028 o_feelin 0.4788 a_tender 4.6343 n_self_c 0.5007 N_adj(ah,imp) 4.7274 o_values 0.6739 o_fantas 5.2641 o_openne 0.7219 n_anxiet 6.0471 a_straig 0.8263 c_self_d 6.6443 e_activi 0.8524 a_modest 7.3293 e_excite 0.9488 o_ideasp 7.9427 a_trustp 1.0332 c_delibe 8.2383 o_action 1.4249 a_altrui 8.3857 n_angryh 1.5035 c_dutifu 9.6412 c_achiev 2.0040 e_gregar 9.6632 n_impuls 2.2437 sum of traits 12.6320 e_positi 2.4789 a_agreea 12.7910 n_depres 2.6920 e_warmth 12.8828 o_aesthe 3.0165 N-adj(angr.h) 17.3819 c_orderp 3.0443 e_extrav 17.9099 n_vulner

Table 4.5: Mean difference between NEO PI-R approximations from D600 data and from our own NRU data.

(26)

Comparing groups relative to ages.

All ages Up to 67 Trait

3.8175 6.6275 n_neurot

0.1887 6.2304 c_compet

12.7910 1.8085 n_depres 12.8828 3.2004 o_aesthe 17.3819 23.4011 c_orderp 17.9099 27.2311 n_vulner

Table 4.6: Mean percentage difference between NEO PI-R approximations from the D600 data vs the NRU data, viewing all subjects and those 67 years old and younger.

Exponential approximation data, N EO(t) = B exp(−ω ∗ t) NEO PI-R all

Positive exponent value Negative exponent value

82.8905 0.0021 n_neurot 9.5948 -0.0004 n_vulner 14.0901 0.0015 n_anxiet 15.3728 -0.0015 o_aesthe 12.6288 0.0038 n_angryh 112.1237 -0.0028 a_agreea 13.1077 0.0015 n_depres 20.7191 -0.0019 a_trustp 14.4392 0.0008 n_self_c 18.3817 -0.0043 a_straig 19.1606 0.0045 n_impuls 15.4553 -0.0038 a_compli 70.2958 0.0018 N_adj(ang.h) 15.4253 -0.0059 a_modest 51.2210 0.0009 N_adj(ah_imp) 19.4231 -0.0016 a_tender 132.9252 0.0035 e_extrav 112.3622 -0.0017 c_consci

23.0405 0.0004 e_warmth 17.4986 -0.0011 c_orderp 22.8577 0.0037 e_gregar 20.2259 -0.0037 c_dutifu 17.1767 0.0027 e_assert 19.2270 -0.0023 c_self_d 22.8795 0.0021 e_activi 14.6410 -0.0040 c_delibe 24.2134 0.0108 e_excite

23.9600 0.0034 e_positi 118.4887 0.0029 o_openne

22.3501 0.0095 o_fantas 22.1704 0.0029 o_feelin 18.2007 0.0021 o_action 20.4238 0.0047 o_ideasp 21.0077 0.0008 o_values 23.1543 0.0004 a_altrui 23.1543 0.0004 a_altrui 21.7459 0.0001 c_compet 19.3919 0.0011 c_achiev

Table 4.7: The exponential approximation functions evaluated from the D600 and our own data combined

(27)

4.3. Results after preprocessing Results

4.3 Results after preprocessing

In the correlation comparison tables (pages 21 - 23 below), I have omitted data from areas of the brain where no receptors are situated, but where there statistically speaking still seem to be some sort of (in some places strong) correlation.

These are the white matter binding (wm), the cerebrospinal fluid csf, and the non-specific binding (non_spec, calculated from PET emissions in the cerebellum). For the subtraction approximations examples of all three approximation varieties are shown. For the division approximations, those correlations viewed where there are no comparisons between age adjusted or untampered data (de- noted as a, b, c and d), the missing correlation value was not large enough to make it to the top ±20 of the viewed data group, and therefore omitted.

The strongest correlations found among the ±20 in all four data groups are emphasized in the tables. They are the positive correlation of vulnerability in the left posterior cingulate gyrus and the negative correlation of self discipline in the same area, which is placed more or less in the middle of the brain. However, the corrected P-values in all groups (including the untampered experiment data) is above 0.05, indicating that the significance of the correlation is low. In fact, even the smallest corrected P-value in all calculations based on the different kinds of approximations is above 0.1, thus statistically irrelevant.

