STUDY OF EFFECTIVE CONNECTIVITY FOR FACE

PERCEPTION IN HEALTHY SUBJECTS AND PARKINSON'S

DISEASE

Elvis Lira da Silva

1,4

, Gabriela Castellano

1,4

, João Ricardo Sato

2,3,4

Ellison Fernando Cardoso

2,4

and Edson Amaro Jr.

2,4

Institute of Physics “Gleb Wataghin”, University of Campinas - UNICAMP, Campinas, Brazil

NIF, LIM-44, Department of Radiology, University of São Paulo, São Paulo, Brazil

Center of Mathematics, Computation and Cognition, Universidade Federal do ABC, Santo André, Brazil

CInAPCe Program (Cooperação Interinstitucional de Apoio a Pesquisas sobre o Cérebro), São Paulo, Brazil

Keywords: Dynamic Causal modelling, fMRI, Connectivity, Parkinson’s disease, Face perception.

Abstract: Facial perception is a fundamental task in our daily life and plays a critical role in social interactions.

Evidence from neuropsychological, neurophysiologic, and functional imaging studies indicated that face

perception is mediated by a specialized system in the human brain. We investigated the neural connectivity

induced by face presentation with different emotional valences in Parkinson's disease (PD) patients and a

control group of healthy, drug-free volunteers, using event-related fMRI in a parametric design. In this

study, we focused on applying Dynamic Causal Modelling (DCM), an approach that allows the assessment

of effective connectivity within cortical networks (Friston et al. 2003), to the study of effective connectivity

between maximally activated brain regions in response to passive viewing of facial stimuli. A connectivity

model was built based on the literature and in our fMRI analyses, which included the fusiform gyrus,

anterior cingulate gyrus, dorsolateral prefrontal cortex (DLPFC) and dorsomedial prefrontal cortex

(DMPFC). The results showed differences in connectivity between the PD group and the control group. We

found that the effective couplings among DLPFC/DMPFC and FG, DLPFC/DMPFC and ACG, were higher

in PD patients than healthy subjects, while the effective coupling among FG and ACG was lower in PD

patients.

1 INTRODUCTION

The branch of Neuroscience that studies functional

integration between cerebral areas has recently

shown a significant growth. Functional integration

refers to the interactions among specialized neuronal

populations, where the integration is mediated by the

so called effective connectivity. Effective

connectivity is defined as the influence that regions,

which encompass given neuronal populations, exert

on each other. It is important to study the effective

connectivity to know how different areas, involved

in a particular brain processing task, are related.

Facial perception is one of the fundamental tasks

in our daily life and plays a critical role in social

interactions. It is a highly developed visual ability in

humans and it is mediated by activation in a

distributed neural system that encompasses visual,

limbic, and prefrontal regions (Fairhall and Ishai,

2007; Haxby et al., 2000). Facial perception with

different emotional valences involves the emotional

recognition that is related to the activity of

amygdala, insula, orbitofrontal cortex and ventral

striatum. The areas linked to emotional regulation

include the anterior cingulate, dorsolateral and

medial prefrontal cortices (Phillips et al., 2003a, b).

In this study we investigated the effective

connectivity induced by face presentation with

different emotional valences in Parkinson's disease

(PD) patients and a control group of healthy, drug-

free volunteers. Depression is the most common

psychiatric disease in patients with Parkinson’s

disease (PD) (Cardoso et al., 2007). Although

several studies have been performed to investigate

the pathophysiology of depression in PD, many

questions remain unanswered.

Lira da Silva E., Castellano G., Ricardo Sato J., Fernando Cardoso E. and Amaro Jr. E. (2010).

STUDY OF EFFECTIVE CONNECTIVITY FOR FACE PERCEPTION IN HEALTHY SUBJECTS AND PARKINSON’S DISEASE.

In Proceedings of the Third International Conference on Bio-inspired Systems and Signal Processing, pages 88-93

DOI: 10.5220/0002743800880093

 SciTePress

To investigate effective connectivity within the

distributed cortical network for face perception, we

combined conventional Statistical Parametric

Mapping (SPM) and the technique of Dynamic

Causal Modelling (DCM) (Friston et al., 2003).

