Learning about Clinical Brain Imaging using R at SAMSI’s Computational Neuroscience Workshop

The following was written by Katy Wang, attendee from the University of California, Riverside, of the Undergraduate Workshop focusing on Computational Neuroscience.

Dr. Ciprian Crainceanu

Dr. Ciprian Crainiceanu.

Of the seven presentations at SAMSI’s Computational Neuroscience Workshop from October 19-20th, the one that was most memorable to me was given by Dr. Ciprian Crainiceanu, a professor in the Department of Biostatistics at Johns Hopkins University. Dr. Crainiceanu’s presentation on neurohacking in R really stood out to me because I learned how to preprocess images, read, write, plot, and manipulate neuroimaging data in R.

You may be wondering “what exactly is neurohacking and how is the application of statistics used in clinical brain imaging?” As defined by Dr. Crainiceanu, “neurohacking is the continuous process of using, improving, and designing the simplest open source scripted software that depends on the minimum number of software platforms and is dedicated to improving the correctness, reproducibility, and speed of neuroimage data analysis.” The goal of neurohacking is the “democratization of neuroimaging data analysis,” in other words, to make neuroimaging data analysis possible for all people to understand. Throughout the presentation, we were shown an image of an axial slice of the T1-w image of the brain that contained a multiple sclerosis lesion surrounded by a hyper intense ring, which indicated blood with a higher concentration of gadolinium chelate. After taking the region of interest, a matrix is used with numbers corresponding to that particular section of the brain in order to understand what the dynamics of blood flow are into the lesion. To see if anything has changed (e.g, Did the brain tumor get bigger? Was the cancer eliminated by surgery?), the follow-up T1-w is subtracted from a baseline T1-w volume. A template-based analysis is also used in which an MNI T1 template is used to see which parts of the brain it maps to. The results are later quantified and mapped to neuroimage.

Students interacting with the lecturer

Students interacting with the lecturer.

Although there was not enough time to actually work with the data during the presentation, Dr. Crainiceanu offered a clear explanation on neuroimaging, an impressive tutorial with Powerpoint slides on how to set up the data, information on data structure and operations (working with various file types, visualization and data manipulation) preprocessing (inhomogeneity correction, intensity normalization, tools in R), registration, segmentation, dynamic visualization in R, and many resources in order to work with and become more familiar with working with the data. Furthermore, we were given suggested prerequisites and coursework, such as (1) Linux/Unix; (2) a basic knowledge of programming; (3) a basic knowledge of array data structures (e.g. 2d and 3d arrays), and most importantly (4) an interest in “hacking” with neuroimaging data! You may also find these Coursera Data Science Specialization courses offered on behalf of Johns Hopkins University as a helpful resource.

All in all, SAMSI’s undergraduate workshop was truly a great learning experience! I went into the workshop with very limited knowledge in computational neuroscience but came out of the workshop with several Word documents of notes, many data files/tutorials, and resources to enhance my knowledge of mathematical and statistical methods in neuroscience.


Learning about the challenges of computational neuroscience

The following was written by Thomas Witelski, Associate Director at SAMSI and Professor at Duke University in the Mathematics Department.

At some level, everyone is aware of the pressing medical and societal
challenges of neuroscience from media coverage of the growing impact of
neurological diseases like Alzheimer’s and Parkinson’s. Understanding the
brain at a scientific level has been identified as one of the central
challenges for this century’s research, as reflected in the magnitude of
resources invested in the NIH’s BRAIN initiative and the European Union’s
Human Brain Project.

attendees sitting in the auditorium

The opening workshop for CCNS was held at the NC Biotech Center.

In August, a diverse community of researchers converged at the NC
Biotechnology Center for SAMSI’s opening workshop for the Challenges in
Computational Neuroscience (CCNS) program. The presentations by leading
researchers on clinical, cognitive, computational and theoretical aspects of
brain research yielded many very lively discussions. Some talks addressed
technical issues, but many pointed to big fundamental questions on
exploring what might be nature’s most intricate black box.

Martin Lindquist speaking at the podium

Martin Lindquist, Johns Hopkins, speaking at the opening workshop.

A long history of anatomical studies has established the general features
comprising the human brain, but great challenges lie ahead in making clear
how the structure and functions of the brain relate to each other. Many of
the talks in the CCNS workshop addressed methods in neuroimaging. Martin
Lindquist (Johns Hopkins Univ) gave a lecture over viewing the various
modalities for functional imaging of brains in vivo, including functional
magnetic resonance imaging (fMRI), positron emission tomography (PET), and
electro/magneto-encephalography (EEG/MEG). These techniques differ in the
technologies used to collect data, but more importantly, they fundamentally
differ in the physiological types of behavior they monitor — in terms of
either blood flow, metabolic activity or electrical activity in the brain.
The methods have different limitations and trades-off in terms of spatial
and temporal resolutions, and represent the current state-of-the art in
clinical methods of collecting neuroimaging data.

Several talks in the meeting addressed fundamental statistical and
mathematical questions on image processing and how to use collected data
(possibly coming from multiple scans) to obtain the most accurate possible
maps of the brain’s structure. Of particular interest is the use of
neuroimaging data to infer the networks of connections among parts of the
brain, called the field of connectomics. In this direction, Max Descoteaux
(Univ of Sherbrooke) showed how diffusion in MRI images could be used to
identify structural connections within the white matter of the brain.

Another major branch of neuroscience research explored in the workshop is
based on “bottom-up” modeling of time series of neural activity in networks
of connected neurons. Physiologically-based models of chemical/electrical
activity like the Hodgkin-Huxley equations can effectively reproduce the
dynamics observed in individual neurons. Equivalent reduced models, like
the “leaky-integrate-and-fire” neuron, can then be used to give statistical
descriptions for the patterns of spikes typically recorded in EEG data.
Workshop presentations in this area included talks by Robert Kass (Carnegie
Mellon), Kenneth Miller (Columbia) and Uri Eden (Boston Univ).

Returning to studies at the “whole-brain” level, many speakers touched on
the computational challenges involved in analyzing the huge datasets that
have been collected in connection with some clinical studies. The importance
of using mathematical and statistical methods to interpret clinical
neuroscience was also highlighted in talks on neurodevelopment by Raquel Gur
(Univ Pennsylvania), behavioral studies by Ruben Gur (Univ Pennsylvania)
and the influence of anesthesia on brain activity by Emery Brown (Harvard).

two people looking at the poster

Looking at a poster during the CCNS Opening Workshop.

Many of the advanced topics addressed in the workshop were also introduced
in a Neuroscience Summer School that was held in connection with the CCNS
program in July. The research focuses begun in the workshop are being
carried forward in several working groups, two graduate courses and further