Professional Documents
Culture Documents
ORIGINAL INVESTIGATION
1
Institute of Sexual Medicine and Forensic Psychiatry and Psychotherapy, Kiel University, Medical School, Kiel, Germany, 2Department of
Neurology, Kiel University, Medical School, Kiel, Germany, 3Department of Radiology and Neuroradiology, Kiel University, Medical School,
Kiel, Germany, 4Institute of Sexology and Sexual Medicine, Charite - Universitatsmedizin Berlin, Berlin, Germany, 5Practice for Sexual
Medicine, Kiel, Germany, 6Danish Research Centre for Magnetic Resonance, Copenhagen University Hospital Hvidovre, Hvidovre, Denmark,
and 7Department of Neurology, Copenhagen University Hospital Bispebjerg, Bispebjerg, Denmark
ABSTRACT
ARTICLE HISTORY
Objectives: Objective assessment of sexual preferences may be of relevance in the treatment and
prognosis of child sexual offenders. Previous research has indicated that this can be achieved by
pattern classification of brain responses to sexual child and adult images. Our recent research
showed that human face processing is tuned to sexual age preferences. This observation prompted
us to test whether paedophilia can be inferred based on the haemodynamic brain responses to
adult and child faces. Methods: Twenty-four men sexually attracted to prepubescent boys or girls
(paedophiles) and 32 men sexually attracted to men or women (teleiophiles) were exposed to
images of child and adult, male and female faces during a functional magnetic resonance imaging
(fMRI) session. Results: A cross-validated, automatic pattern classification algorithm of brain
responses to facial stimuli yielded four misclassified participants (three false positives), corresponding to a specificity of 91% and a sensitivity of 95%. Conclusions: These results indicate that the
functional response to facial stimuli can be reliably used for fMRI-based classification of
paedophilia, bypassing the problem of showing child sexual stimuli to paedophiles.
Introduction
Paedophilic men experience intensive and sustained
sexual attraction to prepubescent children. In many
cases they act on these sexual urges. Despite the
somewhat declining prevalence of child sexual abuse,
an estimated 6.7% of children in the United States have
been sexually abused (Finkelhor et al. 2010). Because
child sexual abuse leads to high tangible and intangible
costs, child sexual abuse is of high public concern.
However, many, but not all, child sexual offences were
committed by paedophiles. Studies that used phallometry to assess sexual preference suggest that about
half of first-time child sex offenders were not paedophilic (Blanchard et al. 2001). This underlines the
importance of a reliable assessment of sexual orientation
in child sex offenders, given that treatment strategies
with respect to child sex offenders depend on the
offenders sexual preference. Assessment of sexual
preferences in child sex offenders is also important for
CONTACT Jorge Ponseti
School, Kiel, Germany.
2015 Taylor & Francis
ponseti@sexmed.uni-kiel.de
Institute of Sexual Medicine and Forensic Psychiatry and Psychotherapy, Kiel University, Medical
40
J. PONSETI ET AL.
paedophiles were heterosexual paedophiles, the remaining two were homosexual paedophiles. Eleven of the
paedophilic participants were part of the Dunkelfeld
prevention project (Beier et al. 2009). The remaining 13
paedophiles were treated in our outpatient department.
Ten paedophiles had been sentenced earlier in life
(mostly because of possession of child pornography),
with only three of them having served a prison sentence.
Thirty-two male participants sexually attracted to adults
(teleiophiles) served as the control group. Of these, 18
were sexually attracted to women (heterosexual teleiophiles, 32.4 [8.2] years [range, 2249 years]) and 14
were sexually attracted to men (homosexual teleiophiles,
28.6 [5.7] years [range, 2342 years]). Groups were
matched for age (F3,52 1.81; P 0.16) and intelligence
(F3,52 1.25; P 0.30).
Procedure
During the fMRI session, participants were exposed to
14 different categories of visual stimuli: pictures of
naked women, men, girls and boys (whole body frontal
views, genitals only and face only) and non-sexual
pictures with either high or low arousal scores from the
International Affective Picture System (IAPS; Lang et al.
1997). Thirty-five pictures within each category were
presented. Only the haemodynamic responses to face
images entered the analysis of the present study,
corresponding to a total of 140 trials per fMRI session.
