Welcome back to Reading the Research, where I trawl the Internet to find noteworthy research on autism and related subjects, then discuss it in brief with bits from my own life, research, and observations.
Today's article is interesting, but badly in need of a reality check.
So, first thing. What is a brain scan? I'm going to grossly simplify, because I presume you value your time and don't want to read a detailed breakdown of the difference between a PET, MRI, and fMRI scan.
A brain scan is like an x-ray, but specifically for your brain. Like an x-ray, it's a snapshot, more or less a picture at a point in time. Where an x-ray shows your bones (and organs, if you know where to look)...
a brain scan shows brain activity.
A trained x-ray technician can look at the x-ray and see cracks in your ribs that explain why you've been having chest pain, or a bulge in one of your spinal disks that explains your lower back pain.
A trained fMRI technician can look at your brain scans, compare them to what little we know about "normal brain function" and note abnormalities.
You might notice I've used specific examples for x-rays, and air quotes and more dubious language for fMRIs. That is very much on purpose. The fMRI technique was invented in 1990, which you'll note was about 30 years ago. The discovery of x-rays was in 1895. In the grand scheme of technology, fMRI is a baby.
A very promising baby, likely with a bright future ahead of it, mind. But a baby all the same. What it might become is unknown, and much of what it can do is theorized and speculated upon.
What does this have to do with the article? Well, here's the thing. The researchers have tossed a thousand or so autistic brain scans at a computer (actually 14 different analysis models, but it's simpler to say "a computer"), and hope to have the computer not only be able to tell which brains are autistic, but explain how it came to that conclusion and present personalized prognoses. That's actually a pretty small sample size for something like this, but it was the convenience sample- the database of these scans is open source and free for research purposes.
This is very much akin to taking that baby, handing them roller skates, and having full expectations of brilliant dance skating.
That's not to say that brain scans can't be exceedingly helpful to psychiatry and even autism treatments in specific. One of the problems is that autism itself isn't well-defined, and as such, any program looking for it is going to be subject to that ambiguity. The study itself even acknowledge that the Related Work section. We know there are some parts of the brain that are associated with autism, but there are many of them. And with no subtypes of autism, saying "this part affected this way means the person will have an increased chance for depression" is just guesswork.
Even if you specifically define autism as only a set of brain differences, you still come out with a metric ton of those differences. And they are not cookie-cutter. You cannot slap 5 autistic brains side by side and expect there to be much by way of commonalities between them. Grandin's book (review linked above) was pretty clear about that.
The real kicker? Brains change. Neuroplasticity is a long word that basically means dogs of any age can learn new tricks. Connections between brain parts can strengthen or weaken (that is, in fact, how LENS works). Any given part of a person's brain might develop more strongly if their circumstances warrant it. When I said brain scans are a snapshot, I quite literally meant it. A brain scan is representative of the brain at the time it was taken. It's silly to say we could give a long-term prognosis based on a single snapshot.
The researchers here hoped that if they throw enough data at the computer, it'd spit out something useful. That's not entirely unprecedented. But it's overly optimistic to say it'd be conclusive. The Results section is pretty dense, but the gist is that they discussed their results with the various models. They weren't what the team were hoping for, but they remain hopeful.
The TL;DR: brain scans are cool and very promising, but analysis of them is prone to human error. Also, autism is not well-defined enough for this line of inquiry to yield good results. Also also, if you're going to throw data at computers, you need a LOT of it. A thousand or so pictures doesn't cut it.
(Pst! If you like seeing the latest autism-relevant research, visit my Twitter, which has links and brief comments on studies that were interesting, but didn't get a whole Reading the Research article about them.)
No comments:
Post a Comment