Correlations after subtraction approximation positive correlation negative correlation l_postci – n_vulner l_hippoc – e_positi

r Puc Pcor r Puc Pcor

lin

NRU 0.358 0.0009 0.3069 -0.343 0.0015 0.4302 D600 0.343 0.0015 0.4378 -0.308^* 0.0046 0.7660 comb. 0.348 0.0012 0.3825 -0.316^* 0.0036 0.6931

quad

NRU 0.286^* 0.0087 0.9196 -0.353 0.0010 0.3440 D600 0.265^* 0.0153 0.9813 -0.306^* 0.0048 0.7754 comb. 0.325 0.0027 0.6082 -0.329^* 0.0024 0.5688

exp

NRU 0.354 0.0010 0.3430 -0.378 0.0004 0.1822 D600 0.346 0.0014 0.4158 -0.322^* 0.0030 0.6396 comb. 0.349 0.0012 0.3811 -0.334 0.0020 0.5169

Table 4.8: Examples of correlation comparisons of three different approximation methods with the time dependent part subtracted from measurement data. The displayed correlations were the strongest positive/negative in eleven of eighteen cases. For the other seven cases (marked with ^(*) ) they were still among the top seven, and the corrected P values of the strongest correlations in those cases were all larger than 0.2, and thus of low interest.

(28)

4.3.ResultsafterpreprocessingResults The correlation coefficients with P-values of the top five positive and negative correlations.

Strongest correlations from experiment data, with comparisons.

experiment data D600 approx. NRU approx. combined approx.

Region Trait r Puc Pcor r Puc Pcor r Puc Pcor r Puc Pcor

r_entorh o_fantas 0.412 0.0001 0.0518 a a a a a a a a a

l_hippoc o_fantas 0.398 0.0002 0.0870 a a a a a a a a a

r_entorh n_impuls 0.389 0.0003 0.1204 a a a a a a a a a

l_postci n_vulner 0.388 0.0003 0.1218 0.346 0.0014 0.4162 0.354 0.0010 0.3396 0.349 0.0012 0.3770

r_hippoc o_fantas 0.371 0.0006 0.2091 a a a a a a a a a

r_entorh c_orderp -0.439 0.0000 0.0192 -0.267 0.0147 0.9790 a a a -0.265 0.0156 0.9838

r_entorh c_consci -0.415 0.0001 0.0488 -0.281 0.0101 0.9394 -0.272 0.0128 0.9664 -0.280 0.0103 0.9392

l_entorh c_dutifu -0.390 0.0003 0.1154 a a a a a a a a a

r_entorh c_self_d -0.377 0.0004 0.1719 -0.289 0.0081 0.9054 -0.291 0.0077 0.8887 -0.290 0.0078 0.8915

r_hippoc c_dutifu -0.370 0.0006 0.2120 a a a a a a a a a

Strongest correlations from D600 approximations, with comparisons

Region Trait r P_uc P_cor r P_uc P_cor r P_uc P_cor r P_uc P_cor

l_postci n_vulner 0.388 0.0003 0.1218 0.346 0.0014 0.4162 0.354 0.0010 0.3396 0.349 0.0012 0.3770 l_postci N_adj(ang,imp) b b b 0.306 0.0049 0.7897 0.306 0.0049 0.7791 0.307 0.0048 0. 7740 l_postci N_adj(angr.h.) 0.333 0.0021 0.5123 0.293 0.0073 0.8846 0.294 0.0069 0.8680 0.294 0.0071 0.8716 l_postci n_neurot 0.337 0.0018 0.4698 0.292 0.0074 0.8876 0.295 0.0068 0.8659 0.293 0.0071 0.8734

r_cau e_excite b b b 0.292 0.0074 0.8879 0.266 0.0150 0.9798 0.285 0.0091 0.9208

r_postci c_self_d -0.360 0.0008 0.2739 -0.328 0.0024 0.5840 -0.328 0.0024 0.5722 -0.328 0.0024 0.5776

r_hippoc e_positi b b b -0.328 0.0025 0.5852 -0.350 0.0012 0.3696 -0.334 0.0020 0.5224

l_hippoc e_positi b b b -0.322 0.0030 0.6451 -0.378 0.0004 0.1763 -0.334 0.0020 0.5208

l_postci c_self _d -0.350 0.0012 0.3531 -0.317 0.0035 0.6934 -0.317 0.0036 0.6851 -0.317 0.0035 0.6864

r_orbfrc e_assert b b b -0.316 0.0036 0.7029 -0.316 0.0036 0.6918 -0.316 0.0036 0.6938

