www.elsevier.com/locate/brainresprot

Brain Research Protocols

Protocol

Regression analysis utilizing subjective evaluation of emotional

experience in PET studies on emotions

Sargo Aaltoa,b,*, Esa Walliusc,1, Petri Naatanend, Jaana Hiltunene, Liisa Metsahonkalab,f,

Hannu Sipilab, Hasse Karlssonb,g

aDepartment of Psychology, Abo Akademi University, FIN-20500 Turku, FinlandbTurku PET Centre, University of Turku, FIN-20520 Turku, Finland

cInstitute of Signal Processing, DMI, Tampere University of Technology, Tampere, FinlanddDepartment of Psychology, University of Helsinki, Finland

eSection of Clinical Neurosciences, Finnish Institute of Occupational Health, Helsinki, FinlandfDepartment of Child Neurology, Helsinki University Hospital, Helsinki, Finland

gDepartment of Psychiatry, University of Helsinki, 00014 Helsinki, Finland

Accepted 27 June 2005

Available online 29 August 2005

Abstract

A methodological study on subject-specific regression analysis (SSRA) exploring the correlation between the neural response and the

subjective evaluation of emotional experience in eleven healthy females is presented. The target emotions, i.e., amusement and sadness, were

induced using validated film clips, regional cerebral blood flow (rCBF) was measured using positron emission tomography (PET), and the

subjective intensity of the emotional experience during the PET scanning was measured using a category ratio (CR-10) scale. Reliability

analysis of the rating data indicated that the subjects rated the intensity of their emotional experience fairly consistently on the CR-10 scale

(Cronbach alphas 0.70–0.97). A two-phase random-effects analysis was performed to ensure the generalizability and inter-study

comparability of the SSRA results. Random-effects SSRAs using Statistical non-Parametric Mapping 99 (SnPM99) showed that rCBF

correlated with the self-rated intensity of the emotional experience mainly in the brain regions that were identified in the random-effects

subtraction analyses using the same imaging data. Our results give preliminary evidence of a linear association between the neural responses

related to amusement and sadness and the self-evaluated intensity of the emotional experience in several regions involved in the emotional

response. SSRA utilizing subjective evaluation of emotional experience turned out a feasible and promising method of analysis. It allows

versatile exploration of the neurobiology of emotions and the neural correlates of actual and individual emotional experience. Thus, SSRA

might be able to catch the idiosyncratic aspects of the emotional response better than traditional subtraction analysis.

D 2005 Elsevier B.V. All rights reserved.

Theme: Neural basis of behavior

Topic: Motivation and emotion

Keywords: Subject specific regression analysis; Regression analysis; Correlation analysis; Subjective rating; Emotion; Sadness; Amusement; Statistical

parametric mapping; Positron emission tomography

1385-299X/$ - see front matter D 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.brainresprot.2005.06.003

* Corresponding author. Department of Psychology, Abo Akademi

University, 20500 Turku, Finland. Fax: +358 2 231 8191.

E-mail address: sargo.aalto@utu.fi (S. Aalto).1 Has contributed more than is typical for 2nd writer.

1. Type of research

The statistical analyses of imaging studies on emotions

have almost exclusively utilized the subtraction method (see

[30]), where two conditions, typically ‘‘activation’’ and

reference (‘‘neutral’’) states, are subtracted from each other,

which is a bipolar and fairly simplistic approach. This

15 (2005) 142 – 154

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154 143

approach is based on external validation of emotion

induction and presumes reliable coupling between the

stimulus and the response, i.e., that almost similar emotional

states can be induced in the study sample as in the validation

sample. However, even in controlled laboratory conditions,

emotional experience varies considerably between subjects

(see, e.g., [43]). Therefore, in parallel with the ‘‘traditional’’

subtraction approach, there is a need for methods of analysis

that could capture more flexibly the inherent individuality of

emotional responses.

The theoretical works by Friston et al. [10–12] have

served as a basis for subject-specific regression analysis

(SSRA) in a voxel-based statistical analysis of imaging data.

With this method, it is possible to examine the association

between regional cerebral blood flow (rCBF) reflecting

neuronal activity and external variables, e.g. subjective

ratings. SSRA with subjective self-evaluation of the emo-

tional experience has several potential advantages compared

to subtraction analysis in studies on emotion. Firstly, the

method does not need a reference (neutral) condition or a

validation procedure of dichotomous emotion induction,

which are required for the subtraction method. Secondly,

this approach allows individual responses to the material

used for emotion induction because it is based on individual

ratings of emotional experience, and the statistical model is

fitted individually for each subject. Thirdly, by using a

versatile rating questionnaire, it is possible to explore the

different emotions or aspects of emotions induced by the

stimuli using the same (multiscan) imaging data. Thus,

SSRA might be able to catch more fully the idiosyncratic

aspects of the emotional response, as it allows and draws on

the natural variation of emotional experience. On the other

hand, despite the individualistic approach and the subject-

specific statistical model, the results of a multi-subject study

can be generalized at the population level, using random-

effects analysis (RFX). Despite its possible methodological

benefits, SSRA utilizing self-evaluation of subjective emo-

tional experience has been utilized only in a few recent

fMRI studies [31,32,46]. However, to the best of our

knowledge, it has been neither applied nor methodologically

evaluated in positron emission tomography (PET) studies on

emotions.

The aim of the study is methodological evaluation of

SSRA utilizing subjective evaluation of emotional experi-

ence in imaging studies on emotion. We present a PET study

utilizing SSRA and subjective evaluation of emotional

experience. Two target emotions, amusement and sadness,

were induced with validated film stimuli. The rCBF values

were measured using PET, and the subjective intensity of

the emotional experience was rated on a category ratio scale

[5] after each PET scan. The correlation between rCBF and

the subjective rating data was analyzed using subject-

specific regression analysis with RFX to achieve results

generalizable at the population level. For comparison,

subject-specific subtraction analysis with RFX was per-

formed as well. In the present study, we utilize the PET data

with eleven subjects used in our earlier study presenting the

results of fixed-effects subtraction analysis [2]. The present

study is mainly methodological, and the discussion focuses

on the usability, benefits, and restrictions of SSRA in studies

on emotions.

2. Time required

Training of subjects to use the CR-10 scale: 30 min per

subject.

Manipulation check: 60 min.

Evaluation of rating data: 60 min.

Magnetic resonance imaging of the brain: 35 min per

subject.

PET data acquisition: 150 min per subject.

PET data preprocessing: 10 min per subject.

Voxel-based statistical analyses: 120 min.

Total time required to run the protocol with 11 subjects:

approximately 45 h.

3. Materials

3.1. Subjects

Eleven right-handed healthy female volunteers (mean

age 33.4; range 18 to 44 years) participated in the study

approved by the Ethics Committee of Turku University/

Turku University Central Hospital. The volunteers gave

written informed consent. (For a detailed description of the

study sample, see [2]).

3.2. Stimulus material

The stimulus material consisted of 12 film clips, each

lasting for an average of 2:30 min, with four films in each

category, to induce amusement, sadness and a ‘‘neutral’’

state. Four of the emotional film clips (When Harry met

Sally (amusement), two clips from Kramer vs. Kramer

(sadness), and The Champ (sadness)) were chosen from

previously validated film sets [16,33]. All the additional

films were validated for the purposes of the present study

with a sample of 51 university students. From this set, three

episodes of the film Bean-The Ultimate Disaster Movie

(amusement) and one episode from Kramer vs. Kramer

(sadness) were chosen. The four neutral films were

professional video material containing views of social

scenes, talking people, and scenes of nature with movement

and sound. The neutral films with minimal emotional

content were chosen in such a way that the basic sensory

properties (sound, movement, color) were similar to those of

the emotional films. To eliminate potential order effects, the

sequence of films was pseudo-random, but similar for all

subjects. The present set of films has been previously used

by Aalto et al. [2].

