A newly-published review of neuroscience research looking at the predictive value of functional and anatomical imaging raises interesting questions about the role of such studies in learning, psychiatric treatment, and even the treatment of criminals. “Prediction as a Humanitarian and Pragmatic Contribution from Human Cognitive Neuroscience” by Gabrieli, Ghosh, and Whitfield-Gabrieli and published in Neuron, does a thorough job of explaining the current state of the research and pointing to where future research is needed.
The basic idea is to use noninvasive imaging to look at the structure or function of the brain as a way of predicting future behavior, and then using those predictions to help guide treatment and education interventions, and perhaps decisions regarding parole or further treatment of criminal behavior. This concept raises many issues, including the technology being used, the state of the research, the ultimate potential for this line of research, and ethical considerations.
The major question underlying this entire endeavor is, to what extent is brain anatomy and function destiny?
There are two basic ways to look at the brain to determine its functional potential with regard to specific tasks (such as reading, impulse control, tendency toward depression, etc.). The first is to look at anatomy independent of any current task being performed. Essentially this approach uses MRI scanning to measure the thickness or overall size of gray matter regions or of white matter tracks. If, for example, the “cable” that connects the two main language processing areas (called the arcuate fasciculus) is thicker, this may correlate with a greater ability to learn a new language or improve reading skills.
The advantage of this approach is that the technology is highly reliable and precise. This is simply a physical measurement.
The second approach is to look at brain activity during a specific task. There are several options for this approach: fMRI scanning (which measures blood flow) has a good spatial resolution but a poor temporal resolution, while looking at electrical or magnetic activity through EEG or MEG respectively has better temporal resolution, but worse spatial resolution.
These approaches have the advantage of looking at actual brain activity – seeing which parts of the brain are active during a specific task. They are currently less precise than purely anatomical techniques. They also have the added difficulty of being dependent on the subject performing the task that they are being instructed to perform (such as attempting to match words that rhyme). The researchers cannot know for sure how focused a subject is on the target task, or what mental method they are using to achieve the task.
Taken together, these techniques are relative new and powerful tools for looking at brain anatomy and function. They are detailed enough to be useful, but still have significant limitations.
The authors do an excellent job of reviewing the logic and methods behind research looking into using imaging technology to predict future outcomes (although their paper can get very technical at times). I will give a simplified summary here. Essentially, the first phase of such research is to look for correlations between some brain attribute and some outcome measure (such as reading skill or level of depression). Researchers can look for correlations within a group, between groups, or with change from baseline.
They correctly caution that correlations do not necessary equal causation. They further point out that such correlations are always overestimates, and that the more variability there is in the brain regions and markers being measured, the greater the ability to tease out some correlation (to find some apparent signal in the noise). Correlations therefore need to be confirmed by making predictions with a fresh data set.
They outline various statistical methods for creating a model based on correlations and then testing the model by using it to make predictions about a different data set.
The bulk of the paper then reviews existing research in various areas. Overall existing research is mostly preliminary. Most studies are small in size and look mainly at correlations. Few have done the follow up of testing their models with predictions using fresh data. Therefore, while this may be an exciting area of research, the published studies have not yet matured to the point where we can make practical use of the results.
What the current research does show is that there is a modest but real correlation between brain imaging and various cognitive outcomes. The data is most robust and convincing with respect to language. This makes sense, since there are brain structures dedicated to language processing, and the robustness of this processing is likely to correlate to language ability.
Another area where the authors point out the research is fairly solid is depression, for example looking at brain responses to depressed faces and outcomes of treatment for depression. I also assess the research looking at impulse control and the frontal lobe structures that correlate with such control as being fairly solid.
However, even in these areas where there is a clear correlation, the amount of predictive value tends to be fairly modest. In many of the studies reviewed, positive imaging predicts only 20% or so of later variability. In a study of reading outcomes, traditional ability measures predicted 65% of later outcome, while brain imaging predicted 57%. However, when the two types of information were combined, predictive value increased to 81%.
Future research and applications
The authors correctly point out that if this line of research is to be useful clinically then the research needs to progress to much larger studies which look at the predictive value of models based on correlations. In addition, we need to get more and more detailed in terms of correlating specific brain structures or activity and specific clinical outcomes.
I do wonder, however, what the ultimate limit will be for this line of research. The brain is a complex system, with many different tasks and abilities interacting in a complex way to yield an end result. There may be some low-hanging fruit to pick, such as with reading and language skill. With such tasks there may be a tight correlation between the organization of the brain and ultimate ability. For other more complex outcomes, such as criminality, there may be significant limits on how predictive such approaches can be.
It may be necessary to take the research farther still – looking at many different parts of the brain and then computer modeling their interaction. We may still get to the limits predicted by chaos theory, however. We can’t predict weather accurately beyond 5 days or so, and we may not be able to predict human behavior and outcomes beyond a certain limit also.
In addition there is the issue of the relative contribution of brain anatomy compared to plasticity (the ability of the brain to change its wiring), environment and learning. We are likely only looking at potential by looking at the brain. This of course will have a statistical predictive value, but this does not mean that brain anatomy is destiny for an individual. If we can’t apply this data to an individual (only statistically to groups) then perhaps the vision of the authors will never be fully realized.
This issue blends into the ethics of looking at the brain in order to determine future outcomes. Should a criminal be denied parole because his brain imaging shows relatively-low impulse control, which predicts a higher probability of recidivism?
Where this type of information can be useful (and the authors do point this out) is adding predictive information to help guide treatment decision in certain psychiatric disorder and education interventions. If we can identify at an early age those students who will have difficulty with language and reading, then they can be given greater resources in that area to help them keep up with their peers. This is already being done, using standard testing, but the evidence suggests brain imaging might add to the predictive power of such testing.
In the clinical setting brain imaging may help predict who will respond better to one type of drug over another, or to cognitive behavior therapy. Medical interventions are largely based on statistical data with large groups, and this approach would be no different.
I largely agree with the authors that the new wave of brain anatomical and function imaging has brought with it a new age of understanding the brain. I also agree that there is tremendous potential to use this type of imaging to guide interventions for education and psychiatry. Applications to criminal justice are more complex and fraught with ethical considerations.
We are, however, still years away from practical applications. We need, as the authors suggest, to do large studies of predictive value. But then we also need, in my opinion, to do clinical outcomes research – looking at the net outcomes from employing this type of data in real world situations.
One fear that I have (and the authors also point this out) is that the allure of such data will give it a mystique that goes beyond its real world applicability. If people think we can peer into someone’s brain and predict their future, there would be the strong temptation to treat brain scans as if they were destiny, and short circuit more thorough and nuanced methods for evaluating individuals and optimizing interventions.
In the end I think that this approach will be one more tool in our toolbox, and it can be a powerful tool. Such data will still need to be used thoughtfully, with a full appreciation of its limitations.