DCM consists of a theoretical-experimental

approach that treats the brain as a nonlinear

deterministic dynamic system. DCM regards an

experiment of fMRI as a designed perturbation of

neuronal dynamics that is distributed throughout a

system of coupled anatomical nodes to change

region-specific neuronal activity (Friston et al.,

2003). In practical terms, a reasonably realistic

neuronal model of interacting cortical regions is

built, with neuro-physiologically meaningful

parameters. This model is supplemented with a

forward model of how neuronal or synaptic activity

is transformed into a measured response, and the

parameters of this model can be estimated by

attempting to match the predicted BOLD signal to

the observed BOLD signal.

DCM has been previously used to investigate

visual perception and visual imagery of faces

(Mechelli et al., 2004) and face perception in healthy

subjects (Fairhall and Ishai, 2007; Rotshtein et

al.2007).

2 METHODS

Eckmann’s faces were morphed to produce neutral,

low and high intensities of sadness, as shown in

Figure 1. An event-related fMRI paradigm, similar

to Fu et al. (2004) was used. Facial stimuli and

baseline trials (crosshair fixation) were presented in

random order. Each trial and control condition was

presented for 2 s, and the inter-trial interval was

randomly varied according to a Poisson distribution

(2–12 s; mean 5 s). All images were acquired in a

1.5 T GE scanner, equipped with a 33 mT/m gradient.

Figure 1: An example of stimuli set used. Eckmann’s faces

were morphed to produce neutral, low and high intensities

of sadness. Facial stimuli and baseline trials (crosshair

fixation) were presented in random order in the event

related fMRI paradigm.

The images were oriented according to the AC–PC

line; and 168 brain volumes were acquired, with 15

slices each (7 mm thickness, 0.7 gap), 64×64 pixels

matrix, 20×20 mm FOV, 90° flip angle, 2.0 s TR, 40

ms TE, using a gradient echo EPI acquisition.

The fMRI statistical and DCM analyses were

performed using the free software Statistical

Parametric Mapping (SPM8, www.fil.ion.ucl.ac.uk

/spm/). All volumes were slice time corrected,

realigned to the middle volume, corrected for motion

artifacts, mean-adjusted by proportional scaling,

normalized into standard stereotactic space (template

provided by SPM8), and smoothed using a 8 mm

full-width at half-maximum (FWHM) Gaussian

kernel. The time series were high-pass filtered to

eliminate low-frequency components (filter width =

128 s) and adjusted for systematic differences across

trials.

DCM is constructed by a bilinear approximation

that allows the dynamics of the system to depend on

three groups of parameters: parameters that mediate

intrinsic coupling among the areas, parameters that

mediate the influence of extrinsic inputs on the areas

and (bilinear) parameters that allow the extrinsic

inputs to modulate that coupling (Friston et al.,

2003). The model depends on the experimental

design, where the extrinsic inputs enter the model by

two ways: directly influencing the areas (driving

inputs) and/or influencing the coupling among the

areas (contextual inputs).

All patients were recruited from the Movement

Disorders Clinics of Hospital das Clínicas –

University of São Paulo (São Paulo – Brazil) and all

gave written informed consent. The study was

approved by the ethics committee of the University

of São Paulo (Project Approval number: 414/03).

2.1 Effective Connectivity Analysis

We studied 19 healthy subjects and 17 patients with

Parkinson’s disease. Initially, fMRI was used for

locating brain responses to the experimental task

(face perception with different emotional valences).

Individual maps of activations were generated using

voxel based analysis. Next, second-level analysis

was used to generate maps of the group using one

sample t-test, with corrected (FWE) p-value < 0.05.

Based on our analysis of the healthy group data and

on the works of Phillips et al. (2003a,b), we

determined which areas should enter in the model of

DCM. The model included 3 areas: left and right

Fusiform Gyri (FG), Anterior Cingulate Gyrus

(ACG), and Dorsolateral Prefrontal Cortex (DLPFC)

STUDY OF EFFECTIVE CONNECTIVITY FOR FACE PERCEPTION IN HEALTHY SUBJECTS AND

PARKINSON'S DISEASE

Figure 2: DCM results for healthy subjects. Black arrows (and values) are the results of intrinsic connections, green arrows

(and values) are the results of modulatory connections, and red arrows (and values) are the results of direct influence of the

stimuli on the FG area.

