Functional magnetic resonance imaging (fMRI) visualizes the blood oxygen level–dependent (BOLD) signal in the brain, which indexes neural activity. However, this signal is noisy and variable. Therefore, most studies average data from many people. “By far the majority of studies have been conducted and analyzed that way, and it has taught us a lot about which regions are generally involved in which tasks across people,” says Emily Finn from Yale University.

Regions in the brain with correlated activity patterns. Reproduced with permission from Finn et al. (2015).

Finn, Xilin Shen and their colleagues in the lab of Todd Constable at Yale University decided to analyze fMRI data from individual subjects. Finn explains that they wanted to prove that data from a single subject are actually meaningful and reliable. The researchers used data from 126 subjects that had been acquired over two consecutive days during two rest sessions and four sessions that involved a variety of different tasks.

Traditionally, changes in the magnitude of the BOLD signal are used to analyze brain activity. Finn and her colleagues chose to look at functional connectivity patterns instead. “We were looking at the pattern synchrony between different pairs of regions over the course of the scan,” explains Finn. Although not novel, this approach gives the data another dimension. More specifically, the researchers divided the brain into 268 regions or nodes and determined whether these regions exhibited correlated activity over time, resulting in connectivity matrices of 268 by 268 nodes for each session. Having established these connectivity profiles, Finn and her colleagues set out to determine whether fMRI as a technique is robust and reliable enough to capture individual variability across several days and cognitive tasks. In other words, does it matter what people are doing, or does it matter who they are?

To answer this question, the researchers compared a person's connectivity matrix from a particular session with the connectivity matrices from all subjects obtained during a different day. “We wanted to show that for the same person doing two different things, you can still pick someone out of a crowd, regardless of how their brain is engaged during the scan,” explains Finn. They indeed found that it was possible to identify a person from his or her brain activity data even when the data were acquired in different sessions on different days.

An interesting question is whether brain connectivity patterns are similar in genetically related subjects. In fact, many subjects in the study were twins or had other relatives who participated. “We expected that when we got the identity wrong we might have been more likely to confuse people with their twin or their sibling,” concedes Finn, but this was not the case. The researchers did not find a strong trend for similar brain activity patterns in relatives.

However, brain connectivity profiles did correlate with an innate form of intelligence called fluid intelligence. Using the connectivity profiles, Finn and her colleagues could predict to a certain degree how individuals would perform in an intelligence test that assessed pattern recognition and problem-solving skills independently of acquired knowledge.

But Finn is also cautious about the implications. Although the results are statistically significant, such individual fMRI analyses are not ready for practical application in a clinical setting (for example, to guide treatment). Nevertheless, the findings are exciting because they suggest “that these connectivity profiles are more stable than we thought and more reliable than we thought. [They] open the door to doing more on an individual-subject level,” says Finn.