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154144

4. Detailed procedure

4.1. Measurement of the intensity of emotional experience

After each PET scan, the subjects rated their subjective

emotional experience during the film using a questionnaire,

which covered the thematic (appraisal), action readiness,

and feeling aspects of sadness, amusement, and a few

covariate emotions (interest, joy, disgust, fear, anger). The

items for amusement were: ‘‘Humorous or funny things

happened’’ (core-relational theme), ‘‘I wanted to smile or

laugh’’ (action readiness), and ‘‘I felt amused’’ (feeling).

The items for sadness were: ‘‘Something important was

lost’’ (core-relational theme), ‘‘I wanted to comfort’’

(action readiness), and ‘‘I felt sad’’ (feeling). The intensity

ratings of the feeling items of sadness and amusement

were used in this study. The intensities of the experienced

emotions were measured with a verbally anchored 12-step

(0, 0.5, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10) category ratio (CR-10)

scale [5], which allows linear quantification of perceptual

intensities [7,39]. To obtain maximal accuracy and

reliability of the intensity ratings, the subjects were trained

and tested with a special blackness test to use the CR-10

scale [6]. The subjects’ experiential responses to linearly

increasing blackness were within the 95% confidence

interval compared to the normal values [6]. The subjects

were instructed to rate their average emotional experience

during the film. Ratings based on one’s subjective

experience instead of normative expectations were encour-

aged. The questions were projected on the film screen

after each PET scan and the subjects answered them

verbally.

4.2. Manipulation check

(1) To check the effectiveness of the films in inducing the

target emotions of the present study, two repeated-measures

analyses of variance (rmANOVA) were conducted on the

amusement and sadness ratings separately across the twelve

films. Simple contrasts with two parameters were used. The

first parameter compared the ratings of amusement or

sadness between the corresponding target emotion films

and neutral films. The second parameter compared the

amusement or sadness ratings between the two target

emotion films. The ratings of both target emotions were

significantly the highest in the corresponding target emotion

films in comparison with the neutral or the other target film,

the lowest F(1, 10) = 57.2, P < 0.001.

(2) To ensure that amusement and sadness were the

most conspicuous components of the induced emotional

states, paired t tests were conducted to compare the mean

ratings of the target emotions to all the other covariate

emotions. Among all the feeling ratings, the sadness ratings

of the sadness films and the amusement ratings of the

amusement films reached significantly the highest level, the

lowest T = 3.4, P < 0.01. Statistical analyses concerning

the rating data were computed with SPSS for Windows,

Release 11.0.1.

4.3. Evaluation of rating data

The inter-subject consistency of the ratings, as measured

with Cronbach alphas of the amusement and sadness ratings

for the target films, varied from high (a = 0.84, amusement

ratings) to very high (a = 0.97, sadness ratings). The

corresponding alphas for both the target and the neutral

films (with very low emotional intensity) varied from 0.70

(amusement, excluding one neutral film of inconsistent

amusement ratings) to 0.79 (sadness). Thus, considering the

inter-personal context, the subjects rated their emotional

experience in a considerably stable manner throughout the

experiment, which supports the reliability of the ratings.

Considering the intra-personal context, the scatter plots

of the CR-10 ratings of the intensity of the emotional

experience show that the variety in the intra-subject ratings

was large and the profiles differed considerably between

subjects (Fig. 1, left panels). The films seem to provide the

necessary intra-subject variation for subject-specific regres-

sion estimates, although some individuals rated the films in

a more dichotomous manner. The scatter plots of the CR-10

ratings (Fig. 1, left panels) also show that some of the films

did not have the expected effect on some individuals (see,

e.g., subject 5, who experienced one of the neutral films as

highly amusing).

4.4. PET data acquisition

The PET data covering the whole brain were gathered in

the 3 D mode using a whole-body PET scanner (GE

Advance, Milwaukee, WI, USA), which provided 35

cantomeatal slices of 4.25 mm thickness and with 2.34 �2.34 mm in-plane voxel size. The performance characteristic

tests made on this scanner ensure that the transaxial spatial

resolution is 6 mm in the radial and 5 mm in the tangential

direction, and the axial resolution is 6.5 mm [24]. In each of

the 12 scans, the film was initiated approximately 20 s

before the injection of an intravenous 300 MBq [15O]-water

bolus (10 ml) into the left antecubital vein. The data were

integrated into a single 90-s frame. (For the details of PET

scanning, see [2]).

4.5. Preprocessing of imaging data

The preprocessing and statistical analyses were per-

formed using the Statistical Parametric Mapping [10]

software version 99 (SPM99), Statistical non-Parametric

Mapping 99 (SnPM99; [27]) and Matlab 5.3 for Windows

(Math Works, Natick, MA, USA). Realignment and spatial

normalization were made using the standard procedure of

SPM99 (see [2]) and, finally, the images were smoothed

with an isotropic Gaussian filter of 14 mm FWHM (full

width at half maximum).

Fig. 1. Scatter plots (left panels) and mean scores (right panels) of subjective intensity ratings of amusement (upper panels) and sadness (lower panels) based on

a category ratio scale (CR-10). In the scatter plots of subjective intensity ratings, the eleven subjects are shown on the horizontal axis and the intensity ratings of

the four emotion-inducing and the four neutral films on the vertical axis. The observations of emotion-inducing films are depicted as circles and the

observations of the neutral films as quadrangles. In the mean scores of the subjective intensity ratings, the CR-10 ratings of the target emotions during the

corresponding target emotion films are denoted by S and the ratings during the neutral films by N. The error bars represent standard error of mean (SEM).

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154 145

4.6. Voxel-based statistical analyses

Statistical testing was done using two types of procedure:

an exploratory analysis covered the entire brain without any

a priori hypothesis about the location of effects, while a

hypothesis-driven (confirmatory) VOI-based (volume of

interest) analysis enabled higher sensitivity.

4.6.1. Exploratory regression analysis

At the first phase of RFX analysis, SSRAs testing the

correlation between rCBF and the subjective ratings (CR-

10) utilized eight scans per subject, out of which four scans

corresponded to the four films inducing specific emotions

(amusement or sadness) and the other four to the ‘‘neutral’’

films. Separate SSRAs for both emotions were conducted

with a model including subject-by-covariate interaction.

Subject-specific AnCova global normalization was used,

and the mean voxel values were subject-specifically scaled

to 50 ml/dl per minute. The model had 33 parameters,

leaving 55 df for the estimation. The parameters in SSRAs

consisted of three parts: a subject part, a block part, and a

nuisance part. The subject part consisted of a block of

subjective ratings for each subject, yielding 11 parameters.

The block part contained 11 parameters from the number of

subjects, and the nuisance part included 11 subject-specific

parameters for global normalization of CBF.

