Reading the contents of working memory

Working memory refers to the process by which small amounts of information relevant to the task at hand are retained for short periods of time. For example, before cellular phones became so ubiquitous, calling someone usually involved first finding the number and then remembering it for a just few seconds by repeating it to oneself several times. Once the digits had been dialled, they are immediately forgotten.

Very little is known about the neural mechanisms underlying working memory, but very recently some advances have been made. Last month, a group from the University of Texas Medical Center described a novel mechanism by which the response of single cells in the prefrontal cortex to a stimulus can persist for many seconds after the stimulus has been removed. They suggested that this could be how cells encode information for short periods of time.    

And now, researchers from Vanderbilt University have made another important finding. In an advance online publication in the journal Nature, they report that the parts of the visual cortex which carry out the earliest stages of visual processing play an important part role in retaining simple images in working memory, and demonstrate that the contents of visual working memory can be accurately predicted by decoding neural activity from those parts of the brain.

Decoding_transverse_FT.jpg

In the new study, Stephanie Harrison and Frank Tong of the Vanderbilt Vision Research Center focused on the visual cortical areas V1-V4, which contain cells that respond selectively to the simplest characteristics of objects within the visual field, such as contrast and the orientation of bars and edges. They also developed an algorithm to decode the activity associated with perception of two grating patterns, each consisting of diagonal lines of a different orientation.

While in the scanner, participants were presented with these two patterns, one after the other and in a random order. Immediately afterwards, they were cued to remember one of them. A few seconds later, they were presented with a third grating pattern, and asked to indicate the direction in which it was rotated relative to the one they had remembered.

Harrison and Tong isolated the activity correlated with visual working memory by analyzing the data obtained during the 11-second remembering period. Their algorithm enabled them to distinguish between the activity patterns associated with each  stimulus and thus to predict, with an accuracy of about 80%, which of the two grating patterns was being retained in visual working memory. It seems that a trace or echo of the activity elcited by the stimuli can maintain an on-going representation of the image viewed seconds earlier.

Thus perception of a stimulus and the temporary retention of that stimulus in visual working memory seem to involve the same subpopulation of neurons. Retention of the images was not related to the overall levels of activity in those groups pf cells- in half of the participants, activity during the remembering period decreased below the baseline level, but could still be decoded accurately. The activity was also very robust: the decoding was equally effective regardless of whether participants retained the first and second pattern, showing that the working memory representations of the images did not interfere with each other.

This study therefore reveals a mechanism by which images can be retained in working memory, and reveals that the visual cortex, whose activity is normally thought of as being driven by external stimuli, is also involved in cognitive function. However, different types of information can be stored in working memory, and it is unlikely that working memory representations of non-visual information would be stored in the visual cortex.

The findings also represent an advance in neuroimaging technology, but their implications should not be exaggerated. The researchers could make their predictions because their algorithm enabled them to distinguish between two different patterns of activity, each associated with the working memory of one of the grating patterns. In the real world, we simutaneously perceive enormous numbers of elaborate visual stimuli, each of which is likely encoded by extremely complex patterns of activity. Our state-of-the-art techniques are therefore nowhere near sophisticated enough to decipher such real world stimuli, let alone the rich and incredibly detailed memories which are stored permanently within the brain.

Related:


Harrison, S.A. & Tong, F. (2009). Decoding reveals the contents of visual working memory in early visual areas Nature. DOI: 10.1038/nature07832

Categories

More like this

Retention of the images was not related to the overall levels of activity

Yikes! Pretty much any other study of working memory with fMRI relies on that assumption. Looks like these cool new pattern classifying fMRI techniques blew that assumption away.

Our state-of-the-art techniques are therefore nowhere near sophisticated enough to decipher such real world stimuli, let alone the rich and incredibly detailed memories which are stored permanently within the brain.

Yeah, you are right. The results shouldn't be exaggerated but they are a step in the right direction and this is very exciting.

I enjoy reading your blog a lot, by the way. :) Keep it up!

Cheers,
Florian

Probably study of cortical columns will allow to understand processes of coding visual info in brain...

By She-Liger (not verified) on 20 Feb 2009 #permalink