SfN Day 5: Finding an fMRI fingerprint

Photo source: Wikimedia Commons

Photo source: Wikimedia Commons

I had a chance to stop by a really intriguing poster this morning on Day 5 (the final day!) of Neuroscience 2013. Brian Mills, a co-author on a poster (765.01) entitled “Model-based functional brain connectivity,” talked me through the pretty complex methods of this project.

In general, functional connectivity studies involving resting state fMRI entail comparing connectivity patterns between different groups of participants (e.g., schizophrenia vs. typically developing). Instead, Mills and colleagues are exploring whether or not resting state fMRI can be used to identify connectivity patterns that are unique to a specific individual. They are seeking an fMRI fingerprint.

In their study, Mills and colleagues developed a model-based connectivity matrix for a set of regions of interest (ROIs) throughout the brain of a participant, in which each cell of the matrix represents not the connectivity between two ROIs (as is traditional in functional connectivity analyses), but instead the relative contribution of neighboring ROIs to the timecourse of a specific ROI. This matrix can be used to model and predict the connectivity patterns of a set of ROIs at another time, such as a second fMRI scan date.

Comparing different linear equations that produce these model matrices, Mills and colleagues have determined the best way to predict one’s resting state fMRI connectivity patterns. And by best, I mean pretty darn accurate. Using their modeling techniques, Mills and colleagues have been able to classify single participants at scans a week after their original scan with up to 100% accuracy.

Even better, this type of modeling only seems to require about 60 frames of resting-state fMRI data to achieve accurate classification of individual participants. These frames, totaling about 2.5 minutes of scanning, don’t even have to be in sequential order. This is especially useful for patient populations, such as autism, in which motion artifact is a particular concern. Mills noted that with this method you can use participants who may otherwise not be included in data analysis due to too much motion in the scanner. This classification method also works in monkeys, proving that human-specific behavior, such as mind-wandering, is not leading to the accuracy of this method.

Additionally, when looking across their pool of participants, Mills and colleagues were able to determine which brain regions were relatively stable, versus dynamic, in terms of connectivity patterns from individual to individual. It appears that connectivity in motor and sensory areas of the brain is conserved across individuals, whereas that of fronto-parietal areas and default mode network areas is more variable between individuals. These findings make natural sense, as I would think that the brain activity that makes someone unique would occur in more frontal, higher order networks of the brain, as opposed to more evolutionarily old regions, such as motor areas.

Mills noted that they hope to apply this method to explore how the classification of individuals changes across development and when patient populations such as ADHD are considered. Mills hopes that much like knowing the variability of one’s genetic makeup, this classification method will someday aid in predicting risk for different disorders as well as the likelihood of response to certain drugs.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s