After the SSRA model specification, subject-specific

contrasts were made and written to contrast images, which

served as the first, within-subject phase of RFX. As the

second, between-subjects phase of RFX, the null hypotheses

of no significant positive or negative correlation effects in

the contrast images were statistically tested with one-tailed

one-sample t tests using SnPM99. A corrected P value of

0.05 at a cluster level was applied, which had been obtained

by a corrected height threshold P value of 0.05 at the voxel

level combined with a critical suprathreshold size of 16

voxels. SnPM’s cluster level inference differs relevantly

from that of SPM. In SnPM, such results are presented that

(i) are significant at the cluster level or (ii) exceed the

critical threshold of the voxel-level inference. In the

analysis, variance smoothing of 12 mm and 2048 permu-

tations were used. Variance smoothing pools variance

estimates over neighboring voxels, giving additional

degrees of freedom and decreasing noise. It also makes

the statistic image, now called pseudo-t image, no longer t-

distributed. With more variance smoothing, the pseudo-t

thresholds for the analysis become lower, increasing the

sensitivity (statistical power) of the analysis. The corre-

sponding pseudo-t height thresholds were 4.55 in amuse-

ment and 4.29 in sadness.

4.6.2. Exploratory subtraction analysis

The comparative method, subject-specific subtraction

analysis, was applied separately for both emotions using a

condition-by-subject interaction model. Otherwise, the

model settings in the first phase of RFX were the same

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154146

as in SSRA. The model specifications resulted in a design

with 44 parameters leaving 55 df for the estimation. In

subtraction analyses, the subject part of the parameters

comprised a block of specific emotion and neutral scans

for each subject, yielding 22 parameters. The block part

(11 parameters) and the nuisance part (11 parameters)

were the same as in the SSRAs. The second phase of

RFX testing the null hypotheses of no significant

activation and deactivation effects was conducted as

described earlier in the context of SSRA. The resulting

pseudo-t height thresholds were 4.92 in amusement and

4.87 in sadness.

4.6.3. VOI-based analysis

The hypothesis-driven VOI analysis performed with

SPM was directed onto the amygdala, medial frontal

cortex (MFC), and subcallocal cingulate (SCC), which are

frequently shown to be involved in sadness, amusement,

or visually induced emotional state (for meta-analysis, see

[30]). The VOIs were drawn with the Imadeus software

(version 1.5, Forima Inc., Turku, Finland) on a ‘‘template’’

image averaged from spatially normalized T1-weighted

magnetic resonance images. The circular VOI with a 10

mm radius was placed bilaterally onto the four consec-

utive slices in the amygdala. The VOI for SCC was

placed onto the eight consecutive planes inferior to the

anterior genu of corpus callosum, corresponding mainly to

the undermost part of the Brodmann areas 32. The VOI

for MFC was placed onto the nine consecutive planes

including the medial aspects of the Brodmann areas 9 and

10, extending from the anterior boundary of the anterior

cingulate to the frontopolar tip. The search volume

covering bilaterally the amygdala, MFC, and SCC was

3300 voxels, which equals 26.4 cm3. Fig. 2 shows the

location and shape of the VOIs. The binary VOI image

was multiplied by the contrast images to obtain images

including only a region corresponding to VOI. A

corrected P value of 0.05 was used in the regression

and subtraction VOI analyses.

Fig. 2. Visualization of anatomical localization and

4.6.4. Localization and calculation of correlation

coefficients

For localization, the coordinates of the local maxima of

the clusters resulting from statistical analysis were converted

into Talairach coordinates [41] using the mni2tal conversion

software by Brett (http://www.mrc-cbu.cam.ac.uk/Imaging/

mnispace.html) and localized using the Talairach Deamon

software [20]. The correlation coefficient for the peak voxel

of each cluster was computed with Matlab after adjustment

for the effects of interest contrast.

5. Results

Fig. 3 presents the results of random-effects regression

and subtraction analyses for amusement (amusement vs.

neutral) and sadness (sadness vs. neutral) with three

orthogonal projections of a glass brain. The corresponding

anatomic regions are given in Table 1 for amusement and in

Table 2 for sadness.

In amusement, the most extensive and intensive (highest

pseudo-t statistics) cluster with a positive correlation

occurred in the right occipitotemporal area (BA19,

BA22, BA39; Talairach coordinates of the peak voxel

[44, �76, �6]) (Fig. 3, upper left panel). Clusters showing

positive correlations were also seen in the occipitotemporal

(BA18; [�36, �82, �4] and BA22; [59, �33, 7]) and

fusiform (BA37; [�42, �51, �13]) areas and in the

superior temporal gyrus on the left (BA38; [�26, 24,

�33]), in the right middle temporal lobe (BA21; [53, 10,

�29]), and in the cerebellum on the left [�10, �71, �15].

Clusters with negative correlations were seen in right

medial frontal gyrus (BA10; [8, 38, �9]) and in the AC

bilaterally (BA24), in the right inferior frontal gyrus

(BA46; [48, 39, 13], BA47; [14, 19, �18]), in the left

superior and middle frontal gyri (BA8; [�22, 24, 45],

BA10; [�22, 52, �3]; BA11, BA46; [�42, 28, 17]), in the

right supraorbital area (BA11; [4, 41, �31]), bilaterally in

the inferior and middle temporal lobes (BA20; right [65,

shapes of VOIs onto a MRI template image.

Table 1

Anatomical brain regions showing statistically significant effects in random-effects regression and subtraction analyses in amusement

Amusement

Random effects regression analysis Random effects subtraction analysis

Positive correlation of rCBF and amusement ratings Amusing > Neutral film blocks

Brain region Talairach

coordinates

of the peak

voxel

Local

maxima

(BA)

Cluster

size

pseudo-t

value

R Brain region Talairach

coordinates

of the peak

voxel

Local

maxima

(BA)

Cluster

size

pseudo-t

value

Right occipitotemporal

area

[44, �76, �6] BA 19,

22, 39

3501 9.23 0.65 Right occipitotemporal

area and gyrus

fusiformis

[48, �74, �3] BA19,

BA22

BA36

4193 12.90

Right middle temporal

gyrus

[53, 10, �29] BA 21 118 5.37 0.67 Right middle temporal

gyrus

[53, 6, �27] BA21 109 5.60

[65, �12, �6] BA 21 27 4.92 0.61

Left occipitotemporal

area

[�36, �82, �4] BA 18 1022 7.77 0.63 Left occipitotemporal

area and gyrus

fusiformis

[�38, �80, 1] BA19,

BA22

BA37

3398 11.53

[�59, �33, 7] BA 22 464 6.01 0.54

Left fusiform gyrus [�42, �51, �13] BA 37 277 6.59 0.70

Left superior temporal

gyrus

[�26, 24, �33] BA 38 34 6.45 0.54 Left superior temporal

gyrus

[�46, 16, �31] BA38 76 5.44

Left cerebellum [�10, �71, �15] 331 5.92 0.54 Left cerebellum [�12, �69, �12] 269 6.82

