How the grid can be used to test the software packages tracking a diagnostic biomarker for Alzheimer’s disease.
Alzheimer’s disease starts slowly and patients may not show symptoms for many years. Yet subtle physical changes will already be occurring, as their brain cells begin to die and the brain itself atrophies. The key to early diagnosis is to find a reliable ‘biomarker’ for Alzheimer’s that researchers can track, in order to monitor the disease and decide on treatments. But how can they identify a reliable biomarker from brain scan images?
One useful clue is the volume of the hippocampus – the region in the brain associated with memory that starts to shrink at the onset of Alzheimer’s. Many software programs have been developed to measure changes in hippocampus size from imaging scans. Now researchers at Vrije Universiteit Amsterdam have used grid computing to compare the performance of several software programs by analysing thousands of MRI scans taken from Alzheimer’s patients.
The study produced a valuable benchmark to evaluate and monitor Alzheimer’s biomarkers. Better biomarkers from brain scan data open the door to earlier diagnosis, effective monitoring, and being able to quickly test new drugs for Alzheimer’s.
Illustration showing a brain at the preclinical stage of Alzheimer's disease, highlighting the location of the hippocampus.
Deciphering MRI data
Alzheimer’s disease kills brain cells and causes the brain to atrophy and shrink. As the disease progresses, patients begin to experience memory loss and inability to carry out physical tasks. The diagnosis is usually made on the basis of these symptoms and the challenge is to distinguish between the normal signs of aging and the beginning of Alzheimer’s.
One way to tell normal aging from actual symptoms is to look for the visual signs of Alzheimer’s disease – the biomarkers of Alzheimer’s. The volume of the hippocampus, one of the first regions to suffer as the disease develops, is one of these biomarkers. Doctors can use a variety of different software programs to measure hippocampus shrinkage from MRI scans. They are computing intensive and may take several days to run on each scan.
Keith Cover, a physicist working at VU University Medical Center (VUmc) in Amsterdam, tested the reliability of several hippocampal shrinkage software packages. This involved analysing approximately 3,300 MRI scans from over 600 patients many times each, requiring many core-years of computation.
The brain scans were collected during the Alzheimer’s Disease Neuroimaging Initiative (ADNI) study, which began in 2004. Each patient provided two back-to-back scans in every visit. This allowed the team to look at two sets of what should be identical data and check how reliable each software package is.
Computing with neuGRID
Software packages FSL and FreeSurfer are widely used in Alzheimer’s research and consume a lot of computational power to piece together data from hundreds of patients. For this study, Keith and his team used the neuGRID – a leading European e-infrastructure for the neuroscience research community.
Through the neuGRID platform, researchers can analyse and share brain scan datasets, use medical software tools and benefit from specialised support.
neuGRID was originally established in 2008. Now, the EU-funded N4U project, led by Giovanni Frisoni, has expanded neuGRID through collaborations with several partners, including EGI, which have provided access to public grid computing resources via the Alternative Energies and Atomic Energy Commission (CEA) and Hellasgrid, the NGI of Greece.
With neuGRID, the calculations for Keith’s study were processed in weeks, instead of years.
“A modern trial could have 100 to 1,000 patients providing these scans,” Keith explained. “Most laboratories have their own MRI scanners but few have their own clusters and the expertise to use them, so they don’t have the computational power to carry out the data analysis. With grid computing, thousands of scans can be processed in a trivial time, days instead of years.”
Thanks to neuGRID, university groups can carry out their own studies, then make use of grid resources to analyse the data.
The study was able to show that FSL and FreeSurfer analyses of hippocampal atrophy rates are similar, while noting slight differences in the current versions of the programs.
With the speed of analysis brought by grid computing, the hippocampus biomarker can be used more widely to help diagnose and monitor individual patients. Until now, this biomarker has only been looked at on averages over groups and clinical trials.
The study has also demonstrated that the ADNI data provides a reliable benchmark that could be used to compare the performance of other similar software packages.
Keith is already looking towards the future: “Now that we’ve got this benchmark in place, we’re now looking at different algorithms and hoping to see if someone has come up with something better.”