Table 4.9: The partial correlation coefficient r calculated for the input data, no nuisance withdrawn; and from the D600 data – where the nuisance of age has been removed by approximations using data from Erik Lykkes danish NEO PI-R manual. P stands for the uncorrected

22

(29)

4.3.ResultsafterpreprocessingResults The correlation coefficients with P-values of the top five positive and negative correlations.

Strongest correlations from NRU approximations, with comparisons

l_postci n_vulner 0.388 0.0003 0.1218 0.346 0.0014 0.4162 0.354 0.0010 0.3396 0.349 0.0012 0.3770 l_postci N_adj(ang,imp) c c c 0.346 0.0014 0.4162 0.306 0.0049 0.7791 0.307 0.0048 0.7740 l_postci n_neurot 0.337 0.0018 0.4698 0.306 0.0049 0.7897 0.295 0.0068 0.8659 0.293 0.0071 0.8734 l_postci N_adj(angr.h.) 0.333 0.0021 0.5123 0.293 0.0073 0.8846 0.294 0.0069 0.8680 0.294 0.0071 0.8716 l_supf c n_vulner 0.326 0.0026 0.5737 0.286 0.0089 0.9203 0.294 0.0070 0.8714 0.289 0.0081 0.8992

l_hippoc e_positi c c c -0.322 0.0030 0.6451 -0.378 0.0004 0.1763 -0.334 0.0020 0.5208

r_hippoc e_positi c c c -0.328 0.0025 0.5852 -0.350 0.0012 0.3696 -0.334 0.0020 0.5224

r_postci c_self_d -0.360 0.0008 0.2739 -0.0.328 0.0024 0.5840 -0.328 0.0024 0.5722 -0.328 0.0024 0.5776 l_postci c_self _d -0.350 0.0012 0.3531 -0.317 0.0035 0.6934 -0.317 0.0036 0.6851 -0.317 0.0035 0.6864

r_orbfrc e_assert c c c -0.316 0.0036 0.7029 -0.316 0.0036 0.6918 -0.316 0.0036 0.6938

Strongest correlations from combined approximations, with comparisons

l_postci n_vulner 0.388 0.0003 0.1218 0.346 0.0014 0.4162 0.354 0.0010 0.3396 0.349 0.0012 0.3770 l_postci N_adj(ang,imp) d d d 0.346 0.0014 0.4162 0.306 0.0049 0.7791 0.307 0.0048 0.7740 l_postci N_adj(angr.h.) 0.333 0.0021 0.5123 0.293 0.0073 0.8846 0.294 0.0069 0.8680 0.294 0.0071 0.8716 l_postci n_neurot 0.337 0.0018 0.4698 0.306 0.0049 0.7897 0.295 0.0068 0.8659 0.293 0.0071 0.8734

r_orbfrc n_vulner 0.335 0.0020 0.4877 d d d d d d 0.289 0.0080 0.8958

l_hippoc e_positi d d d -0.322 0.0030 0.6451 -0.378 0.0004 0.1763 -0.334 0.0020 0.5208

r_hippoc e_positi d d d -0.328 0.0025 0.5852 -0.350 0.0012 0.3696 -0.334 0.0020 0.5224

r_postci c_self_d -0.360 0.0008 0.2739 -0.0.328 0.0024 0.5840 -0.328 0.0024 0.5722 -0.328 0.0024 0.5776 l_postci c_self _d -0.350 0.0012 0.3531 -0.317 0.0035 0.6934 -0.317 0.0036 0.6851 -0.317 0.0035 0.6864

r_orbfrc e_assert d d d -0.316 0.0036 0.7029 -0.316 0.0036 0.6918 -0.316 0.0036 0.6938

Table 4.10: The top five positive and negative partial correlation coefficients r from the NRU data – where the nuisance of age has been removed by approximations using Vibe Frøkjærs research data; also top five positive and negative partial correlation coefficients from the combined data – approximations using both the D600 and the NRU data sets. P stands for the uncorrected significance value, P stands for the

23

(30)

Figure 4.3: Openness and Extraversion scores, with three (division) decorrelations. “Own decor” is decorrelation from the NRU data.