Negative correlation of rCBF and amusement Neutral > amusing film blocks

Right medial frontal

gyrus

[8, 38, �9] BA10 825 6.40 0.56 Right medial frontal

gyrus, left medial and

middle frontal gyrus

and anterior cingulate

[�24, 40, �17] BA11 3111 7.83

Anterior cingulate BA24 BA32

Right orbital gyrus [4, 41, �31] BA11 22 5.26 0.54

Left superior and

middle frontal gyrus

[�22, 52, �3] BA10,

BA11

417 5.59 0.53

Left middle frontal

gyrus

[�42, 28, 17] BA46 125 5.29 0.64

Right inferior frontal

gyrus

[48, 39, 13] BA46 236 5.87 0.65 Right inferior

frontal gyrus

[48, 37, 11] BA46 379 7.81

Right inferior frontal

gyrus

[14, 19, �18] BA47 30 4.90 0.53

Right middle frontal

gyrus

[32, 20, 47] BA8 27 5.16

Left superior frontal

gyrus

[�22, 24, 45] BA8 77 5.45 0.63 Left middle frontal

gyrus

[�22, 22, 45] BA8 80 5.64

Right inferior parietal

lobule

[57, �39, 44] BA40 138 5.46 0.58 Right inferior parietal

lobule

[57, �37, 44] BA40 208 6.35

Right inferior

temporal gyrus

[65, �24, �19] BA20 108 5.20 0.56 Right inferior temporal

gyrus

[67, �30, �17] BA20 494 7.25

Left inferior parietal

lobule

[38, �64, 42] BA39 23 4.81 0.60 Left inferior parietal

lobule

[�36, �64, 42] BA39 400 7.02

Left middle temporal

gyrus

[�61, �45, �11] BA20 35 4.85 0.61 Posterior cingulate and

paracentral lobule

bilaterally

[�4, �28, 33] BA31

BA5,

BA6

295 6.04

Brain regions showing statistically significant effects in random-effects regression (left tables) and subtraction analyses (right tables) in amusement. The upper

left table represents the areas showing a positive correlation, the lower left table contains negative correlations, while the upper right table shows activations

and the lower right table deactivations, respectively. A corrected P value of 0.05 at the cluster level (corrected height threshold P value of 0.05; critical

suprathreshold size of 16 voxels) was considered statistically significant in SnPM analyses. The results are localized according to the coordinates of the peak

voxel of the cluster and its local maxima. Anatomical brain regions, Brodmann areas (BA, for all local maxima of the cluster), cluster sizes in 2 � 2 � 2 mm

voxels, pseudo-t statistic values, and for regression analyses, correlation coefficients (denoted by R) for the peak voxels of the clusters are given.

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154 147

�24, �19], left [�61, �45, �11]), and bilaterally in the

parietal lobules (right BA40; [57, �39, 44]; left BA39;

[38, �64, 42]).

The results of subtraction analyses (Fig. 3, upper right

panels) differed slightly from those of regression analyses

in amusement. A small cluster in the right middle temporal

Table 2

Anatomical brain regions showing statistically significant effects in random-effects regression and subtraction analyses in sadness

Sadness

Random effects regression analysis Random effects subtraction analysis

Positive correlation of rCBF and sadness rating Sad > Neutral film blocks

Brain region Talairach

coordinates

of the peak

voxel

Local

maxima

(BA)

Cluster

size

pseudo-t

value

R Brain area Talairach

coordinate

of the peak

voxel

Local

maxima

(BA)

Cluster

size

pseudo-t

value

Right middle occipital

gyrus

[44, �76, �10] BA 18 214 4.81 0.70 Right occipital

area and cerebellum

[44, �80, �9] BA18 1992 8.84

Right fusiform gyrus [40, �51, �14] BA 36, 37 62 4.60 0.58

Right superior temporal

gyrus

[34, 26, �33] BA 38 30 4.96 0.60 Right fusiform

gyrus

[44, �46, �18] BA37 55 5.43

Right cerebellum [32, �81, �25] 41 4.72 0.53 Right middle

temporal gyrus

[65, �14, �3] BA21 715 8.72

[55, 7, �21] BA21 380 6.68

[46, 0, �42] BA38 65 5.70

Left inferior occipital

gyrus

[�38, �82, �4] BA 19 387 5.40 0.69 Left superior and

middle temporal

area

[�61, �16, 1] BA22,

BA38

3664 11.24

Left temporal lobe [�61, �16, 1] BA 22 824 6.99 0.73 Left inferior

occipital gyrus

[�40, �82, �6] BA18 841 8.74

[�26, 24, �33] BA 21, 38 465 6.71 0.57

[�40, �8, �42] BA 20 20 4.59 0.55

Left precentral gyrus [�44, �10, 43] BA 4 114 5.14 0.61 Left precentral gyrus [�48, �8, 41] BA4 92 5.55

Left superior frontal

gyrus

[�6, 54, 29] BA9 59 5.68

Right superior frontal

gyrus

[6, 9, 57] BA6 18 5.15

Negative correlation of rCBF and sadness rating Neutral > sad film blocks

Right middle frontal

gyrus

[26, 36, �17] BA11 432 7.61 0.90 Right inferior frontal

gyrus

[26, 34, �18] BA11 316 7.34

Right inferior frontal

gyrus

[40, 45, 0] BA46 189 4.88 0.88 Right inferior frontal

gyrus

[48, 39, 7] BA46 225 5.98

Right inferior parietal

lobule

[53, �35, 44] BA40 336 5.49 0.91 Right inferior

parietal lobule

[61, �37, 42] BA40 325 6.99

Right inferior temporal

gyrus

[65, �32, �20] BA20 816 8.70 0.81 Right inferior and

middle temporal

gyrus

[63, �30, �22] BA20 797 8.93

Left superior frontal

gyrus

[�26, 59, 6] BA10 78 5.00 0.87 Left middle frontal

gyrus

[�24, 38, �19] BA11 171 6.89

Left inferior frontal

gyrus

BA11 50 4.95 [�30, 59, 6] BA10 144 6.27

Left middle frontal

gyrus

[�26, 38, �19] BA11 50 4.95 0.88

Left middle and inferior

temporal gyrus

[�61, �45, �13] BA20 266 5.89 0.83 Left middle and

inferior temporal

gyrus

[�61, �45, �13] BA20 348 7.27

Left inferior parietal

lobule

[�40, �60, 45] BA40 128 4.69 0.89 Left parietal lobule [�48, �50, 39] BA40,

BA7

527 6.22

Right middle frontal

gyrus

[47, 27, 32] BA9 18 4.60 0.90 Anterior cingulate [�2, �28, 33] BA23 382 6.24

Left inferior frontal

gyrus

[�44, 37, 6] BA46 138 5.64

Right precuneus [18, �68, 44] BA7 112 5.80

Brain regions showing statistically significant effects in random-effects regression (left tables) and subtraction analyses (right tables) in sadness. The upper left

table depicts the areas showing a positive correlation, the lower left table includes negative correlations, and the upper right table shows activations and lower

right table deactivations, respectively. A corrected P value of 0.05 at the cluster level (corrected height threshold P value of 0.05; critical suprathreshold size of

16 voxels) was considered statistically significant in SnPM analyses. The results are localized according to the coordinates of the peak voxel of the cluster and

its local maxima. Anatomical brain regions, Brodmann areas (BA, for all local maxima of the cluster), cluster sizes in 2 � 2 � 2 mm voxels, pseudo-t statistic

values, and for regression analyses, correlation coefficients (denoted by R) for the peak voxels of the clusters are given.

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154148

Fig. 3. SnPM results of random-effects regression analyses (left panels) and random-effects subtraction analyses (right panels) of amusement (upper two rows)

and sadness (lower two rows) projected on a glass brain. The upper row in both emotions depicts positive correlations (left panel) and activation (right panel)

and the lower row negative correlations (left panel) and deactivation (right panel), respectively. The statistically significant effects of using a corrected P value

of 0.05 at the cluster level are shown (corrected height threshold P value of 0.05; critical suprathreshold size of 16 voxels). The images are presented in

accordance with the neurological convention (Right is right).