Each stimulus was presented for 1 s with a variable
interstimulus interval (of 26 s) in a pseudorandom
order. To ensure that participants paid attention to the
stimuli, they manually responded when an oddball
stimulus (green circle) appeared on screen. The oddball
was presented 20 times during the fMRI session.
Participants rated the attractiveness of the stimuli
after the fMRI session.
Methods
Participants
41
Figure 1. Differences in brain activity between paedophiles and healthy volunteers. The statistical maps show regions where the brain
response is increased for the preferred stimuli contrary to the non-preferred stimuli (with a threshold at P50.001, uncorrected). Panel
A shows differences for the male faces and panel B shows differences for the female faces. The t-score maps were overlaid on a
structural T1-weighted MRI. The numbers represent the stereotactic z coordinate corresponding to each axial slice.
42
J. PONSETI ET AL.
Results
In our previous fMRI study we found that the ratings of
face pictures consistently reflected the sexual preference
of the participants (Ponseti et al. 2014). Two-sample
t-test (step 2) revealed significant differences of the
individual difference maps between paedophilic and
teleiophilic participants in extended brain areas. This was
found in the girls vs. women as well as in the boys vs.
men comparison (Figure 1 and Table 1). However, group
differences were more extended in the girls vs. women
than in the boys vs. men comparison.
The brain responses of each participant were characterised by two individual expression values, one
corresponding to the girls5women comparison and
one corresponding to the boys5men comparison.
Figure 2 shows the individual expression values for all
56 participants. Cross-validation showed that a total of
four participants were misclassified. Three teleiophilic
participants were classified as paedophiles (false positives), and one paedophile was classified as teleiophile
(false negative), corresponding to a specificity of 91%,
a sensitivity of 95% and a mean classification accuracy
of 93%.
By analysing fMRI images in response to body and
genital images in our previous classification study
(Ponseti et al. 2012) we found a total number of three
misclassified participants. However, this was based on
twice the number of observations per participant (35
whole body images and 35 genital images per condition). In order to better compare the discriminative
power of face, body and genital stimuli we re-analysed
these data by submitting the fMRI data in response to
body images and the fMRI data in response to genital
images to our classification procedure separately. After
cross-validation of the BOLD responses to body images
we found seven misclassifications (five false positives).
The classification of BOLD responses to genital images
resulted in 18 misclassifications (10 false positives).
43
Table 1. The discriminative pattern underlying paedophilic versus control group classification*.
Brain area
Paedophiles4controls (girlswomen)
Hippocampus
Fusiform
Fusiform
Cerebellum
Lingual gyrus
Cerebellum
Thalamus
Inferior occipital gyrus
Lingual gyrus
Middle occipital gyrus
Thalamus
Superior frontal gyrus
Anterior cingulate cortex
Anterior cingulate cortex
Superior frontal gyrus
Supplementary motor area
Superior parietal lobule
Paedophiles4controls (boysmen)
Fusiform gyrus
Cerebellum
Inferior occipital gyrus
Middle occipital gyrus
Inferior parietal lobule
Precentral gyrus
Inferior occipital gyrus
Fusiform gyrus
Inferior temporal gyrus
Putamen
Side
L
R
L
L
R
R
R
R
L
L
L
L
R
L
L
L
39406
L
L
L
L
L
L
R
R
R
L
3200
1633
184
268
986
1405
292
t value maxima
z score maxima
x
8.41
8.20
7.52
7.43
7.05
6.89
6.88
6.87
6.65
6.36
6.34
5.25
4.84
4.59
4.49
4.88
4.59
6.65
6.54
6.16
6.10
5.88
5.78
5.77
5.77
5.63
5.45
5.43
4.68
4.38
4.19
4.10
4.41
4.19
14
34
38
34
14
30
20
32
8
34
16
12
10
4
14
10
26
4
70
46
66
38
50
28
84
82
70
12
50
36
30
52
14
62
16
12
20
12
4
22
0
8
8
16
14
22
22
24
24
62
54
5.67
4.88
4.63
4.55
4.31
5.25
5.41
5.16
4.89
4.79
4.98
4.41
4.21
4.15
3.97
4.68
4.80
4.62
4.42
4.34
42
30
42
38
26
36
42
40
46
12
68
46
72
60
48
2
68
58
56
14
14
22
4
0
50
54
8
18
16
8
Figure 2. The classification of paedophiles and healthy controls using individual expression values according to Fishers linear
discriminant analysis. Participants with Ppaedophilic40.50 (dark area) were classified as paedophiles. Note, individual expression values
of those sexually attracted to female faces (heterosexual teleiophiles and heterosexual paedophiles) differed mostly along the y-axis
(girls5women comparison) and the participants who were sexually attracted to male faces differed mostly in regard to the x-axis.