(31)

Figure 4.4: Agreeableness and Conscientiousness scores, with three (division) decorrelations.

(32)

Figure 4.5: Neuroticism and the two adjusted N data scores, with three (division) decorrelations.

(33)

Chapter 5

Discussion, conclusion, recommendations

5.1 Discussion

According to equation 2.1 (p.9), the binding potential of a specific region is calculated partially by subtracting the mean count from the cerebellum in the same brain. This sometimes lead to a resulting negative value of the binding potential in certain regions; in our examination these regions were the right caudates, the midbrain, left entorhinal cortex and left and right hippocampus.

The cerebrospinal fluid and the white matter, being completely free of neuronic cells, both get negative potentials in all subjects, the largest value found as - 4.6589. This is of course physiologically impossible, as it would imply some sort of anti-neurons; or – given that the BP1 is calculated from β⁺ emissions from the radioligands – nuclear reactions from the absorption of positrons.

It is plausible that the age dependencies of serotonin vary throughout the brain, as my preprocessing calculations have shown (p. 12). One might improve the correlations and diminish the p-values, thus also diminishing type II errors in hypothesis testing, by letting the age dependency in each ROI kick in at different ages. For certain regions there is probably no age dependency at all. Further, in certain areas of the brain the ration between white and gray matter changes a lot with age, possibly producing errors in the SPM fitting where older brains are normalized to look like younger.

By decorrelating for age dependencies before calculating the correlation between regions and traits one has removed a large fluctuation in the data. Plausibly however, one has also added an equally large error as the calculation of the age dependency functions – be it linear, quadratic or exponential – is very difficult to get totally correct. Prior researches have stated very different declines relative to age.

Using the D600 data in fitting our psychological scores to age-independent values gives a possible bias. In the fact that the D600 subjects all were twins, the test took place in the autumn in Århus while our took place in the spring in Copenhagen and probably a number of other differences one can assume that parts of the score result differences came out of there. On the other hand, using

(34)

5.2. Conclusion Discussion, conclusion, recommendations

our own test subjects data for the fitting risks an even higher bias, as any error in our input propagates through the approximation function, which when used with the same input risks causing an even larger output error.

The above mentioned problems are most certainly causes of the low significance found in the trait-binding potential correlations made after the age dependency removal (see table 4.9). Thus, the calculated correlations are most likely false.

It can be noted that, when calculating the correlations by making linear or quadratic approximations of the psychological traits, we get just as high or even higher Pcor-values not only for the subtraction approximation method, but also for the division approximation method, these data are however not displayed.

5.2 Conclusion

The above used methods to convert input data to age independent data, for the purpose of finding greater correlations between data sets, are clearly not optimal.

5.3 Recommendations

To get better input data for cross correlation analysis, one could make experiment setups with PET and MRI scanners with higher resolutions, resulting in smaller voxels. Using a combined PET-MRI apparatus would also be good, as to do both scans simultaneously. It might also be good to make a simultaneous fMRI scan for the blood flow, to better calculate the arterious decay of [¹⁸F ] as a possible nuisance. A larger number of subjects is always a good thing for improved statistics. Although 83 subjects is seen as a large group when looking at medical research, it is still a very small sample size compared to the 7.2 billion humans alive today, or even to the 5.6 million Danes of which the group all volounteers were members. Further one could use other methods to calculate the age dependencies. According to D600 ([11] pp 25-26), the traits vary differently over age, thus one should not necessarily assume them all to behave as the same variety of time dependent function. In the paper mainly resulting from the experiment data [8], calculations where the levels of serotonin (5HT2A) receptor binding were adjusted to simulate that expected for a 40-year old subject, while the NEO PI-R data was untouched. There strong indications of correlations between neuroticism and serotonin binding in the frontolimbic system¹were found.

1Putamen, pallidus, anterior cingulate gyrus, caudate nucleus et. alt.