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154 149

gyrus (BA21; [65, �12, �6]) showed only a positive

correlation and no activation (amusement > neutral) in the

subtraction analysis. One cluster in the left middle

temporal gyrus (BA20; [�61, �45, �11]) showed a

negative correlation in the regression analysis, but did

not show deactivation (amusement < neutral) in the

subtraction analysis. The posterior cingulate gyrus

(BA31; [�4, �28, 33]) and the paracentral areas (BA 5,

BA6) showed only deactivation and no negative correla-

tion. The left frontal (BA 11; [�24, 40, �17]) and right

temporal areas (BA 20; [67, �30, �17]) with deactivation

were more extensive than the corresponding areas with

negative correlations.

In sadness, the cluster with the most extensive and

intensive positive correlation was situated in the left

temporal lobe (BA22; [�61, �16, 1]) (Fig. 3, lower left

panel). Several other clusters showing positive correla-

tions were also situated in the left occipitotemporal area

(BA21; [�26, 24, �33], BA38, BA20; [�40, �8, �42]),

BA19; [�38, �82, �4]). Positive correlations were

further seen in clusters in the superior temporal gyrus

(BA38; [34, 26, �33]), the middle occipital gyrus (BA18;

[44, �76, �10]), and the fusiform gyrus (BA36; [40,

�51, �14], BA37) on the right. Clusters showing positive

correlations were also found in the precentral gyrus (BA4;

[�44, �10, 43]) on the left and in the cerebellum [32,

�81, �25] on the right. In sadness, clusters with negative

correlations were found bilaterally in the inferior and

middle temporal gyri (BA20; right [65, �32, �20], left

[�61, �45, �13]), in the right inferior (BA46; [40, 45,

0]) and middle frontal gyri (BA9; [47, 27, 32], BA11;

[�26, 38, �19]), in the left superior and middle frontal

gyri (BA10; [�26, 59, 6], BA11; [�26, 38, �19]), and

bilaterally in the parietal lobe (BA40; right [53 ,�35, 44],

left [�40, �60, 45]).

In sadness, too, there were some differences between the

results of subtraction (Fig. 3, lower right panels) and the

regression analyses. The activated (sadness > neutral)

clusters in the cerebellum and in the right temporal lobe

(BA18; [44, �80, �9], BA37; [44, �46, �18]) were larger

than the clusters with positive correlations in the corre-

sponding brain areas. A small cluster in the superior frontal

gyrus (BA 9; [�6, 54, 29]) on the left and another in the

superior frontal gyrus (BA 6; [6, 9, 57]) on the right also

showed only activation and no positive correlation. A small

cluster in the right middle frontal gyrus (BA9; [47, 27, 32])

showed only a negative correlation and no deactivation

(sadness < neutral). Clusters in the left inferior frontal gyrus

(BA46; [�44, 37, 6]), in the anterior cingulate gyrus (BA

23; [�2, �28, 33]), and in the right precuneus (BA7; [18,

�68, 44]) showed only deactivation and no negative

correlation.

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154150

VOI-based analysis directed onto the amygdala, the

medial frontal cortex (MFC), and the subcallocal cingulate

(SCC) did not yield correlation or activation in either of the

emotions.

6. Discussion

Imaging studies of the neurobiological basis of emotions

have so far almost exclusively applied the subtraction method

(see [30] for a review), which reveals only dichotomous

activation or deactivation and does not allow a more versatile

evaluation of the neurobiology of emotional experience.

Subject-specific regression analysis (SSRA) utilizing sub-

jective evaluation of emotional experience might open up

new possibilities in studies of emotions. Although analysis of

covariance has been utilized earlier in PET studies on

emotions [23], SSRA utilizing self-ratings has not been

previously applied to PET studies exploring the correlation

between rCBF and subjective emotional experience. In the

present study, we utilized SSRA to explore the association

between rCBF and post-film subjective evaluation of the

intensity of emotional experience. The generalizability of the

results and inter-study comparability were ensured using

random-effects analysis (RFX). For comparison, subject-

specific subtraction analysis with RFX was also performed.

We found that the areas revealed by regression analyses

were mainly the same as those found using subtraction

analyses (see Fig. 3). In general, the comparison of the

results of SSRAs to the results of the previous studies on

emotions is not straightforward, as the earlier studies have

not utilized a subject-specific statistical model to explore the

correlation between self-evaluation of emotional experience

and rCBF. Furthermore, as far as we know, the previous PET

studies on emotions utilizing subtraction analysis have not

utilized RFX which limits the generalizability and compar-

isons between studies. Despite the considerable methodo-

logical differences between the present and previous studies,

the overlapping brain areas seen in both regression and

subtraction analyses have been detected in several earlier

studies on emotions using film stimuli. Activation of the

occipital cortex is a common finding when using visual

stimuli for emotions [30]. Activation in the temporal [22,28]

and occipital areas bilaterally [28,44] has been seen during

amusement in studies using film stimuli. Deactivation in the

left frontal and right parietal areas and in the anterior and

middle cingulate has also been reported previously to be

associated with film-induced amusement/happiness [28]. As

regards sadness, even other studies using film induction have

revealed increased activation in the cerebellum and bilat-

erally in the temporal area [4,22]. As the common and

distinct neural substrates of amusement and sadness were

examined and discussed in our earlier study [2], their

discussion is not included in the present study.

In some areas, however, the findings of regression and

subtraction analyses differed. The most prominent exclusive

findings of subtraction analyses were the deactivation of the

posterior cingulate without a negative correlation in amuse-

ment and the inactivation of the ventral part of AC (BA23)

bilaterally without a negative correlation in sadness.

Although activation of the AC would better agree with its

suggested role in emotional awareness [21], deactivation of

the AC related to emotions has also been reported [26,28].

In general, the lack of correlation might indicate a nonlinear

neural response or non-additive interaction [11]. On the

other hand, there were also regions that were only found in

the regression analyses. In amusement, a cluster in the right

middle temporal gyrus showed only a positive correlation

and no activation, and one cluster in the left middle temporal

gyrus showed a negative correlation but no deactivation. In

sadness, a cluster in the right middle frontal gyrus showed

only a negative correlation and no deactivation. As far as we

know, precisely, these areas have no previously identified

function in the neurobiology of emotion. It is not easy to

give a convincing interpretation of the deactivations or the

corresponding negative correlations found in the frontal

lobe. They might relate to functional interaction between

cognition and emotion (e.g., [15]). However, at least a

partial explanation might be that distinct emotional stimuli

probably provoke less cognition in comparison to clearly

non-emotional stimuli with otherwise equal content (for

further discussion, see [2]).

To enhance the sensitivity of the analysis, we conducted

VOI-based analyses focused onto the amygdala, the sub-

callosal cingulate (SCC), and the medial frontal cortex

(MFC), which did not reveal any findings related to sadness

or amusement. Although visual emotional stimuli generally

activate the amygdala [30], our results are in line with some

previous studies showing that film-induced emotional states

do not activate the amygdala [22,28]. In addition, our results

of a female sample are well in line with the findings reported

by Schneider et al. [37], who used ROI analysis to

demonstrate that females, unlike males, fail to show

amygdala activation or any association between amygdala

activity and mood change. VOI analysis did not reveal

activation or correlation in SCC, either, which is considered

to be involved in sadness [30]. However, even this agrees

with the previous studies showing that film-induced sadness

does not activate SCC [4,22,44]. Finally, VOI analysis did not

yield a positive correlation or activation in theMFC, although

MFC activation was seen in almost 40% of emotion studies

using visual induction [30]. However, some studies have also

shown deactivation of the MFC [14,26]. Although the lack of

findings in the amygdala, SCC, and MFC is not surprising in

view of the results of the above-mentioned eight studies, one

should also note that almost all earlier studies have been

performed without random-effects analysis, which weakens

the generalizability of their results. Apart from this methodo-

logical factor, the differences in the stimuli and the imaging

protocols might also partially explain why the recent meta-

analysis of emotion studies showed no specific brain area to

be activated consistently in the different studies [30].

