New Machine Learning Algorithm Can Identify IQ Just by fMRI Scans

The technology with the ability to scan people’s brains to predict whether they have a propensity of commit crime have long been a topic of dystopian science fiction.  Though the innovation to do this is still a ways off, each emerging discovery seem to show a progression towards this. Just this month Canada’s University of Alberta had used machine learning to identify people with schizophrenia by analyzing MRI scan with 78 percent accuracy. Now machine learning has been taken a step farther, as it can now predict our IQ scores with just fMRI scans.
 
A team comprised of researchers from America’s California Institute of Technology, Cedars-Sinai Medical Center and the University of Salerno have developed a machine learning algorithm that can successfully predict the IQ scores of several hundred patients. The team acquired IQ scores and brain scans from nearly 900 subjects from the Human Connectome Project for their machine learning algorithm.
 
Each subject were in a resting-state fMRI scan, meaning the patients did not need to do anything while being scanned. Professor Ralph Adolphs explains, “We found if we just have people lie in the scanner and do nothing while we measure the pattern of activity in their brain, we can use the data to predict their intelligence.” The patients had taken 10 different cognitive ability tasks such as working memory tests, oral recognition tests, sequence memory tests, and control and attention tests, to name a few. By testing the patients on a wide range of cognitive abilities, it allows the researchers a more precise measure of the patient’s intelligence.
 
The fMRI creates a 3d model of activity in the brain by using a magnetic field and radio waves to detect the changes in blood flow from various areas of the brain. The advantage of fMRI from technology such as electroencephalography or EEG is that the EEG can only detect electrical activity at the surface level while the fMRI can get a picture deep within the brain. Although the detection in change in blood flow works as a proxy for activity for neurons it is still blind to the exact number of neurons being fired as well as a large amount of noise in the data.
 
It is also important to note that although the intelligence of an individual is stable over time, the actual cause of intelligence is hard to find a definitive correlation. The team’s report notes “individuals may score identically on an IQ test by using different cognitive strategies, or different brain structures.” Because of this it has not been possible for human experts to parse out predictable patterns with fMRI brain images.
 
This is where our old friend machine learning comes. To give a birds eye view of how it works, researchers gave the network IQ scores and the brain scan images. By feeding several hundred of these, the neural network was able to find statistical correlations between brain scans and IQ scores, i.e. that there were certain patterns in how to brain was wired that correlated to what the IQ score would be.
 
Along with this study, the research team also conducted another study in parallel using the same pool of subjects to find predict personality traits through fMRI brain scans. They tested personality on five attributes:
“

1. Openness to experience: Preference for new experiences and ideas vs. preference for routine and predictability
2. Conscientiousness: Self-discipline and thoughtfulness vs. spontaneity and flexibility
3. Extraversion: Sociability and talkativeness vs. shyness and reservation
4. Agreeableness: Friendliness and helpfulness vs. antagonism and argumentativeness
5. Neuroticism: Confidence and predisposition to positive emotions vs. nervousness and predisposition to negative emotions

“
Perhaps unsurprisingly, it did not yield any conclusive results. Professor Dubois reported: “The personality scores in the database are just from short, self-report questionnaires; That's not going to be a very accurate measure of personality to begin with, so it is no wonder we cannot predict it well from the MRI data.”
 
IQ on the other hand changes very little in weeks, months or even years and being tested in an objective manner gives it a more concrete standard to be able to test upon. When thinking about the applications for this, one may come to the idea of using it to determine the aptitude of students before they go into schooling to determine what type of teaching style is best suited for the students or if they are suited to skip grades. Perhaps along with standardized testing, such screening would be utilized. It is still just speculation but the possible implications are potentially heavy.

Original Article: http://www.caltech.edu/news/caltech-scientists-can-predict-intelligence-brain-scans-82675