Shared High Value Research Resources: The CamCAN Human Lifespan Neuroimaging Dataset Processed on the Open Science Grid

Don Krieger
Neurological Surgery University of Pittsburgh Pittsburgh, PA, USA kriegerd@upmc.edu Paul Shepard
Physics and Astronomy
University of Pittsburgh Pittsburgh, PA, USA shepard@pitt.edu
Ben Zusman Neurological Surgery
University of Pittsburgh Pittsburgh, PA, USA
zusmanbe@upmc.edu
Anirban Jana Pittsburgh Supercomputing Center Pittsburgh, Pa anirbanjana@cmu.edu

David Okonkwo Neurological Surgery University of Pittsburgh Pittsburgh, PA, USA okonkwodo@upmc.edu Abstract—The CamCAN Lifespan Neuroimaging Dataset [1,2], Cambridge (UK) Centre for Ageing and Neuroscience, was acquired and processed beginning in December, 2016. The referee consensus solver deployed to the Open Science Grid [3,4] was used for this task. The dataset includes demographic and screening measures, a high-resolution MRI scan of the brain, and whole- head magnetoencephalographic (MEG) recordings during eyes closed rest (560 sec), a simple task (540 sec), and passive listening/viewing (140 sec). The data were collected from 619 neurologically normal individuals, ages 18-87. The processed results from the resting recordings are completed and available for download at http://stash.osgconnect.net/+krieger/ . These constitute ≈1.7 TBytes of data including the location within the brain (1 mm resolution), time stamp (1 msec resolution), and 80 msec time course for each of 3.7 billion validated neuroelectric events, i.e. mean 6.1 million events for each of the 619 participants. The referee consensus solver provides high yield (mean 11,000 neuroelectric currents/sec; standard deviation (sd): 3500/sec) high confidence (p < 10-12 for each identified current) measures of the neuroelectric currents whose magnetic fields are detected in the MEG recordings. We describe the solver, the implementation of the solver deployed on the Open Science Grid, the workflow management system, the opportunistic use of high performance computing (HPC) resources to add computing capacity to the Open Science Grid reserved for this project, and our initial findings from the recently completed processing of the resting recordings. This required ≈14 million core hours, i.e. ≈40 core hours per second of data. Keywords—magnetoencephalography, MEG, referee consensus, opportunistic computing, shared data, concussion, TBI I. INTRODUCTION It has been the informed expectation for a century that the keys to understanding the human brain will be found in measuring and understanding the electrical activity of neurons. Today, clinical neurophysiologists routinely measure single neurons to aide implantation of therapeutic devices deep in the brain [5]. Epileptologists use arrays of implanted “stereo EEG” electrodes and the population recordings obtained from them to diagnose and guide the treatment of intractable seizure disorders [6]. It is population activity which is thought to be the basis for brain function. Stereo EEG and comparable invasive methods produce voltage recordings with resolution of a few millimeters at best from up to a few hundred recording sites. Because the electric field interacts strongly with the conducting tissue in the brain, these measures are difficult to localize if there is much tissue between the field source and the electrodes. This problem is particularly pronounced when the recordings are made noninvasively from electrodes placed on the scalp.

Magnetoencephalography (MEG) provides an alternative noninvasive measurement approach with several advantages over scalp and even implanted EEG recordings. A typical MEG scanner is shown in Figure 1. The magnetic fields produced by minute electric currents within the brain are measured at the MEG sensor array with high fidelity. Unlike electric fields, magnetic fields do not interact with brain tissue; they pass through it as if it weren’t there. So in principle, the current which is the source of such a field is more readily localized. If the contribution to the MEG measurements of a single neuroelectric current source can be identified, the corresponding current can be accurately estimated using the Biot-Savart law [7].

Fig 1. 306 channel whole head VectorView MEG scanner (Elekta, Inc., Stockholm, Sweden). The sensor array is composed of 102 planar “chips,” each with 3 superconducting magnetic field sensors: one magnetometer, two figure-eight gradiometers at right angles to each other, and a superconducting quantum interference device (SQUID) coupled to each. The SQUID’s are used to couple the minute currents induced by magnetic flux in the sensors to the room temperature electronics. The measurements at the MEG sensor array are due to an unknown and almost certainly large number of neuroelectric currents.

…(Full text truncated)…