(35)

Appendix A

Glossary of Terms

A.1 NEO PI-R traits

n_neurot, neuroticism dim. a_agreea, agreeableness dim.

n_anxiet, anxiety facet a_trustp, trustp— facet n_angryh, angry hostility f. a_straig, straightforwardness f.

n_depres, depressin facet a_altrui, altruism facet n_self _c, self conciousness f. a_compli, compliance facet n_impuls, impulsiveness facet a_modest, modesty facet n_vulner, vulnerability facet a_tender, tender-mindness f.

e_extrav, extraversion dim. c_consci, concientiousness dim.

e_warmth, warmth facet c_compet, competence? facet

e_gregar, gregariousness f. c_orderp, orderp??? facet e_assert, assertiveness facet c_dutif u, dutifulness facet e_activi, activity facet c_achiev, achievement-striving f.

e_excite, excitement seeking f. c_self _d, self-discipline f.

e_positi, positive emotion f. c_delibe, deliberation facet

o_openne, openness dimension N _adj(ang.h.), neuroticism dimension with

o_f antas, fantasy facet angry hostility removed

o_aesthe, aesthetical emotion f. N _adj(ah, imp), neuroticism dimension

o_f eelin, feelings facet with angry hostility and

o_action, actions facet impulsiveness removed

o_ideasp, ideas facet o_values, values facet

(36)

A.2. Cerebral regions Glossary of Terms

A.2 Cerebral regions

Figure A.1: Regions of interest (ROI) displayed on the left hemisphere.

1 (green) l_supf c, r_supf c l. & r. superior frontal cortex 2 (yellow/blue) l_orbf rc, r_orbf rc l. & r. orbital frontal cortex 3, (apricot) l_suptem, r_suptem l. & r. superior temporal gyrus 4 (dark salmon) l_medinf , r_medinf l. & r. medial and inferior

temporal cortex

5 (mustard) l_entorh, r_entorh left & right entorhinal cortex 6 (sand) l_hippoc, r_hippoc left & right hippocampus 7 (apple green) l_th, r_th left & right thalamus 8 (blue) (see non_spec below) cerebellum

9, (piggy pink) l_occ, r_occ left & right occipital cortex 10 (pink/blue) l_parc, r_parc left & right parietal cortex 11 (yellow & pink

w. red stripes) l_senmot, r_senmot left & right sensory motor cortex 12 (yellow) l_medin2, r_medin2 l. & r. medial and inferior

frontal gyrus

13 (beige) l_postci, r_postci l. & r. posterior cingulate gyrus 14 (purple) l_cau, r_cau left & right caudate nucleus 15 (lavender) l_put, r_put left & right putamen

and pallidus

16 (dark brown) l_antcin, r_antcin l. & r. anterior cingulate gyrus

(37)

A.3. Miscellaneous Glossary of Terms

Regions of interest continued, those not displayed in figure A.1 midbrain, spans the midline

(behind thalamus and hippocampus) l_ins, r_ins left & right insula cortex (behind superior

temporal gyrus and inferior frontal gyrus)

wm white matter

csf cerebrospinal fluid

non_spec non-specific binding calculated from the cerebellar binding gm_no_re mean count density in gray matter

that does not belong to any region

A.3 Miscellaneous

CSV-file Comma separated values (data) in a raw data file.

CCA Canonical correlation analysis

D600 Abbreviation refering to the NEO PI-R Danish Manual ICA Independent component analysis

ROI Regions of interest VOI Volume of interest

PCC Partial correlation coefficient

Nota Bene that articles [8], [12], [13] in the bibliography below all are predominantly based on the same data as this thesis.

(38)

A.3. Miscellaneous Glossary of Terms

(39)

Bibliography

[1] John OP, Srivastava S (1999) The Big-Five Trait Taxonomy: Histo- rym Measurement, and Theoretical Perspectives Handbook of Per- sonality: Theory and Research

[2] Allport GW, Odbert HS (1936) Trait-names: A psycho-lexical study Psychological Monographs 47

[3] Catell RB (1943) The descripton of personality: Basic traits re- solved into clusters Journal of Abnormal and Social Psychology 38, pp 476-506

[4] Tupes EC, Christal RC (1992, reprint from 1961) Recurent personality factors based on trait ratings Journal of Personality 60, pp 225-251 [5] Fiske DW (1949) Consistency of the factorial structures of person-

ality ratings from different sources Journal of Abnormal and Social Psychology 44, pp 329-344

[6] Norman WT (1963) Toward an adequate taxonomy of personality attributes:Replicated factor structure in peer nomination personality ratings Journal of Abnormal and Social Psychology 66, pp 574-583 [7] Goldberg LR (1981) Language and individual differences: The search for universals in personality lexicons Review of personality and social psychology 1, pp 203-234

