Using high performance computing to help build the whole brain structural connectome
Karen joined the School of Psychology, now part of the Faculty of Health Sciences at ACU, in January 2015 as Senior Research Fellow. She has a strong background in behavioural neuroscience and experience with a variety of brain imaging methods, including Diffusion Tensor Imaging (DTI), which makes it possible to visualise the white matter connections of the brain. Karen tries to reveal critical insights at the interface between brain structure and connectivity, in relation to behaviour of different patient groups, mostly children with Traumatic Brain Injury (TBI).
She briefly presents her research below:
Brain injury is the most frequent cause of disability and death among children, often as a result of traffic accidents. Many children with TBI are faced with persistent cognitive and motor deficits, which have substantial negative consequences for their quality of life. In previous DTI studies, we have related integrity of white matter connections and behavioural deficits in children with TBI, with the majority being focused on regional changes in specific brain tracts. This is possibly due to the continuing enthusiasm of researchers for the traditional localisationist view of the brain and the increasing accessibility of brain-mapping tools to researchers. While particular brain tracts are important for specific functions, the capacity of information flow within and between brain regions is also crucial. Within this perspective, the concept of ‘brain connectome‘ has emerged, referring to a comprehensive map of neural connections in the brain.
Such a framework has the advantage of being closer to our understanding of brain organisation. Specific brain functions, like executive functions, cognition and language, depend on the coherent activity of widely distributed networks. However, building this whole brain structural connectome as a computer model requires a huge amount of memory and long calculation times. Therefore, such analysis is only possible using High Performance Computing (HPC). Intersect’s HPC systems use large-scale clusters of computers and parallel processing techniques that enable research activities through complex system modeling and solve sophisticated computational problems.
In addition, HPC systems have vastly increased memory and storage. The single compute nodes in the cluster are connected via a high speed, low latency Infiniband interconnects which allows for parallel usage of many nodes in a single job. The availability of the cluster are above 92% allowing for long running jobs (up to 8 days for a job on an unlimited number of nodes). The disk storage subsystem offers high speed and high availability. The used file system is PanFS which a parallel, global file system. In addition, each compute node has a local scratch disk for temporary results. All this leads to a high performing system.
While the connectome analysis for a set of participants would take 48 to 60 hours on a desktop machine, same analysis on Intersect HPC clusters would typically yield high quality results in less than one hour.
Using Intersect for the current project, we will be able to address the effects of a Virtual Reality training on structural networks, providing a window into neuroplasticity in TBI patients. These insights will provide a foundation for therapy to maximise recovery after brain damage.