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154 151

There were only a few regional differences between the

results of the regression and subtraction analyses, which

might be partially due to the properties of our data. Despite

the considerable variation in the subjective ratings (see Fig.

1, left panels), the intensities of the emotions were fairly

polarized at the group level due to the ‘‘neutral’’ films (see

Fig. 1, right panels), which increased the convergence of the

results. On the other hand, the very low intensity of

emotional involvement during the neutral films may have

decreased the accuracy of the ratings. This was demon-

strated by the lower reliabilities of the emotional ratings

over the neutral films, possibly influencing the results of

SSRA. There is also evidence to suggest that the activation

of the emotional system must reach a certain level to cause

congruence between the different response systems (e.g.,

facial expression and experience) [42,35], which might also

apply to the relationship between brain activity and self-

evaluation (ratings). Thus, emotional stimuli that induce a

response of at least moderate intensity and also cover the

entire range of intensity of a certain emotion would be most

suitable for performing SSRA (see [1]). However, SSRA

also seems feasible for the present fairly polarized data in

spite of the above-mentioned factors that detract from the

accuracy of ratings.

The rating of emotional experience may be considerably

inaccurate, especially when the object of a single rating is

the average intensity of the emotional experience over the

whole scanning session (at least 1 min in PET). An average

retrospective rating used in the present study does not reflect

the primary emotional experience in the best possible way

because there is no guarantee that the rated state remains

stable over this period. In addition, several factors, such as

the demand characteristics, stereotypes and temporal fea-

tures of the response, may hamper the validity of

retrospective subjective ratings and should be taken into

consideration when planning the experimental design (see,

e.g., [8] and [45]). However, the moderately high alphas of

the ratings through the films give some support for the

notion that there is true variation behind the retrospective

ratings of this study. The usability of subjective ratings

made after the scanning session was also demonstrated in a

recent fMRI study [31]. There is also empirical support for

the accuracy of retrospective rating of emotional state, as the

facial expressions induced by 1- to 2-min films and the

retrospective ratings made after a lag of at least 30 s

considerably overlap ([35], see also [36]). As fMRI studies

allow operation at short time intervals, they might offer

good opportunities to catch the immediate and momentary

aspects of emotional experience (see [31,32]).

All first-phase SPM analyses are fixed-effects analyses

that ignore the between-subject component of variance and

thus produce results that cannot be generalized beyond the

study sample [18]. Hence, when generalizable results and

reliable inter-study comparisons are wanted, RFX is the

only appropriate method of analysis. Due to the low degrees

of freedom, RFX was performed using SnPM with variance

smoothing, to increase the statistical power (sensitivity) of

the analysis. Although additional hypothesis-driven VOI

analysis enhanced the sensitivity of the analyses, the

relatively small study sample (11 subjects) may have

precluded us from detecting correlations in some regions

that might play a role in the neurobiology of emotional

experience. One should note that the fairly small sample size

also weakens the generalizability of the results. Finally,

although we have conducted mainly a comparison between

the statistical properties of the results, one should note that

statistical significance, per se, does not guarantee biological

relevance of the results.

SSRA utilizing subjective evaluation of emotional

experience has some methodological properties that open

up interesting possibilities for studies on emotion. It enables

emotions to be studied basically from an individual

subjective point of view, contrary to subtraction analysis,

which is based on validation and dichotomous selection of

the stimuli beforehand. Validation may not be appropriate

when stimuli are used in an actual imaging situation or for a

dissimilar group of study subjects, e.g., psychiatric patients.

SSRA is based on individual ratings made in the actual

experimental conditions and acknowledges that subjects can

experience a given stimulus in considerably different ways,

which enhances its psychological validity. Another advant-

age of regression analysis is that no particular reference

condition of the kind needed in subtraction analysis is

required. This makes regression analysis more specific

because the lack of sensory or cognitive equivalence

between neutral and emotional conditions easily hampers

the specificity of subtraction analysis (see, e.g., [34]). A

special advantage of the regression method is the possibility

to study the different nuances of an emotional experience.

An emotional experience is typically a multidimensional

construct containing elements from the emotion-eliciting

appraisal processes and the following action tendencies and

feelings [9,38]. With an appropriate rating questionnaire and

separate regression analyses, it would be possible to locate

brain areas associated with the different dimensions of

emotional experience. This novel approach also seems to be

applicable when utilizing an objective variable, e.g., in

research on the association between rCBF and the EEG

entropy index of the depth of anesthesia [25].

Subject-specific regression analysis utilizing subjective

evaluation of emotional experience turned out a promising

method of analysis in imaging studies of emotions. Our

results give support to the idea that, in several regions, the

neural responses related to amusement and sadness associate

linearly with the self-evaluated intensity of emotional

experience. The most significant benefit of SSRA using

subjective ratings is the possibility to explore the neuro-

biological correlates of actual and individual emotional

experience. Using a random-effects analysis, the results of a

multi-subject study can be generalized at the population

level, enabling reliable comparisons between the results of

different studies. The regression approach seems more

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154152

suitable than the traditional subtraction methodology in

studies focusing on subjective and experiential aspects of

emotions and might reveal completely new aspects in

studies on emotion-related phenomena.

7. Quick procedure

7.1 Measurement of emotional experience;

7.2 Manipulation check;

7.3 Evaluation of rating data;

7.4 PET data acquisition;

7.5 Preprocessing of image data;

7.6 Voxel-based statistical analyses;

(i) Exploratory regression analysis;

(ii) Exploratory subtraction analysis;

(iii) VOI-based analysis;

(iv) Localization and calculation of correlation

coefficients.

8. Essential literature references

[5–7,10,16,18,27,30,33,43]

Acknowledgments

The study has been funded by Signe and Ane Gyllenberg

Foundation. E.W. was funded by the Academy of Finland

and Tampere Graduate School of Information Science and

Engineering (TISE). The authors thank the following

persons for their valuable comments: Dr. Jyrki Mottonen,

Institute of Signal Processing, Tampere University of

Technology, Finland, and Virpi Ahola, M.Sc., Department

of Statistics, University of Turku, Finland.

Appendix A. Theory of subject-specific regression and

random effects analysis

A.1. General linear model

The subject-specific regression model was presented by

Friston et al. [10] in the framework of a general linear model

with the following matrix equation:

Y ¼ Gcb c þ Hlcl þ Hccc þ e; ð1Þ

where Y is the response matrix consisting of the voxels j =

1,. . .,J as columns and the observations (e.g. scans) i = 1,. . ., I

as rows. The columns of Gc contain the covariates of interest

(e.g. subject-by-covariate interaction), Hl is a matrix con-

taining the effects of no interest (e.g. subject or block effects),

Hc contains the nuisance variables (e.g. global activity),

matrix bc includes the regression coefficients of interest,

matrix cl contains the effects of no interest, matrix cc includes

the regression coefficients for the nuisance variables, and ( isthe residual matrix. It is assumed that the residuals are

independent and normally distributed with zero mean and

variance rj2 [N(0, rj

2)]. In the regression model, it is possible

to test statistically the distinct contribution of the covariate of

interest (e.g. subjective rating) in Gc to the variance of the

response (rCBF) while simultaneously controlling for the

effect of the nuisance covariates in Hc. However, one should

be careful in choosing the covariates, since the use of

correlated covariates may confound the results (see [3]).

