Early Graduate Projects (Pitt / CMU)
Neural activity in the frontal cortex has been hypothesized to represent three stages of motor processing: feedforward planning, motor command, and visual feedback. This processing can be considered serial, having latencies corresponding to each stage. I spent the first years of graduate school incorporating these latencies into prosthetic decoding algorithms using digital filters. This work was a collaboration with my mentoring post-doc, Meel Velliste.
A.) Psychophysics: A monkey tracing an ellipse in virtual reality. I delayed the displayed movement to determine how robust the motor system is when performing the task with late visual feedback. Subjects still performed adequately with up to 200mS of delay. Furthermore, we found that certain portions of the trace were more dependent on visual feedback than others.
B.) Brain-Controlled Cursor: During neuroprosthetic applications, we use neuronal population activity to control the trajectory of a cursor tracing an ellipse. Latency values were determined for each neuron and incorporated into the decoding system. We demonstrated improved control, surprisingly, when average latencies were used.
C.) Brain-Controlled Robotic Arm: While controlling a robot arm with population activity, we found improved control when assigning individual latencies to each neuron. The differences in results between B and C may be explained by the method we used to find the latencies. In part C, we used a target position change and observed the corresponding activity change to calculate the empirical latency (right side, C), whereas in part B, we used cross correlation.
Anatomical Coordinate Mapping: I regularly performed surgery on monkeys to implant recording chambers and microelectrode arrays into specific cortical areas. Meel Velliste and I developed software in house to rotate MRI coordinates to surgical stereotaxic coordinates. Each set of MRI images (2D slices) can be thought of a 3D matrix of intensity values. We utilized anatomical features of the cranium to calculate the relevant rotation matrices, applied them to the set of images, and then re-sliced them in surgical coordinates. We also implemented a cortical shrink-wrap that allowed us to visualize the cortex before surgery.
Fabrication: I designed and fabricated much of my experimental setup in-house using Solidworks, Mastercam, and a Hass 3-axis CNC. Above is a a three part recording chamber that was surgically implanted into the cranium in two stages. First, I would install the bottom "octopus" ring on top of the skull to allow the bone to grow around the screws (green). After a month or so, I would perform a craniotomy and install the second and third pieces. Designed in collaboration with George Fraser.
The Neurosciences Institute (San Diego, CA)
Image Processing: Before graduate school, I spent a year at the Neurosciences Institute, developing software (Matlab) for histological assays of brain slices of the auditory cortex. Most of the results from the research group I belonged to (Principal Investigator: Weimin Zheng) relied on a time-consuming process of hand-counting neurons stained with chemical markers. I independently developed a reliable automated cell counter that convolved a "mexican hat filter" over slice images. This software was apparently used in the lab until NSI discontinued laboratory research in 2012.
Neuropace Inc. (Mountain View, CA)
My first job out of undergrad was a research internship at Neuropace, a startup company developing an implantable deep brain stimulator for epilepsy. The device monitors gamma activity deep in the brain and stimulates to reset activity before the onset of a seizure. I was investigating signal coherence from two electrodes as an effective flag for pre-onset epileptiform activity. My results were modest, but I loved working in the startup environment.