[8] Frøkjær VG, Mortensen EL, Nielsen FÆ, Haugbøl S, Pinborg LH, Adams KH, Svarer C, Hasselbalch SG, Holm S, Paulson OB, Knudsen GM (2008) Frontolimbic Serotonin 2A Receptor Binding in Healthy Subjects Is Associated with Personality Risk Factors for Affective Disorder Biological Psychiatry Volume 63, Issue 6, pp 569-576

[9] Pinborg LH et al. 2003 Quantification of 5HT2A receptors using [¹⁸F ]altanserin-PET and the bolus/infusion approach J. Cereb Blood Flow Metab 23, pp 985-96

[10] FÅ Nielsen 2002 Neuroinformatics in Functional Neuroimaging Ph.D. Thesis IMM-DTU ISSN 0909-3192

[11] Costa Paul T, McRae Robert R 2003(US), H Skovdal Hansen, E Lykke Mortensen (DK) NEO PI-R Manual, danish version ISSN ? ISBN?

A 0355-01

(40)

Bibliography Bibliography

[12] Svarer C, Madsen K, Hasselbalch SG, Pinborg LH, Haugbøl S, Frøkjær VG, Holm S, Paulson OB, Knudsen GM (2005) MR-based automatic delineation of volumes of interest in human brain PET images using probability maps NeuroImage 24, pp 969-979

[13] Adams KH, Pinborg LH, Svarer C, Hasselbalch SG, Holm S, Haugbøl S, Madsen K, Frøkjær VG, Martiny L, Paulson OB, Knudsen GM (2004) A databas of [¹⁸F ] altanserin binding to 5HT_2A receptors in normal volunteers: normative data and relationship to physiological and demographic variables NeuroImage 21, pp 1105-1113

[14] Fazio F, Perani D (2000) Importance of Partial-Volume Correction in Brain PET Studies Journal of Nuclear Medicine vol 41, pp 1849-1859 [15] Costa P, Terraciano A, McRae R (2001) Gender differences in personality traits across cultures: Robust and surprising findings.

Journal of Personality and Social Psychology Vol 81(2), pp 322-331 [16] Costa P, McRae R (2006) Age Changes in Personality and Their Ori-

gins: Comment on Roberts, Walton, and Viechtbauer Psychological Bulletin 132 (1), pp 26-28

[17] Roberts BW, Walton KE, Viechtbauer W (2006) Patterns of Mean- Level Change in Personality Traits Across the Life Course: A Meta-Analysis of Longitudinal Studies Psychological Bulletin 132 (1), pp 1-25

Can psychological aspects ibe localized in the brain? – Comparison of PET data with psychological mapping

Can psychological aspects be localized in the brain?

-

Comparison of PET data with Psychological Mapping

Carl Forsmark

November 14, 2014

A Master of Science Thesis in Engineering Physics

Abstract

Analys över hjärnans funktionalitet -

Jämförelse mellan PET-data och psykologisk kartläggning

Sammanfattning

Who, Where, Why, What?

Thank you

Contents

Chapter 1

Background and theory

1.1 NEO PI-R

1.1.1 Psychological taxonomy

1.1.2 The Big Five

1.1.3 Big Five in NEO PI-R

1.1.4 NEO PI-R in Danish

1.2 Positron emission tomography

1.3 Magnetic resonance imaging

1.4 Statistics

1.4.1 Data correlation

1.4.2 Independent component analysis

1.4.3 Statistical Parametric Mapping

1.4.4 Permutation test

Chapter 2

Experimental setup

2.1 The psychological questionnaires

2.2 The PET scan

2.3 The MRI

2.4 Quantification of receptor binding

2.5 Co-registration

Chapter 3

Methods

3.1 What seems to be the problem?

3.2 How will you fix it?

3.2.1 Control for dependency on sex (gender)

3.2.2 Preprocessing with respect to age dependency

3.3 Calculating significance values

Chapter 4

Results

4.1 Differences between the sexes

4.2 Preprocessing data

4.3 Results after preprocessing

Chapter 5

Discussion, conclusion, recommendations

5.1 Discussion

5.2 Conclusion

5.3 Recommendations

Appendix A

Glossary of Terms

A.1 NEO PI-R traits

A.2 Cerebral regions

A.3 Miscellaneous

Bibliography