/ Dorsomedial Prefrontal Cortex (DMPFC). These

regions of interest (ROIs) were defined using masks

created with the WFU Pickatlas software (Maldjian

et al., 2003 and Maldjian et al.,2004).

After delimiting the brain areas aforementioned

in the individual brain activation map, time series of

voxels limited by a sphere of 8 mm were extracted.

These spheres were located in the local maxima of

the activation map for each anatomical area included

on the model. This procedure was performed for

each of the subjects. These three volumes of interest

(VOIs) were identified for each individual subject.

Mean localization and t-values of these areas are

shown in Table 1 for healthy subjects and PD

patients. All three VOIs were reliably delineated in

16 of the 19 healthy subjects and 10 of the 17 PD

patients (Table 1) (p < 0.05, uncorrected).

Initially the DCM model was estimated

separately for each subject. In order to generalize

our results to the population level, the estimated

connection strengths from that analysis were then

subjected to a second-level analysis (using Matlab

functions) where the significance of inferred

connections was tested using one-sample t-tests

against the null hypothesis that the connection

strength is equal to zero.

Table 1: Regions of the DCM model for healthy subjects

and PD patients. The x, y, z, columns give the average

coordinates across all subjects for the location of each

region (with the standard deviation – SD – in parentheses).

The T column shows the average T-statistics across

subjects in the first-level analysis (and the SD in

parentheses). “L” and “R” mean Left and Right,

respectively.

Healthy subjects

Regions x y z T

FG (R) 31.3(7.1) -59.5(6.9) -18.4(3.1) 4.0(1.5)

ACG (R) 3.7(4.0) 14.2(9.9) 20(16,9) 2.53(0.8)

DMPFC/

DLPFC(R)

21.8(13.8) 19.3(14.7) 56(18) 4.6(1.5)

FG (L) -31.3(7.9) -58(14.4) -20(4.9) 3.8(1.1)

ACG (L) -7(2) 22.3(8.2) 8.3(14.9) 2.5(0.5)

DMPFC/

DLPFC(L)

-30.1(13.1) -2.5(8.7) 63.2(8.7) 4.0(1.4)

PD patients

Regions x y z T

FG (R) 30(8.6) 48.6(16.5) -21.1(7.9) 4.1(1.2)

ACG (R) 5.2(3.2) 20.3(5.6)

28.1(12,8

)

2.9(0.6)

DMPFC/

DLPFC(R)

17(15.2) 9.0(8.3)

59.7(13,3

)

4.1(1.0)

FG(L) -22(0) -67.5(10.6) -16.0(0) 4.9(1.2)

ACG (L) -11(0) 26.0(0) 28.0(0) 2.9(0)

DMPFC/

DLPFC(L)

-32(2.8) -3.5(10.6) 63.0(4.2) 4.3(0.7)

BIOSIGNALS 2010 - International Conference on Bio-inspired Systems and Signal Processing

Figure 3: DCM results for PD patients. Black, green and red arrows represent the same as for Figure 2.

3 RESULTS

We found that emotion perception and recognition in

faces involve activity in the FG, ACG, DLPFC and

DMPFC brain areas. The activation of these areas

was found through the fMRI analysis. A model that

connects the different areas was defined; taking into

account the FG, ACG and prefrontal cortex regions.

As a first model we admitted that all regions

interacted with one another; and that the face stimuli

entered the model as driving inputs on FG only. The

face stimuli also entered the model as contextual

factors on all connections. From this model, the

intrinsic connections between each of the

components of the model were estimated using a

DCM analysis.

The results of the groups are shown in Figure 2,

for healthy subjects, and Figure 3, for PD patients.