Heterosexual, heterosexual teleiophilic participants; homosexual, homosexual teleiophilic participants.
44
J. PONSETI ET AL.
Discussion
To our knowledge, this is the second study applying a
neurofunctional pattern classification to assess paedophilia. In our first study (Ponseti et al. 2012), we
employed this type of pattern classification algorithm
to distinguish participants with paedophilia from healthy
controls based on the between-group differences in
functional brain response to sexual stimuli (with three
out of 56 participants misclassified). Analysing the same
group of participants, we show here that our pattern
classification algorithm performed similarly well when
classification is based on brain responses to face images
(with four out of 56 participants misclassified). In
comparison to our previous classification study, the
present classification results show a de crease in
specificity (91 vs. 100%) and an increase in sensitivity
(95 vs. 88%). Further, mean classification accuracy of our
classification approach based on brain responses to
facial stimuli is somewhat superior as opposed to the
phallometric assessment (Freund et al. 1989; Freund and
Watson 1991). In the present study, most of the false
positives were homosexual teleiophiles. This might
account for the smaller group differences in the boys
vs. men comparison in contrast to the girls vs. women
comparison (Figure 1A vs. 1B).
The aetiology for paedophilia remains unresolved.
Previous brain research in paedophiles failed to show
congruent structural brain differences between paedophiles and teleiophiles (Schiffer et al. 2007; Schiltz et al.
2007; Cantor et al. 2008; Poeppl et al. 2013). So far,
functional MRI studies also provided little to the
understanding of this paraphilia: When exposed to
sexually preferred stimuli paedophilic probands
showed brain activity in rather the same areas as
teleiophilic probands did (Schiffer et al. 2008a, 2008b;
Poeppl et al. 2011; Habermeyer et al. 2013). Thus, brain
structure as well as the processing of sexual stimuli
appears to be similar in paedophiles and teleiophiles.
How does a pattern recognition algorithm work reliably
under these circumstances? The human brain shows an
extended and very fast response to emotional salient
stimuli, like fearful faces (Vuilleumier and Pourtois 2007).
In case of sexual stimuli (as a special type of emotional
stimuli) this kind of fast and extended brain response is
preference specific, i.e. generally it can be observed
only when the sexual stimulus matches the sexual
orientation of the observer (Ponseti et al. 2006). Given
the extended and most probably automatic brain
response to sexually preferred stimuli (Childress et al.
2008; Gillath and Canterberry 2012; Oei et al. 2012), fMRI
pattern classification does a rather easy job. Previous
research demonstrated that not only sexual stimuli,
but also facial stimuli evoke brain responses according to the sexual preferences of the observer
(Kranz and Ishai 2006). Recently this was shown in
paedophiles as well (Ponseti et al. 2014). We therefore
expected promising results in case of pattern classification based on BOLD responses to facial stimuli.
Facial stimuli might even be better suited for classification of sexual age preferences than sexual stimuli
given that the present classification analysis was only
based on half of the number of observations used in our
previous study. In fact, when holding the number of
observations constant, face images did better than body
images and much better than genital images. Perhaps
faces contain more age-relevant cues than whole body
views or genital images. Faces trigger a specialised
network of age detection, which is common in
teleiophiles and paedophiles (Ponseti et al. 2014). This
network involves core areas of the human face processing system. Group differences in response to facial
stimuli (which we used as discriminative pattern for
our classification procedure; see Figure 1 and Table 1)
showed that face processing also involves brain areas
that have been found to be involved in sexual processing (e.g. thalamus, cerebellum, superior parietal cortex,
inferior temporal cortex) (Stoleru et al. 2012). This link
between facial and sexual processing (see also Kranz and
Ishai 2006) probably allows for the successful classification of sexual age orientation based on facial stimuli.
