It was an fMRI study of 24 self-identified paedophiles (recruited through a clinic offering anonymous treatment) and 32 male controls. Everyone was shown a series of images of naked men, women, boys and girls. The neural response to child vs. adult images was the main outcome measure.
Respect to the authors for getting that past the ethics committee.
However, what's more interesting is that by comparing each individual's brain activity to the average activity of the paedophile group and the control group, it was possible to diagnose people as paedophiles or not with high accuracy (90+%).
The statistics seem kosher: they used leave-one-out cross-validation to avoid the error of double dipping.
What's not clear is whether this was measuring sexual attraction as such. All it's measuring is how much each person's activity correlated with the paedophile group average. Maybe it's picking up on the shame paedophiles feel over being reminded of what they've done. Maybe the controls were just averting their eyes when the child porn came on.
However, you could say that if you're just interested in the practical business of catching paedophiles, that's academic. More concerning is the question of whether it would be possible to fool the technique. A recent study showed that it's easy to fool a brain scan designed to detect lying.
But let's suppose it does work out. Would that be a good thing? What is "a paedophile", anyway? Is it someone's who's attracted to children, or someone who acts on that attraction?
For example, there are people who are caught with child porn, and who admit they downloaded it, but who deny being attracted to children. The Who shredder Pete Townsend and comedian Chris Langham being two British examples. Both admit downloading illegal images, but say it was for 'research purposes'.
Now it might be possible, using fMRI, to find out if they're telling the truth. Let's suppose it was doable.
So what? Downloading child pornography is a crime - whatever your motivation. Being attracted to children is legal, in itself. So from a legal perspective it should make no difference at all in cases like this.
Of course, we don't in fact go around seeing things from a purely legal perspective. We care whether someone is attracted to children or not. But should we care? Is that fair? You don't choose your sexual orientation. What you choose is whether to break the law by commiting the crime.
There are surely people out there - no-one knows how many - who are attracted the children, and never act on it. Do we want to be able to "catch" them?
Edit: The original version of this post linked to the wrong paper, an older paper by the same authors. This has been fixed now.
Ponseti, J., Granert, O., Jansen, O., Wolff, S., Beier, K., Neutze, J., Deuschl, G., Mehdorn, H., Siebner, H., & Bosinski, H. (2011). Assessment of Pedophilia Using Hemodynamic Brain Response to Sexual Stimuli Archives of General Psychiatry DOI: 10.1001/archgenpsychiatry.2011.130
22 comments:
Nice post. Thanks for that.
Again a reason why fMRI isn't a valid tool. As you so correctly stated nothing can be concluded from the 'blobbogram' but other then that the brain reacted. What that reaction was will be in the eye of the beholder.
Disgust, shame, lust, pure intellectual interest, intent, whatever. The blobbogram won't tell you.
It's about time to scrap this machine or sell it to mindreaders and mystic mediums.
Genuine question (somewhat related to the post):
The whole leave-one-out mularkey is standard practice in all of these diagnosing X with (f)MRI studies. Here X is paedophilia.
Typically all the participants are included and the threshold optimized to get a headline sensitivity / specificity measure.
Then use the leave-one-out method to show that it generalizes. But it's only remotely useful if it generalizes at high levels of specificity or sensitivity. And leave-one-out doesn't tell you that. Just that it generalizes above chance.
Please tell me I'm wrong. Then I'll stop getting annoyed with all these papers!
Hi,
interesting post, thanks. The link is to the wrong paper (seems to point to a previous paper by the same authors rather than the recent archives of general psychiatry paper you discuss in your post).
Given that (I presume) sexual arousal is associated with cardiovascular changes, I wonder what the activation maps in this paper really represent? The regions of different activity seem to cluster around the lateral ventricles/mid-brain which is where you often see the effects of cardiovascular (e.g. heart rate) confounds. It would be interesting to know what happens to these results if you covary for heart rate/BP. I guess the functional interpretation of the blobs is not central to the authors' argument, that the data is able to classify group membership-- however if these effects are driven by the cardiovascular changes of arousal, fMRI is probably a fairly expensive way to measure them.
Mike
MB: Thanks for tipoff about the paper mistake, I got fooled by the similar titles.
Physiological confounds are another possibility. To interpret these results you'd want to do eye tracking and physiological monitoring.
The "so-what?" response is crucial here. Free societies don't scan brains to obtain evidence of commission of crimes. Crimes are acts. Blobs on fMRI are, well, blobs.
Following up on rob lindeman's comment, it seems to me that one problem with thinking about bringing this sort of analysis into the legal sphere relates to people who are aroused by images of naked children but do not act upon those urges. It seems likely that (a) the attraction is not all-or-nothing, and that (b) some people are better able to manage their urges than others. The fMRI results would seem unlikely to be able to distinguish between people who will act and those who won't. As a result, this is not viable as a screening tool.
But certainly interesting.
