Virginia Tech’s Dr. Anders Eklund discusses employing GPUs for neurological medical research.

By Chris Barylick | 09-15-2012 | 10:30AM

He survived both a master’s and a doctoral program. Now, Virginia Tech’s Dr. Anders Eklund is using every GPU he can get his hands on, Cuda and Matlab to study neural activity in real-time. Read on to find out what he’s up to as well as what advice he’d give to anyone wanting to follow in his academic footsteps.


Dr. Eklund, what exactly is your role in the science field?

I work as a postdoc at Virginia Tech Carilion Research Institute. I conduct research within the field of medical image processing, especially fMRI (functional magnetic resonance imaging), and write scientific papers. Currently I’m finalizing a review paper on how graphics processing units historically have been used for medical imaging.

What exactly IS functional magnetic resonance imaging and how does it work?

fMRI is a way to non-invasively measure brain activity with high spatial resolution. To make this possible, we use magnetic resonance (MR) scanners which have extremely powerful magnets (50,000 to 100,000 times stronger than the magnetic field of the earth). The key component is that active parts of the brain consume more oxygen, which changes the mix of oxygenated and de-oxygenated blood. The oxygen is transported by hemoglobin, which is magnetic since it contains iron. The main assumption in fMRI is that the change of iron concentration is caused by a change of brain activity, but no one knows the exact link between brain activity and blood flow.

How do you employ GPUs for scientific research?

Analysis of fMRI data is normally based on a number of statistical assumptions. I try to use methods that are based on as few assumptions as possible, in order to get more reliable results. A general problem with these so called non-parametric methods is that they are computationally intensive, since they, for example, require 10,000 permutations of the original dataset to be analyzed. Here the GPU plays a crucial role, as it can decrease the processing time from hours to minutes.

How do you utilize Nvidia technology at your company?

I use an ordinary desktop with several Nvidia Geforce graphics cards. I program in CUDA and run the calculations through Matlab.

GPU computing is advancing at a tremendous rate. Is there anything you hope you’ll be able to do in the future?

A problem is that the temporal and spatial resolution of fMRI data also increases. Here at Virginia Tech Carilion we especially work with real-time fMRI, where the data analysis is performed while the subject is in the MR scanner. With more powerful GPUs we can apply advanced analysis of brain activity in real-time, and thereby increase the understanding of the human brain.

How did you get into your current line of work? And what advice would you give to someone interested in that career?

I applied for a Phd program in medical image processing, since I’ve always been interested in image processing. During my Phd, I got interested in GPGPU and thought that CUDA was a really nice programming interface. I would advice anyone to focus on what they think is fun. To obtain a Phd degree requires a lot of work, and that is not possible if you don’t love what you are doing.

Anders is a member of Stephen LaConte’s research group at Virginia Tech Carilion Research Institute. For more information, see .

Leave a Reply

Your email address will not be published.