Only connections that were significant (non-zero

mean) after the one-sample t-test (p < 0.05) are

shown. Black arrows (and values) are the results of

intrinsic connections, which do not depend on

external stimuli. Green arrows (and values) are the

results of modulatory connections given by

contextual inputs; these values increase or decrease

the influence of the intrinsic connections and depend

on the external stimulus. Red arrows (and values)

are results of direct influence of stimuli on the areas.

Values are shown when significant (one-sample t-

test) for each of the experimental conditions

(neutral, low and high sadness).

In principle we can see, in control subjects

(Figure 2), an increase in activity in the FG areas

induced by modulation of connectivity by neutral

and high sadness faces (FG → FG connection) and

in the DLPFC/DMPFC areas induced by modulation

of connectivity by all faces (DLPFC / DMPFC →

DLPFC / DMPFC and ACG → DLPFC / DMPFC

connections).

Neutral faces increase activity in ACG areas (FG

→ ACG connection) and decrease activity in FG

(DLPFC / DMPFC → FG connection).

For PD patients (Figure 3), we can notice an

increase in activity in the DLPFC and DMPFC areas

induced by the modulation of connectivity by low

sadness (DLPFC/DMPFC → DLPFC/DMPFC

connection) and high sadness faces (DLPFC/

DMPFC → DLPFC/DMPFC and ACG →

DLPFC/DMPFC connections). We can also see an

increase in activity in the ACG areas induced by the

modulation of connectivity by high sadness faces

(ACG → ACG connection). In addition, we see a

decrease in activity of the FG area induced by the

modulation of connectivity by neutral faces (FG →

FG).

Using a two-sample t-test (through Matlab

functions) we compared the connections of the two

groups (healthy and PD) and found a significant

difference among the intrinsic connections (black

lines in Figures 2 and 3) DLPFC/DMPFC → ACG

(p-value of 0.0345), DLPFC/DMPFC → FG (p-

value of 0.0303) and FG → ACG (p-value of

0.0487).

We found that the effective coupling

DLPFC/DMPFC → ACG and DLPFC/DMPFC →

ACG were higher in PD patients than in healthy

subjects, while the effective coupling FG → ACG

was lower in PD patients. The results are shown in

Table 2.

STUDY OF EFFECTIVE CONNECTIVITY FOR FACE PERCEPTION IN HEALTHY SUBJECTS AND

PARKINSON'S DISEASE

To compare the connectivity patterns between

the stimuli of high sad and neutral faces, we used a

paired t-test among these conditions within each

group (PD patients and controls), for every

connection (intrinsic, modulatory and direct

influence) in every region. We found a significant

difference between the connectivity for these two

conditions in healthy subjects. The connectivity for

the sad faces stimulus has greater modulation by

prefrontal areas in the ACG and FG, which is in

agreement to the article by Philips et al. (Phillips et

al., 2003a). According to Phillips et al., the

prefrontal areas are responsible for the regulation of

the emotional state and the perception of emotion.

Therefore, we may conclude that in healthy subjects,

the prefrontal area regulated the emotional state due

to the presentation of the sad faces stimulus. On the

other hand, we did not observe this difference in the

group of Parkinson's disease patients: the t-test did

not show significant differences in connectivity

between the different face conditions. In fact, many

studies (Assogna et al. 2008; Dujardin et al 2004;

Sprengelmeyer et al. 2003) have described the

disability that Parkinson's patients have for the

recognition and perception of emotion, suggesting

that the decline in dopaminergic nigrostriatal leads

not only to motor disorders but also to a deficit in

processing facial emotions. This suggests that the

connection between the prefrontal cortex areas and

the other areas of the model is affected by

Parkinson’s disease, and causes this deficit in

processing emotions.

Results about the comparison between different

conditions for healthy subjects and PD patients are

shown in Table 3. For clarity only the results for the

comparison between the neutral and high sadness

conditions are shown. The connection FG →

DLPFC (intrinsic and modulatory) is not shown in

the table because it was not included in the model.

4 CONCLUSIONS

Using DCM, we explored the effective connectivity

of the main cerebral regions involved in responses to

facial stimuli with different intensities of sadness,

for both PD patients and healthy subjects. The

results showed differences in connectivity between

the PD group and the control group, suggesting that

these changes in connectivity can play an important

role in Parkinson's disease and may thus provide

insights on the underlying mechanisms of PD.

