Courtesy Mike Wilde and Jed Dobson

"We had people come in and listen to short clips of songs, and asked them which they knew and which they didn't," said Assistant Professor William Kelley, one of the study's researchers. "Then we sent them into the fMRI scanner, where we played the same songs, with short gaps inserted in the music."

The goal was to analyze each subject's brain activity during the moments when there were gaps in the songs, comparing brain activity during silence in songs people knew to activity during silence in songs they didn't. Before the analysis could be performed, however, all the fMRI scans had to be normalized.

"There's a whole set of processing you have to do before you can even look at the data," explained Kelley. "You have to make everybody's brain the same size, eliminate effects from brain motion during the scan, and make many other adjustments to normalize the scans. Grid computing is very useful for this."

Before computer scientist James Dobson and collaborators at Dartmouth turned to grid computing for fMRI normalization, researchers were tying up clusters and desktop computers for days at a time running the 20,000-file scans through the necessary steps.

"The normalization is very computationally and data intensive," said Dobson. "We started working with the Virtual Data System group in the Grid Physics Network, and found that a lot of the methods they had developed for physics worked quite well for our applications."

The Dartmouth team can now run their fMRI normalization on their own computer clusters, on their campus grid, or on Grid3 using the same infrastructure. As a result, the data can be ready for analysis in much less time, and studies like William Kelley's can be completed and published more quickly.

So what does your brain look like when you're listening to silence during a song you know? The same brain regions that are active when you're hearing the song remain active when the actual music is silenced, so you "hear" music that isn't actually playing.