AI (Hopefully) Will Be Able to Diagnose ASD Soon
For those of you who aren't aware of what ASD (or Autism) is, it's a developmental condition that affects a person's social skills, communication, and relationships. And as of today the way to diagnose ASD is to take a series of tests that involve testing the areas that ASD affects. Every clinician tests kids differently, though there are some more standardized tests, and the process is long and arduous for both parents and kids.
However, a new Stanford AI has shown progress in what could be a new way of diagnosing ASD in a definitive manner. The AI uses fMRI scans can detect the ASD "fingerprint" in the brain. The initial study had a sample size of about 1,100 brain scans and from them, the AI was able to correctly identify the scans that were people who were clinically diagnosed with ASD with about 82% accuracy. This is one of the first points of evidence for a definitive biomarker for ASD and which once acquired will allow for a faster and more accurate diagnosis of ASD.
The Stanford AI used an XAI model instead of a normal AI. They then had the AI scan three specific parts of the brain that have shown relation to ASD. These areas were the posterior cingulate cortex and precuneus, which are part of the default mode network; the dorsolateral and ventrolateral prefrontal cortex, which are involved in cognitive control; and the superior temporal sulcus, which processes human sounds. Using these parts of the brain the AI then differed the scans with ASD from the neurotypical scans. With this AI, Stanford hopes that they could start diagnosing ASD as early as 6 months.
Standford actually released another study earlier this year focusing on ASD and fMRI. While the above study was focused more on differing neurotypicals from those with ASD, the earlier study focused on seeing if the AI (also using the XAI model) could use the fMRI's to tell the difference between boys diagnosed with ASD and girls diagnosed with ASD. The sample size for this study was slightly smaller using a total of 773 scans, 637 from boys and 136 from girls. Given the fact that girls get diagnosed with ASD significantly less than boys do (due to a variety of factors such as less research being done on how ASD affects girls), the fact that Stanford acquired 136 fMRI scans is impressive.
The AI was able to tell the difference between the sexes with about 86% accuracy and when given 976 neurotypical fMRI scans, the AI could not tell the difference between the sexes at all. What this tells us about ASD and the AI that was scanning them is that there are unique brain differences between the sexes in children with ASD that aren't found in neurotypical children. The AI was looking for the specific differences between the two sexes using these unique markers and henceforth was unable to detect them in the neurotypical scans.
These two AI together will make diagnosing ASD significantly more efficient and possibly even help better understand how ASD affects girls. The thing about ASD is that the earlier the diagnosis, the faster the child can get help. And the faster the child gets help, the more effective the help is for the child so that they can live their best most adjusted life. Hopefully, these AI can do more testing and soon the old ways of testing ASD will be gone and a more effective method will be in its place