The findings of this study are limited: (i) by the fact
that our paedophilic participants admitted their sexual
orientation and we therefore do not know whether
classification based on facial stimuli may be falsified by
manipulation of non-admitting paedophiles and (ii) by
the fact that we do not know the test/retest reliability.
Moreover, the decreased specificity of our face-based
classification approach raises the question whether
individuals who are frequently exposed to childrens
faces (e.g. parents, teachers) are more likely to be
misclassified than others. Once these questions are
resolved by future research, the fMRI classification based
on face images may be a promising technique to assess
the sexual orientation of child sex offenders as it
circumvents the need to present child sexual images.
Finally it has to be emphasised that the pattern
classification algorithm which we used in the present
study shows a promising ability to detect paedophilia but
it does not detect whether a paedophile will act on his
urges or not. Paedophilia is a major risk factor for sexual
child offending; however, there are paedophilic men who
effectively refrain from offending children sexually.
Meanwhile there is evidence for a potential to enhance
behavioural control and reduce associated dynamic risk
factors in self-motivated paedophiles in the Dunkelfeld
(Beier et al. 2015). Hence there is a growing understanding of the psychological factors which enable paedophilic
men to successfully control their sexual behaviour but
no comparable progress in knowledge with respect to
the underlying neurobiological factors. If these factors
were to be identified by future research, the clinical
relevance of fMRI pattern classification would lie in
considering the individual classification result as predictors in assessment and treatment.
Acknowledgments
Statement of interest
HRS has received honoraria as speaker from Lundbeck A/S,
Valby, Denmark, Biogen Idec, Denmark A/S, Genzyme,
Denmark and MerckSerono, Denmark, honoraria as editor
from Elsevier Publishers, Amsterdam, The Netherlands and
Springer Publishing, Stuttgart, Germany, travel support from
MagVenture, Denmark, and grant support from Biogen Idec,
Denmark A/S.
References
Association AP. 2000. Diagnostic and Statistical Manual of
Mental Disorders, Fourth Edition, Text Revision. Washington,
DC: American Psychiatric Association.
Beier K, Ahlers CJ, Goecker D, Neutze J, Mundt IA, Hupp E. 2009.
Can pedophiles be reached for primary prevention of child
sexual abuse? First results of the Berlin Prevention Project
Dunkelfeld (PPD). Journal of Forensic Psychiatry &
Psychology 20:851867.
Beier KM, Grundmann D, Kuhle LF, Scherner G, Konrad A,
Amelung T. 2015. The german dunkelfeld project: a pilot
study to prevent child sexual abuse and the use of child
abusive images. J Sex Med 12:529542.
Blanchard R, Klassen P, Dickey R, Kuban ME, Blak T. 2001.
Sensitivity and specificity of the phallometric test for
pedophilia in nonadmitting sex offenders. Psychol Assess
13:118126.
Cantor JM, Kabani N, Christensen BK, Zipursky RB, Barbaree HE,
Dickey R, et al. 2008. Cerebral white matter deficiencies in
pedophilic men. J Psychiatr Res 42:167183.
Childress AR, Ehrman RN, Wang Z, Li Y, Sciortino N, Hakun J,
et al. 2008. Prelude to passion: limbic activation by unseen
drug and sexual cues. PLoS ONE 3:e1506.
Finkelhor D, Turner H, Ormrod R, Hamby SL. 2010. Trends in
childhood violence and abuse exposure: evidence from 2
national surveys. Arch Pediatr Adolesc Med 164:238242.
Freund K, Watson R. 1991. Assessment of the sensitivity
and specificity of a phallometric test: an update of
phallometric diagnosis of pedophilia. psychological assessment 3:254260.
45
46
J. PONSETI ET AL.
orgasm in healthy men and women: a review and metaanalysis. Neurosci Biobehav Rev 36:14811509.
Vuilleumier P, Pourtois G. 2007. Distributed and interactive
brain mechanisms during emotion face perception: evidence from functional neuroimaging. Neuropsychologia
45:174194.