Jon: Leave one out does tell you how well it generalizes. In this case the mean accuracy (sensitivity + specificity / 2) ranged from 89 to 95% using the Fisher linear discriminant.
K-Nearest Neighbor was 75-91%.
Chance would be 57% (I think) - given that there were 24 paedophiles and 32 controls so just guessing that everyone was a control would give you 32/56 = 57%.
One thing I didn't mention though is that the scatterplot does not seem to have use the leave one out procedure, so it's a bit optimistic.
Interesting research on brain activation. However, the difference between mere attraction to children and acting on that attraction and actually victimizing children points to how this method could be abused by the legal system. Pedophilia doubtless contains, along with attraction, a large element of lack of impulse control and/or a sociopathic lack of concern with the child's welfare.
Thanks Neuroskeptic.
So in this case, it does seem to generalize quite well (all other caveats notwithstanding).
But for the next paper I read, K-Nearest Neighbour is what I should look for to see how well it generalizes?
The papers I've read recently claiming they can "diagnose" autism using MRI scans either just report p-values for the leave-one-out (Ecker et al., 2010, J Neurosci) or don't bother doing it at all (Uddin et al., 2011, Biol Psychiat).
Anyway, good to know. Thanks for the learning!
Interesting stuff, especially the implications of the application of technology like this. Orwellian/discriminatory on one hand, protective towards predatory behaviour on the other.
Interesting, but I agree that conclusions can't be drawn with regard to actual behaviours, but it's probably prettttty close.
Hi. I just found your blog and fell in love with it. Hope for more interesting blog posts in the future. /Amanda, Swedish Skeptic och journalist
Thanks!
Neuroskeptic wrote:
"However, you could say that if you're just interested in the practical business of catching paedophiles, that's academic."
If the MRI scanner and associated EPI time-series are functioning as a very expensive motion detector then it would be more than of academic interest. There are less expensive ways of detecting motion.
It will be a problem when they will use brain scans to approve your health insurance, during job interviews and stuff like that. One might even be denied entrance to swimming pools.
One can even imagine a future society where brain scans will be mandatory and certain brain patterns will earn you a mandatory "pedophile star" on your coat.
Where this would come in to play is when someone is accused of molesting a child. The outcome of a this type of evidence could very easily push a jury one way ("She couldn't have done this... look! her brain scan shows she isn't attracted to kids!") or the other.
The sample doesn't really focus on the attraction to girls, but seems to be largely focused on men attracted to boys. The perceived dangers and fears of paedophiles in current times tends to be associated to the sexual abuse of girls. What was the gender and nature of the images shown to subjects? Was there any gender difference? Studies like these can conflate the domains in which paedophile acts occur. There only seems to be one heterosexual paedophile in the sample? Lots of homosexual paedophiles in the sample? They do not help us to understand where dangers exist most and for whom? Are the self identified paedophiles individuals who have physically abused children? Do they have children in their care? Where are children most at risk?
To my eye these BOLD maps look like they could easily be due to motion. The motion correction (registration) algorithms that are used in this and most fMRI studies work poorly. These algorithms depend upon an assumption that the volume of 2D slices (not simply the head) behave as a rigid body. When motion occurs it occurs at anytime during the acquisition of a volume. This motion causes the 2D slices to shift relative to one another and thereby violate the rigid-volume assumption. This will lead to poor estimates of motion and flat out wrong corrections.
This paper does not mention which type of head coil was used. Multi-channel phased array head coils are the norm in medical MRI scanners these days. The method used to obtain a composite image by combining the individual images from each channel have many implications when motion is present. These facts make all fMRI suspect - not just this study - and this is even more so the case if temporal interpolation (so that all slices can be taken to be sampled simultaneously) is used.
Simply put, the temporal interpolation and motion correction, taken as separate operations upon the data, do not commute. A specific ordering (for example temporal interpolation first) does not fix the problem because the problem is that the operations can not (except for special cases) be taken to be separate. All such corrections should be made simultaneously. To do this will require a lot of work. Until this is done all fMRI studies that do not use reliable methods of head constraint should be highly suspect.
Interesting discussion and I agree with your conclusion. I'd go further and claim that if people with paedophilic affects were shown more understanding and support and less stigma, they'd be better equipped and motivated to act lawfully, with empathy and restraint. I think perhaps paedophilia is as misunderstood today as homosexuality was in the 1950's.
As you said: <>
People attracted to adults also have limits placed on their sexual expression. A capacity for empathy and the ability to live within social mores is a consequence of socialization, not of sexual orientation.
Sean
[correction] ..As you said: =We care whether someone is attracted to children or not. But should we care? Is that fair? You don't choose your sexual orientation. What you choose is whether to break the law by committing the crime.=
Well said.
Sean.
Attn: Anonymous on 7 October 2011 20:25.
Re. comments on array coils and motion, I'm working on this very issue at the moment. You are clearly informed, do me a favor and drop me a note with a contact email? I'd like to discuss the (few) possibilities.
Cheers!
Ben (aka PractiCalfMRI)
Post a Comment