News

August 1, 2019

The NEMAR project to build a neuroelectromagnetic data portal to the NIMH OpenNeuro human neuroimaging data archive funded by the National Institute for Mental Health (NIMH)

UCSD scientists Scott Makeig and Arnaud Delorme at the Swartz Center for Computational Neuroscience (SCCN) of the Institute for Neural Computation (INC), together with co-PIs Amitava Majumdar from the San Diego Supercomputer Center (SDSC), UCSD, and Russ Poldrack of Stanford University, have received a five-year , nearly $5M grant from the U.S. National Institutes of Health (NIH) to create a portal (‘NEMAR’) for human electrophysiological data to the OpenNeuro data archive and tools resource (OpenNeuro.org) developed by Poldrack and colleagues. OpenNeuro has been designated by the National Institute of Mental Health (NIMH) as a repository for NIH-funded and other human neuroimaging data. Currently it mainly hosts data acquired using functional magnetic resonance imaging (fMRI), but will over time accumulate data acquired using other brain imaging modalities. The new project will add a web and data processing front door to OpenNeuro specifically designed to archive, combine, and further analyze human electrophysiological data – scalp EEG, its magnetic counterpart, MEG, and, eventually, intracranial iEEG and ECoG data sometimes recorded to plan brain surgeries for epilepsy or other brain pathologies.

Previous neuroimaging data archives were designed primarily for use only as data libraries and required users to use a complex query language to find and copy data. Typically, these served the few users who were prepared explore the downloaded data using their own tools and devices. The title, of the new project, NEMAR, stands for ‘NeuroElectroMagnetic data Archive and tools Resource.’ The new data archive and tools resource (DATR) concept integrates an archive of sufficiently annotated data, stored in compatible formats, with an easily extended resource of data analysis tools built to efficiently explore the archived data, all housed in a widely accessible and computationally powerful cloud computing environment – the latter becoming necessary to allow new machine learning tools to be applied across archived studies that together constitute a very large corpus of archived data, one that in toto could not be quickly copied to another computing resource.

To build a working DATR requires that the included datasets be well organized and described in a way suitable for machine search and comparison. Over the last few years, a new set of neuroimaging data archiving standards, extensions of the Brain Imaging Data Standard (BIDS) for fMRI data, have been created by neuroimaging research communities to organize EEG, MEG, and ECoG or iEEG, and other types of data, an effort spearheaded in part by Dr. Poldrack and colleagues, with contributions by Delorme, Makeig, NEMAR project consultants Cyril Pernet (University of Edinborough) and Robert Oostenveld (Nijmegan University), and many others. ‘BIDS-app’ tools efficiently read and process BIDS data; else, BIDS data sets may be readily exported to other formats for further analysis.

A specific need for research using human electrophysiological data has been a system for carefully describing the nature of experimental events experienced or produced by the subject during the data recording. For this purpose, BIDS incorporates the Hierarchical Event Descriptor (HED) system and tools first developed in SCCN by project consultant Nima Bigdely-Shamlo during his Ph.D. studies at UCSD. HED tagging, also an extensible standard under continuing development by the electrophysiology research community, allows researchers to use new statistical approaches to uncover hidden patterns visible only when viewed across a large enough amount of data, a new data analysis approach termed data mining. Data mining using machine learning approaches, essential tools in the new science of neuroinformatics, present exciting new horizons for researchers in human electrophysiology, who for decades could only could only compare results of their studies to those of past studies by visually comparing their plotted results against similar plots in previously published journal articles by themselves and others.

The NEMAR project has a second goal of linking the OpenNeuro data archive to high-performance computing (HPC) resources of the NSF-supported XSEDE network of supercomputers including Comet (and its planned successor) at SDSC. The NEMAR researchers at INC and SDSC are already collaborating on an NIH project to include the world-leading EEGLAB software environment for electrophysiological data analysis of Drs. Delorme and Makeig (sccn.ucsd.edu/eeglab) in the suite of neuroscience software made freely available to researchers for computing on the XSEDE network by the Neuroscience Gateway (NSG) project of Drs. Majumdar, Sivagnanam, and Yoshimoto at SDSC and Carnevale at Yale, funded by the U.S. National Science Foundation. NSG (nsgportal.org) currently provides neuroscience simulation and data processing tools and pipelines on multiple XSEDE HPC resources freely and openly for neuroscience researchers.

The NSG facility will support the development of NEMAR project data quality evaluation and visualization tools that will allow electrophysiology researchers using OpenNeuro to visualize and compute on archived data --- both raw data as recorded by the sensors themselves, and transformed data identified as generated in part by brain and in part by non-brain (‘artifact’) processes. Computing on data portions associated with cortical sources will allow data from multiple studies, recorded with different numbers, placements, and even types of sensors to be aggregated and compared within a common brain space. Brain sources of NEMAR-processed data accompanied by subject magnetic resonance (MR) head images can be localized using advanced individual electromagnetic head models.  In future, NSG users may also be able to work on OpenNeuro data ported to UCSD XSEDE resources via a high-bandwidth pipe between SDSC and commercial cloud providers such as Amazon, where the OpenNeuro data will reside.

For more information contact: Ms. Rhonda McCoy (rmccoy@ucsd.edu) or Scott Makeig (smakeig@ucsd.edu) Institute for Neural Computation; Mr. Jan Zverina (jzverina@sdsc.edu) or Amitava Majumdar (majumdar@sdsc.edu), San Diego Supercomputer Center

NIH award notice: https://projectreporter.nih.gov/project_info_details.cfm?aid=9795341&icde=45812664