A.2. Application to multi-subject repeated-measures data

SSRA is based on the subject-specific regression model

(SSRM), where a parametric regressor is fitted at a single-

subject level for each subject of the study. It differs

fundamentally from the group-level regression model

known as ‘‘simple regression’’ in Statistical Parametric

Mapping (SPM) or, more generally, as ‘‘correlation analy-

sis’’. The group-level model is inappropriate in multi-subject

repeated-measures designs with multiple pairs of observa-

tions (scans and corresponding ratings) per subject because

it leads to a within-subject covariance structure, which

violates the fundamental independence assumptions under-

lying the analyses. SSRM models the within-subject

covariance structure by decomposing the observations into

blocks for each subject, and the model therefore includes

only within-subject variation, so that the independence of

residuals can be maintained. SSRM also allows subject-by-

covariate interactions, i.e., individual variation in the slopes

of the regression lines, which may be considerable,

especially in the context of study on emotions.

A.3. Types of inference and generalizability of results

All first-level SPM analyses are fixed-effects analyses

(FFX), but subject-specific models enable random-effects

analysis (RFX) [18] as a second-level analysis. The type of

statistical inference and the generalizability of the results

differ essentially between FFX and RFX (see [12,13] for a

review). In FFX, the error variance is estimated on a scan-

to-scan basis, whereas in RFX, it consists of both scan-to-

scan error variance and a between-subject component of

variance. FFX tests statistically the subjects’ mean effect,

which, however, might be due to a large effect in some

subjects, but no effect in the others. Whenever there is a lot

of between-subject variation in the effects, which is typical

in emotional responses, the within-subject variance in FFX

is inappropriate for between-subject (group-level) inference

[18]. Thus, FFX is principally a case study [18] allowing

only inference for the subjects studied.

RFX or, to be exact, mixed-effects analysis enables

population level inference. This is achieved by testing

statistically the significance of the mean effect in the within-

subject parameter estimates (h:s) for the sum of scan-to-scan

error variance and a between-subject component of var-

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154 153

iance. RFX is the only method that yields generalizable

results and reliable inter-study comparisons. However, this

is accomplished at the cost of degrees of freedom (df),

which collapse from N (scans)-rank(X) to N (subjects)-1,

where N (scans) is the number of scans, N (subjects) is the

number of subjects, and rank (X) is the rank, i.e., the

number of linearly independent columns of the design

matrix X, which consists of Gc, Hl, and Hl in (1).

A two-phase RFX can be made using SPM as well as the

first-level analysis. However, the study sample should be

fairly large to yield a reliable estimate of the between-

subject variance needed in RFX [13]. Moreover, with low df

or low smoothness, the multiple-comparison correction in

SPM based on the theory of random fields [47] has been

found too rigorous for voxel-level inference [17,40],

resulting in insensitivity of analysis. Thus, in studies using

typical samples (n < 15), the recently developed Statistical

non-Parametric Mapping 99 software (SnPM99) [27] is a

promising tool for RFX. In SnPM, the multiple comparison

correction is based on permutations, which enables analyses

with small samples [19,27]. Moreover, it is possible to use

non-parametric pseudo-t statistics [27,29] with local

smoothing of noisy variance images. This increases the

statistical power of the analysis, making SnPM even more

appropriate for RFXs with a modest number of subjects.

References

[1] S. Aalto, P. Naatanen, L. Metsahonkala, E. Wallius, J. Hiltunen, H.

Karlsson, H. Sipila, Neuroanatomical correlates of film induced

disgust: a PET activation study, NeuroImage 13 (2001) S376.

[2] S. Aalto, P. Naatanen, E. Wallius, L. Metsahonkala, H. Stenman, P.M.

Niemi, H. Karlsson, Neuroanatomical substrata of amusement and

sadness: PET activation study using film stimuli, NeuroReport 13

(2002) 67–73.

[3] A. Andrade, A.-L. Paradis, S. Rouquette, J.-B. Poline, Ambiguous

results in functional neuroimaging data analysis due to covariate

correlation, NeuroImage 10 (1999) 483–486.

[4] M. Beauregard, J.-M. Leroux, S. Bergman, Y. Arzoumanian, G.

Beaudoin, P. Bourgouin, E. Stip, The functional neuroanatomy of

major depression: an fMRI study using an emotional activation

paradigm, NeuroReport 9 (1998) 3253–3258.

[5] G. Borg, A category scale with ratio properties for intermodal and

interindividual comparisons, in: H.G. Geissler, P. Petzold (Eds.),

Psychophysical Judgment and the Process of Perception, Deutscher

Verlag der Wissenschaften, Berlin, 1982, pp. 25–34.

[6] G. Borg, E. Borg, A general psychophysical scale of blackness and its

possibilities as a test of rating behavior. Reports from the Department

of Psychology, No. 737 Stockholm University, Stockholm, 1991.

[7] G. Borg, H. Diamant, L. Strom, Y. Zotterman, The relation between

neural and perceptual intensity: a comparative study on the neural

and psychophysical response to taste stimuli, J. Physiol. 192 (1967)

13–20.

[8] G. Braucht, Nicholas, Analysis and reduction of components of

systematic error in ratings,Multivariate Behav. Res. 7 (1972) 203–222.

[9] N.H. Frijda, The Emotions, Cambridge Univ. Press, Cambridge, 1986.

[10] K.J. Friston, A.P. Holmes, K.J. Worsley, J.-B. Poline, C.D. Frith,

R.S.J. Frackowiak, Statistical parametric maps in functional imaging:

a general linear approach, Hum. Brain Mapp. 2 (1995) 189–210.

[11] K.J. Friston, C.J. Price, P. Fletcher, C. Moore, R.S.J. Frackowiak, R.J.

Dolan, The trouble with cognitive subtraction, NeuroImage 4 (1996)

97–104.

[12] K.J. Friston, A.P. Holmes, C.J. Price, C. Buchel, K.J. Worsley,

Multisubject fMRI studies and conjunction analyses, NeuroImage 10

(1999) 385–396.

[13] K.J. Friston, A.P. Holmes, K.J. Worsley, How many subjects constitute

a study?, NeuroImage 10 (1999) 1–5.

[14] J. Geday, A. Gjedde, A.-S. Boldsen, R. Kupers, Emotional valence

modulates activity in the posterior fusiform gyrus and inferior medial

prefrontal cortex in social perception, NeuroImage 18 (2003) 675–684.

[15] J.R. Gray, T.S. Braver, Integration of emotion and cognitive control: a

neurocomputational hypothesis of dynamic goal regulation, Emotional

Cognition: From Brain to Behaviour, John Benjamins Publishing

Company, Amsterdam, Netherlands, 2002, pp. 289–316.

[16] J.J. Gross, R.W. Levenson, Emotion elicitation using films, Cogn.

Emot. 9 (1995) 87–108.

