How accurate are brain scans in predicting behavior?

This is a deep question, and it all turns on what you mean by “predict”. Current machine learning technologies coupled with high-resolution MRI are able to successfully classify scans with near-100% accuracy in simple settings. But that’s for fairly contrived scenarios such as classifying the orientation of a line that the subject is viewing, or the direction of a moving stimulus. Other work has shown that it’s possible to decode natural images, movies and even dreams from the pattern of activations in the visual cortex with fairly high accuracy.

None of this should surprise us - the activity in visual cortex is topographically organized, with different regions dedicated to different regions of space, and tightly coupled to the subject’s perception of the world. With the right technology and decoding algorithms it’s possible to pull out this signal. What’s much less understood is the neural “code” in higher brain areas such as the prefrontal cortex, that subserve abstract thought. But even here it’s possible to classify abstract operations such as addition or subtraction from activity patterns, albeit with lower accuracies around 65%.

However the way most of these decoding algorithms work is that they are first “trained” on a dataset where the experimenter knows the ground truth (such as whether the subject was adding or subtracting) and then applied to a test set to predict mental states or behaviour. Without the training set decoding is often impossible - e.g. it’s not currently possible for us to scan your brain and decode your thoughts unless a suitable training set is available for each possible thought. This is unlikely to change as technology advances - while ultimately every behaviour and mental state is a product of patterns of brain activity, the distributed nature of neural coding and the uniqueness of individual brains means that our thoughts will remain safely locked up for some time to come.

Find the answer here: http://bit.ly/1OSGcH3

This answer was provided by : Steve Flemin,

Cognitive neuroscientist, University of Oxford

More about me here:

Twitter:@fouad_khafaga
Hotmail: Fouad_khafaga@hotmail.co.uk
Google+:  khafagafouad@gmail.com