There are more than two million concussions seen in emergency rooms in the US every year; yet fewer than 20% show structural abnormalities on computed tomography (CT) scan or magnetic resonance (MR) imaging. Functional tests using neuropsychological or computerized behavioral measures are sensitive, but they are not specific for the diagnosis of concussion. Our effort combines structural images with functional brain recordings obtained during task performance.
The clinical purpose of the effort we describe is to develop methods to improve diagnosis and treatment of concussion, also known as mild traumatic brain injury. The effort is under the auspices of the Brain Trauma Research Center (BTRC) at the University of Pittsburgh, US. The primary goal of the BTRC is to improve outcomes following brain injury; it is a multidisciplinary research program funded by the US National Institute of Neurological Disorders and Stroke.
The aim is to generate information that is both sensitive and specific enough to promote understanding, diagnosis, and treatment of concussion. Even the simple advance of confirming that a traumatic brain injury has occurred could make a huge difference. For patients and their families this would markedly reduce the anxiety and frustration of dealing with a significant illness for which no objective diagnostic measures exist. And for the physician, “If we don’t know what it is, we can’t treat it …” would no longer determine the quality of care.
Electrical current flow within populations of neurons is a fundamental constituent of brain function. The resulting fluctuating magnetic fields may be sampled noninvasively with an array of magnetic field detectors positioned outside a patient’s head. This is magnetoencephalography (MEG). Our analysis of MEG measurements is for the first time enabling observation of those neuroelectric currents as if from one million plus virtual recording electrodes.
The extracted virtual waveforms, represented as dots in the video, are generated by (1) currents in nearby fibers due to the passage of volleys of action potentials and (2) postsynaptic currents in the dendritic domains of nearby neuronal populations (due to arriving volleys of action potentials). The sources of these volleys of action potentials are populations of neurons from which the fibers originate.
High definition fiber tractography is used to assign each source’s location on the fiber tracts near it. Using this information and the patient’s virtual recordings, propagation velocities for each fiber tract, and the coupling strength of each neuronal population to each tract can be obtained. This will produce a personalized functional wiring diagram of the brain – the individual patient’s functional “Connectome.”
Both the MEG and Connectome optimization problems involve systems of thousands of equations/measurements in hundreds of variables. Their solutions require replacement of the familiar mean squared error cost function with a new measure of goodness of fit: “referee consensus.” This distribution-free metric enables optimization with very high statistical power. And it can be effectively applied for one or a few variables at a time while holding all others constant.
For the MEG problem, the solution produces spatial resolution approaching 1mm from raw data and multiplies the information about ongoing brain function by more than 1000 compared with any other functional brain mapping method. Implementation for the Connectome problem is under way.
The referee consensus metric converges only within a few millimeters of a true current source. Optimization therefore requires a search. For this the brain is divided into more than 3000 1/2 cm3 cubes. The independence and parallelism of the method and its small memory size ideally suits it to the Open Sciences Grid (OSG). The computing requirements for the Connectome problem are similar and we anticipate that it will also run efficiently on OSG.
The Glidein Workflow Management System (GlideinWMS) works on top of Condor to give us easy access to grid resources. We use generic tools, e.g. csh, scp, rsync, tar, and ssh, to interface with OSG. These work reliably, and script development has required only a modest effort and time commitment. Access to OSG via XSEDE provides full-scale implementation of this revolutionary method for extracting functional brain imaging information.
David O. Okonkwo is the clinical director of BTRC and director of the scientific effort. Walter Schneider is senior scientist at the Learning Research and Development Center and lead developer of High Definition Fiber Tractography. Donald N. Krieger gathers magnetoencephalography (MEG) data from patient and normal control volunteers and devises data analysis methods. Mats Rynge, OSG User Support, assists with the program’s high throughput computing efforts.