[17] A.P. Holmes, Statistical issues in functional brain imaging. Doctor of

Philosophy thesis, University of Glasgow, Glasgow, 1994.

[18] A.P. Holmes, K.J. Friston, Generalisability, random effects and

population inference, NeuroImage 7 (1998) S754.

[19] A.P. Holmes, R.C. Blair, J.D.G. Watson, I. Ford, Non-parametric

analysis of statistic images from functional mapping experiments,

J. Cereb. Blood Flow Metab. 16 (1996) 7–22.

[20] J.L. Lancaster, M.G. Woldorff, L.M. Parsons, M. Liotti, C.S. Freitas,

L. Rainey, P.V. Kochunov, D. Nickerson, S.A. Mikiten, P.T. Fox,

Automated Talairach atlas labels for functional brain mapping, Hum.

Brain Mapp. 10 (2000) 120–131.

[21] R. Lane, Neural correlates of conscious emotional experience, in: N.

Lane (Ed.), Cognitive Neuroscience of Emotion, Oxford Univ. Press,

New York, 2000, pp. 345–370.

[22] R.D. Lane, E.M. Reiman, G.L. Ahern, G.E. Schwartz, R.J. Davidson,

Neuroanatomical correlates of happiness, sadness, and disgust, Am. J.

Psychiatry 154 (1997) 926–933.

[23] R.D. Lane, E.M. Reiman, B. Axelrod, L.S. Yung, A. Holmes, G.E.

Schwartz, Neural correlates of levels of emotional awareness:

evidence of an interaction between emotion and attention in the

anterior cingulate cortex, J. Cogn. Neurosci. 10 (4) (1998) 525–535.

[24] T.K. Lewellen, S.G. Kohlmyer, R.S. Miyaoka, M.S. Kaplan, C.W.

Sterans, S.F. Schubert, Investigation of the performance of the general

Electric ADVANCE positron emission tomograph in 3 D mode, IEEE

Trans. Nucl. Sci. 43:2 (1996) 1999–2206.

[25] A. Maksimow, K.K. Kaisti, S. Aalto, M. Maenpaa, S. Jaaskelainen, S.

Hinkka, S. Martens, M. Sarkela, H. Viertio-Oja, H. Scheinin, EEG

Spectral Entropy Correlates with Regional Cerebral Blood Flow

During Sevoflurane and Propofol Anesthesia. In press, Anaesthesia.

[26] H.S. Mayberg, M. Liotti, S.K. Brannan, S. Mcginnis, R.K. Mahurin,

P.A. Jerabek, J. Arturo Silva, J.L. Tekell, C.C. Martin, J.L. Lancaster,

P.T. Fox, Reciprocal limbic-cortical function and negative mood-

converging PET findings in depression and normal sadness, Am. J.

Psychiatry 156 (1999) 675–682.

[27] T.E. Nichols, A.P. Holmes, Nonparametric permutation tests for

functional neuroimaging: a primer with examples, Hum. Brain Mapp.

15 (2001) 1–25.

[28] S. Paradiso, R.G. Robinson, N.C. Andreasen, J.E. Downhill, R.J.

Davidson, P.T. Kirchner, G.L. Watkins, L.L. Boles Ponto, R.D.

Hichwa, Emotional activation of limbic circuitry in elderly normal

subjects in a PET Study, Am. J. Psychiatry 154 (1997) 384–389.

[29] K.M. Petersson, T.E. Nichols, J.-B. Poline, A.P. Holmes, Statistical

limitations in functional neuroimaging II. Signal detection and statistical

inference, Philos. Trans. R. Soc. London, B 354 (1999) 1261–1281.

[30] K.L. Phan, T. Wager, S.F. Taylor, I. Liberzon, Functional neuro-

anatomy of emotion: a meta-analysis of emotion activation studies,

NeuroImage 16 (2002) 331–348.

[31] K.L. Phan, S.F. Taylor, R.C. Welsh, L.R. Decker, D.C. Noll, T.E.

Nichols, J.C. Britton, I. Liberzon, Activation of the medial prefrontal

cortex and extended amygdala by individual ratings of emotional

arousal: a fMRI study, Biol. Psychiatry 53 (2003) 211–215.

S. Aalto et al. / Brain Research Protocols 15 (2005) 142–154154

[32] K.L. Phan, S.F. Taylor, R.C. Welsh, S.-H. Ho, J. Britton, I. Liberzon,

Neural correlates of individual ratings of emotional salience: a trial-

related fMRI study, NeuroImage 21 (2004) 768–780.

[33] P. Philippot, Inducing and assessing differentiated emotion-feeling

states in the laboratory, Cogn. Emot. 7 (1993) 171–193.

[34] E.M. Reiman, R.D. Lane, G.L. Ahern, G.E. Schwartz, R.J. Davidson,

K.J. Friston, L.S. Yun, K. Chen, Neuroanatomical correlates of

externally and internally generated human emotion, Am. J. Psychiatry

154 (1997) 918–925.

[35] E.L. Rosenberg, P. Ekman, Coherence between expressive and

experiential systems in emotion, Cogn. Emot. 8 (1994) 201–229.

[36] W. Ruch, Will the real relationship between facial expression and

affective experience please stand up: the case of exhilaration, Cogn.

Emot. 9 (1995) 33–58.

[37] F. Schneider, U. Habel, C. Kessler, J.B. Salloum, S. Posse, Gender

differences in regional cerebral activity during sadness, Hum. Brain

Mapp. 9 (2000) 226–238.

[38] C.A Smith, R.S. Lazarus, Appraisal components, core relational

themes, and the emotions, Cogn. Emot. 7 (1993) 233–269.

[39] S.S. Stevens, On the psychophysical law, Psychol. Rev. 64 (1957)

153–187.

[40] J. Stoeckel, J.-B. Poline, G. Malandain, N. Ayache, J. Darcourt,

Smoothness and degrees of freedom restrictions when using SPM99,

NeuroImage 13 (2001) S259.

[41] J. Talairach, P. Tournoux, Co-Planar Stereotaxic Atlas of the Human

Brain: Three-Dimensional Proportional System, Georg Thieme,

Stuttgart, 1988.

[42] L.G. Tassinary, J.T. Cacioppo, Unobservable facial actions and

emotion, Psychol. Sci. 3 (1992) 28–33.

[43] R. Westermann, K. Spies, G. Stahl, F.W. Hesse, Relative effectiveness

and validity of mood induction procedures: a meta-analysis, Eur. J.

Soc. Psychol. 26 (1996) 557–580.

[44] B.E. Wexler, I. Prohovnik, R.K. Fulbright, P. Skudlarski, M.

Anderson, C.M. Lacadie, J.C. Gore, Brain activation by happy and

sad emotional stimulation, NeuroImage 9 (1999) S369.

[45] R.J. Wherry, C.J. Bartlett, The control of bias in ratings: a theory of

rating, Pers. Psychol. 35 (1982) 521–551.

[46] J.S. Winston, B.A. Strange, J. O’Doherty, R.J. Dolan, Automatic and

intentional brain responses during evaluation of trust-worthiness of

faces, Nat. Neurosci. 5 (2002) 277–283.

[47] K.J. Worsley, S. Marrett, P. Neelin, A.C. Vandal, K.J. Friston, A.C.

Evans, A unified statistical approach for determining significant

signals in images of cerebral activation, Hum. Brain Mapp. 4 (1996)

58–73.