Table 2: Comparison between healthy subjects and PD

patients connections. Significant differences are marked

with bold-face. Connections marked with asterisk are

higher in PD patients than healthy subjects.

Comparison between groups

Connections

Healthy

Subjects

Patients

P-value

Intrinsic Connections

FG→FG -1.000 -1.000 -

FG→ACG 0.4908 0.2620 0.0487

ACG→FG 0.2080 0.2090 0.9850

ACG→ACG -1.000 -1.000 -

ACG→DLPFC 0.5636 0.5003 0.6691

DLPFC→FG 0.0361 0.1596 0.0303*

DLPFC→ACG 0.0855 0.2404 0.0345*

DLPFC→DLPFC -1.000 -1.000 -

Modulatory Connections (Neutral)

FG→FG 0.0587

0.0590 0.9896

FG→ACG 0.0557

-0.0820 0.0047

ACG→FG -0.0059

0.0017 0.6207

ACG→ACG 0.0014

0.0404 0.1135

ACG→DLPFC 0.0530

0.0390 0.6259

DLPFC→FG -0.0373

0.0128 0.0345*

DLPFC→ACG -0.0300

-0.0055 0.3046

DLPFC→DLPFC 0.0183

0.0261 0.6131

Modulatory Connections (Low Sadness)

FG→FG

-0.0192 0.0180 0.2513

FG→ACG

0.0102 0.0236 0.7676

FG → DLPFC

ACG→FG

-0.0245 -0.0010 0.3873

ACG→ACG

0.0042 0.0387 0.1190

ACG→DLPFC

0.0943 0.0671 0.5274

DLPFC→FG

-0.0246 0.0239 0.0324*

DLPFC→ACG

-0.0024 0.0011 0.8531

DLPFC→DLPF

0.0621 0.0660 0.8996

Modulatory Connections (High Sadness)

FG→FG

0.0337 0.0234 0.6424

FG→ACG

0.0047 0.0076 0.9513

ACG→FG

0.0226 0.0111 0.5818

ACG→ACG

0.0087 0.0477 0.0344*

ACG→DLPFC

0.0649 0.0754 0.7691

DLPFC→FG

0.0179 0.0068 0.6267

DLPFC→ACG

0.0061 0.0095 0.8716

DLPFC→DLPF

0.0404 0.0403 0.9974

Extrinsic Connections

Stimulus → FG (All others connections not included in

the model)

Neutral

0.1530 0.0845 0.0330

Low Sadness

0.1469 0.0999 0.2165

High Sadness

0.1390 0.0939 0.2400

BIOSIGNALS 2010 - International Conference on Bio-inspired Systems and Signal Processing

However, as the success of DCM is dependent on

the experimental design and on the specified

interacting regions model, other models involving

those regions should be tested for a more definitive

conclusion.

Table 3: Comparison between the neutral and high sadness

faces conditions, for healthy subjects and PD patients

connections. Significant differences are marked with bold-

face. Connections marked with asterisk are higher in the

high sadness than in the neutral condition. Healthy

subjects show many more significant differences between

high sadness and neutral faces than PD patients.

REFERENCES

Assogna F. , Pontieri F., Caltagirone C., Spalletta G. 2008.

The recognition of facial emotion expressions in

Parkinson's disease. European

Neuropsychopharmacology 18, 835–848.

Dujardin K., Blairy S., Defebvre L.,, Duhemb S., Noël Y,

Hess U., Destée A. 2004. Deficits in decoding

emotional facial expressions in Parkinson’s disease.

Neuropsychologia 42, 239–250

Cardoso, E., Fregni F., Maia F., Boggio, P., Myczkowski,

M., Coracini K., Vieira A., Melo L., Sato, J., Marcolin

M., Rigonatti S., Cruz A.,Barbosa, E. and Amaro, E.,

2007. rTMS treatment for depression in Parkinson’s

disease increases BOLD responses in the left refrontal

cortex. International Journal of

Neuropsychopharmacology, 11(2), 173 – 183.

