Cloud based neuroimaging analysis for identifying traumatic braininjuries and related changes

Lovelace Biomedical Research Institute, Albuquerque NM

Investigators

Linked publications & trials

Abstract

Project Summary (30 lines max) This proposal outlines plans to evaluate the performance and utility of cloud-based data processing for computationally demanding analysis of MRI-based brain imaging data. This administrative supplement would build on the aims of a recently awarded R01 which develops classification algorithms for identifying and tracking progressive pathology associated with mild traumatic brain injury (mTBI) in a population of high-risk individuals. Over the last decade, our team has been continuously funded by NIH to collect detailed clinical and neuroimaging protocols from over 4000 high-risk men and women. Our extant data include multimodal neuroimaging protocols (sMRI, fMRI, DTI), thorough clinical assessments, neuropsychological evaluations, and histories of TBI. The aims of the current project are to generalize existing classification algorithms for mTBI from community samples to high-risk forensic samples and to improve on an objective neuroimaging-based measure of cognitive decline. On traditional platforms, these neuroimaging-based classification tools involve hundreds of thousands of potential features and require running times of several weeks, even for relatively small numbers of subjects. Given the computational complexity of the analyses required for this project, cloud-based computing platforms could be highly advantageous in terms of efficiency. We propose, first, to containerize our customized neuroimaging pipelines for pre-processing, followed by implementation of our current locally implemented classification algorithms. A cloud-based solution will allow us to explore several algorithmic approaches towards feature selection and union in a shorter time frame than using a local server-based solution. In order to test the feasibility and advantages of cloud-based processing, we will build data processing pipelines and validate them using existing data. Specifically, we would like to prototype algorithmic approaches towards detecting trait related changes in neural connectivity and test these using extant data collected under NIH support and from publicly available neuroimaging databases (e.g. FITBIR). Indeed, one of the aims of our R01 award is to test the generalizability of our algorithms to data in FITBIR (readily available). This testing could begin as soon as supplement was received. The cloud-based platform versus local-server-based processing will be evaluated in terms of data processing speed and costs (including human working hours). These objective measures will give us a clear picture of the value of implementing cloud-based processing on a larger scale, including applications for the longitudinal aims of the current grant.