Alex Rockhill

Hello! I am postdoctoral researcher at Oregon Health & Science University in the lab of Dr. Ahmed Raslan where I study movement and cognition using intracranial electrophysiology recordings. I completed my PhD at the University of Oregon in the Department of Human Physiology researching in the Swann Lab studying the electrophysiology of movement and movement disorders using computational neuroscience. Check out some of the science I have done that is displayed below. I am a Pacific Northwesterner who loves neuroscience, the outdoors, running, dogs and all things sustainable. I worked with Alik Widge MD PhD at the Martinos Center (Harvard/Massachusetts General Hospital affiliation) studying psychiatric patients with deep brain stimulators using electrophysiology and neuroimaging for two years. Before that, I worked in the Bair Lab studying 3D shape representation in area V4. I also worked with Erik Carlson MD PhD while at UW studying cerebellar serotonin knockout mice. I have degrees in neurobiology (BS) and applied math (BS) while also completing the computational neuroscience program. I love the discovery, community and shared sense of purpose of being a scientist and software engineer.

Classification of Movement-Related Oscillations in sEEG Recordings with Machine Learning

For my first results paper with stereoelectroencephalography (sEEG), I wanted to use interpretable machine learning to study how populations of neurons synchronize their communication via oscillations when humans produce movement (here is the preprint). Nicki, my advisor, has done a lot of work on how oscillations are related to Parkinson's disease and healthy stopping movement (just to name two of many examples), and I wanted to extend that to the brain regions sampled by sEEG as well as trying to take a bigger bite out of the apple by going for all the movement-related oscillations at once with a machine learning approach. Our relatively simple support vector machine (SVM) approach worked really well and was able to provide good classification right where you thought they would be (i.e. primary motor cortex) as well as a wide range of other areas ranging from somewhat related to movement to more closely related to movement. A cluster permutation analysis got almost exactly the same features as the SVM, which was really reassuring that this was legitimate. We could have used a more advanced, deep learning technique, and probably will in the future, but the interpretability of the analysis was great and I hope having it be a bit simpler will make it so that a wider range of audience finds it accessible and enjoys it!

Here the paper.

MNE-Python Intracranial Analysis

Analysis of intracranial electrophysiology data takes many steps to process to find out where the contacts are recording from, which I implemented in MNE-Python. The computed tomography (CT) image has to be aligned to the magnetic resonance (MR) image to link the positions of the contacts (seen in the CT) to the brain (shown in the MR), the locations of the contacts have to be found in the CT and the brain of the individual has to be warped to a template for group analyses. Once these steps are done, the intracranial EEG data can be analyzed similar to other time-series data with positions and labels of brain areas that are being recorded from, like EEG and MEG, which works really well with the rest of MNE-Python which deals with all the cool analyses you can do with this kind of data. For instance, you could use machine learning to classify which signals are when someone is performing a movement and which are when then they are not, like this example. More on this coming soon!

pd-parser

As a start to my PhD work, I made a software package that processed photodiode events from a corrupted channel as that is a big challenge in the epilepsy monitoring unit where patients have to have the task laptop on their lap, causing lots of noise. This task sounds somewhat trivial, but due to computer clock drift bbetween the laptop and the recording computer and photodiode deflection-like artifacts, it's a bit more difficult than it sounds. I implemented continuous integration, used up-to-date python style and tested and documented all the functions with the idea that if this task is done correctly once, it will be done in a way that is rushed and potentially error-prone many fewer times. pd-parser was published in JOSS as well, which is a journal that I whole-heartedly support as it is open access and doesn't require any publication fees due to it's structure of requiring authors to format their own manuscripts and its use of GitHub for reviews.

Multi-Modal Visualization Tool (MMVT)

When working with high-dimensional and especially multi-modal datasets, it is absolutely crucial to visualize them with a powerful tool. One of the most fun projects I've worked on is developing mmvt. This amazing tool uses an open-source animation tool, Blender to create meshs of tiny 3D planes that have a virtual camera detect virtual photons in the process of rendering that underlies top Hollywood CGI. This process can be used to make very cool pictures of brains, including in augmented reality. Check out the example gallery to see the many amazing images made in MMVT.

MNE-BIDS Software Development and Reproducible Science

I work on developing mne-bids because I am passionate about reproducibility in science. Papers that I publish contain quality science that I would like to be seen and, for those interested, downloaded and worked through. BIDS has been a leader in setting pragmatic but efficiency-promoting standards that has made neuroscience much more reproducible and accessible, and I am glad to play a role in this effort.

Approach-Avoidance Conflict Paper

I've written a paper with Sam Zorowitz on hierarchical Bayesian modeling of an approach-avoidance task as applied to an fMRI analysis. The Bayesian framework is a powerful way to estimate laten states accounting for individual differences simulatenous with and constrained by a group estimate.

Closed-Loop Brain Stimulation with an Analog Filter

I've written with analysis of a paper with Matt Boggess on the effects of deep brain stimulation (DBS) on endogenous firing patterns in a rhesus macaque model. It turns out that non-phase-locked power increases in the stimulated frequency but not phase-locked power whereas open-loop stimulation and playing back the brain's own endogenous activity decrease power in the frequency(likely because anything that is not aligned with endogenous rhythms has a tendancy to disrupt endogenous rhthyms due to enthropic principles). This paper used a novel, throwback approach. This may be a contradiction in terms, but we used a modern digital recording setup while basing stimulation on the analog signal which is an approach generally not used in decades. This allowed us to stimulate one cycle or less then 100 milliseconds later after processing the frequency content of the signal through an analog filter. This would have been far too slow to digitize, compute a fourier transform and then inform stimulation.

Neuretymology

Here is an art project I made for Biomedical Imaging-themed art event that I hosted art in mARTinos. How many etymological roots of brain regions can you spot in the piece?

Art Neureau Piece/V4 Stimuli Representation

I showed a piece similar to this on loop at Art Neureau, a neuroscience-themed art show put on by UW graduate students. Stills of these images were shown to rhesus macaque monkeys with an electrode in area V4 to determine which shapes elicted maximal responses from a neuron. The shape selectivity followed the angular position and curvature model wherein each neuron responded most to a shape with a particular feature such as a concavity to the right or a protrusion to the upper-left. Along with the work I did on the latency of the responses of these neurons (shown below), this positioned future research to show videos of 3D stimuli to neurons at speed that they would respond to in order to characterize neuronal responses in this much larger shape space. Here is my undergraduate honors thesis.

Here is my CV.