Fairhall, S. and Ishai, A., 2007. Effective Connectivity

within the Distributed Cortical Network for Face

Perception. Cerebral Cortex, 17(10), 2400–2406.

Friston, K., Harrison, L., Penny, W., 2003. Dynamic

causal modeling. NeuroImage 19, 1273– 1302.

Fu C., Williams S., Cleare A., Brammer M., Walsh N.,

Kim J., Andrew C., Pich E., Williams P., Reed L., et

al. 2004. Attenuation of the neural response to sad

faces in major depression by antidepressant treatment:

a prospective, event-related functional magnetic

resonance imaging study. Archives of General

Psychiatry 61, 877–889

Haxby, J., Hoffman, E., Gobbini, M., 2000. The

distributed human neural system for face perception.

Trends in Cognitive Sciences, 4(6), 223–233.

Mechelli A., Price C., Friston K., Ishai A., 2004. Where

bottom-up meets top-down: neuronal interactions

during perception and imagery. Cerebral Cortex. 14,

1256-1265.

Maldjian, J., Laurienti, P., Burdette, J., Kraft, R., 2003. An

Automated Method for Neuroanatomic and

Cytoarchitectonic Atlas-based Interrogation of fMRI

Data Sets. NeuroImage, 19, 1233-1239.

Maldjian, J., Laurienti, P., Burdette, J., 2004. Precentral

Gyrus Discrepancy in Electronic Versions of the

Talairach Atlas. Neuroimage, 21(1), 450-455.

Phillips M., Drevets W., Rauch S., Lane R., 2003a.

Neurobiology of emotion perception I: The neural

basis of normal emotion perception. Biological

Psychiatry 54, 504–514.

Phillips M., Drevets W., Rauch S., Lane R., 2003b.

Neurobiology of emotion perception II: Implications

for major psychiatric disorders. Biological Psychiatry

54, 515–528.

Rotshtein P., Vuilleumier P., Winston J., Driver J., Dolan

R., 2007. Distinct and convergent visual processing of

high and low spatial frequency information in faces.

Cerebral Cortex 17, 2713–2724.

Sprengelmeyer R., Young A., Mahna K., Schroeder U. ,

Woitalla D., Büttner T. , Kuhn W. , Przuntek H. 2003.

Facial expression recognition in people with

medicated and unmedicated Parkinson’s disease.

Neuropsychologia 41, 1047–1057.

Healthy Subjects

(Comparison between conditions)

Modulatory Connections

Connections Neutral

High

Sadness

P-value

FG→FG 0.0587

0.0337

0.2430

FG→ACG 0.0557

0.0047

0.1149

ACG→FG -0.0059

0.0226

0.0459*

ACG→ACG 0.0014

0.0087

0.6311

ACG→DLPFC 0.0530

0.0649

0.5454

DLPFC→FG -0.0373

0.0179

0.0100*

DLPFC→ACG -0.0300

0.0061

0.0346*

DLPFC→DLPFC 0.0183

0.0404

0.0470*

Extrinsic Connections

Stimulus → FG

(All others

extrinsic

connections were

not included in the

model)

Neutral

High

Sadnes

P-value

0.1530 0.1390 0.0414

PD Patients

(Comparison between conditions)

Modulatory Connections

Neutral

High

Sadnes

P-value

FG→FG

0.0590 0.0234 0.1538

FG→ACG

-0.0820 0.0076 0.0427*

ACG→FG

0.0017 0.0111 0.5809

ACG→ACG

0.0404 0.0477 0.7515

ACG→DLPFC

0.0390 0.0754 0.2473

DLPFC→FG

0.0128 0.0068 0.7276

DLPFC→ACG

-0.0055 0.0095 0.6463

DLPFC→DLPFC

0.0261 0.0403 0.5526

Extrinsic Connections

Stimulus →FG

(All others

extrinsic

connections were

not included in the

model)

Neutral

High

Sadnes

P-value

0.0845 0.0939 0.4041

STUDY OF EFFECTIVE CONNECTIVITY FOR FACE PERCEPTION IN HEALTHY SUBJECTS AND

PARKINSON'S DISEASE