Funding for Algorithms: Dietmar Cordes leads project in brain data analysis
June 24, 2013
Ryerson professor Dietmar Cordes and two post-doctoral fellows have embarked on a newly-funded project. This spring, the National Institute of Biomedical Imaging and Bioengineering (through the NIH) awarded Cordes a grant worth nearly $700,000 over the next three years. Cordes is developing algorithms to analyze data generated by functional magnetic resonance imaging (fMRI), a procedure that measures brain activity by the detection of blood flow. In his view, too much data is missed by the algorithms used in 95 per cent of today’s labs – data that can give insight into diseases such as Alzheimer’s and schizophrenia.
Cordes is one of the few researchers in North America who develop new mathematical and statistical methods for analyzing fMRI data. For the purpose of his current project, he is focusing on the area of the brain known as the hippocampus. “Think of it as the brain’s dictionary,” he explains. “It enables you to look up the memories stored in the neocortex.” In a brain affected by Alzheimer’s, the hippocampus loses function. Access to recent (and eventually, all) memory is lost, along with the ability to create new memories. Cordes wants to provide better tools to help researchers understand why.
Our understanding of the brain – and memory – has been greatly advanced by our ability to map brain activity through imaging equipment and software tools. Most labs use public software that creates an activation pattern from the analysis of single-voxel time series. (A voxel is a point of data on a 3D grid: in video games, voxels render terrain, as in Minecraft; they are a 3D equivalent of pixels.) The brain, notoriously complex, is hard to map. Borders between active and non-active regions are complicated by cortical folding, cerebrospinal fluid, white matter, and blood vessels. Standard tools of analysis are based on “univariate analysis” and look at each voxel in isolation. However, brain activity is a network activity and involves extended neighbourhoods with complicated activation embedded in an even more complicated physiological noise structure, which is difficult to model accurately. Cordes is developing tools that analyze voxels in the context of their neighbours leading to a “spatially constrained multivariate analysis” of brain activation. Accuracy in this field is like a balancing act between too much and too little data. Early studies suggest that Cordes has found a way to improve accuracy of detecting brain activation by up to 20 percent.
To compare his algorithm with others in the public domain, Cordes developed tasks that test “recognition memory”—i.e., the memory of things in the recent past, but further back than “working memory.” Cordes is able to distinguish different levels of activation within small fields of the hippocampus. Currently, the subjects participating in the project are healthy young adults; but Cordes can imagine developing a tool of analysis for monitoring symptoms or maybe predicting the progression of diseases such as mild cognitive impairment and Alzheimer’s.
A physics professor in the Faculty of Science at Ryerson, Cordes will be working closely with the University of Colorado at Boulder and the MIND Institute in New Mexico. He is affiliated with both institutions. US experts in recognition memory (Tim Curran), multivariate statistics (Rajesh Nandy), and optimization theory (Richard Byrd) will give breadth to his Ryerson team, consisting of post-doctoral fellows Martin Merener and Adnan Sljoka, both mathematicians.
Cordes has very specific targets: to map brain activity with more sensitivity and specificity (e.g., by finding better spatial dominance constraints to analyze voxel neighbourhoods of the brain in a multivariate setting); to compare different algorithms using a recognition memory test; and to offer his new methods to the scientific community in the form of a free software package. The ultimate goal is a more profound understanding of neurological and psychological conditions.
Cordes recently finished an NIH project entitled “Functional MRI and Alzheimer’s Disease” which led to multiple publications in multivariate modelling and task development to study memory activation. His latest project points toward an even wider scope of application of fMRI data analysis and recognition memory.
Get in touch with individual members of our faculty and staff