Brain representations of social thoughts predict autism

Post Reply New Topic

here are the links for the original sources of both articles, if you want to look for the methods:

http://www.plosone.org/article/info%3Ad ... ne.0113879

http://brain.oxfordjournals.org/content/134/8/2422

That point is considerably blunted by the sample size and assumptions made in the conclusion as well as in the design of this experiment.

Each research study has flaws that the researchers themselves recognize.
Often, the media blows up research studies into big claims that the researchers aren't making and wouldn't make.
Most researchers I know will go so far as to say what is new, interesting, and important about a study, and what understanding or applications it could lead to.
Most research in cognitive neuroscience related to autism is about finding clear differences between autistic people and neurotypical people and identifying subgroups of autism.

I agree, InThisTogether: the distinction of subtypes is a really important issue, and Just doesn't seem to say anything about it.

Suppose that this is the case: some ASD people rely principally on visual-spatial strengths, while the clinically defined Asperger subgroup, tend to have maximised language-based strengths. Some have speech delay, and some have lower or higher vascularisation in the brain (which affects MRI results).

So just rounding up 17 high functioning people doesn't do what the researchers claim it does. This seems straightforward to me, though obviously there are people who consider this viewpoint irrelevant.

What is interesting to me is not the 17 people representing the autistic population, but machine learning to analyze fmri data and distinguish between two groups with high accuracy, even in small sample, as one must start somewhere. As I said earlier, this is not common in research on autism, as there is generally a lot of overlap between nt and asd groups on whatever is measured. Being able to develop tasks that do find clear differences in small sample is important before eggstending to larger samples. Even if small sample is not representative of large sample, it could be a subgroup of autism spectrum. New data analysis techniques for large, complex datasets like EEG, fMRI, and eye-tracking are just as important for finding differences.

It seems to me that it would have been better for Just to refer to it as a pilot study then.

Calling it a study is fine, the term pilot study has a different meaning which is that the researchers are still testing the eggsperimental setup or code or trying to work out a data analysis chain that makes sense for the topic studied. It is much more preliminary than an ackshul study involving 30+ participants. All researchers know that published studies have flaws and limitations, and results of published studies are subject to further investigation and potential refutation, that is the nature of the scientific method.

Also, researchers doing brain imaging are aware of technical issues in brain imaging and do question brain imaging results, e.g. if certain types of artifacts contributed to the data and if the results are due to artifacts instead of the phenomena of interest. They also work on methods to address these issues, so sometimes, previous datasets will be re-analyzed using new and improved methods. Most researchers are pretty paranoid about these issues.

failure posting

failure posting

It's ok for there to be a range of views on this topic, as nothing about this field of research re autism is yet written in stone. It's too soon to make definitive conclusions. I think there may have been a rather hasty readiness to ignore the confounding issues, assume that they must have been controlled for (?evidence) and/or a tendency to brush them off as unimportant or irrelevant.

My perspective is a cautionary one.

Here is one more commentator who also expresses some doubt:
http://crackingtheenigma.blogspot.co.nz ... -five.html

Doubt is good, at every stage of the experimental process particularly when evaluating new claims.

A single fMRI test to diagnose autism is a bad idea.
It is better to use multiple behavioral tasks on a computer to supplement and inform current diagnostic procedures.

97% accuracy claim in another context:

http://www.theguardian.com/science/sift ... er-matters