661270 event 1663351521 1663351521 <![CDATA[PhD Proposal by Yahia Ali]]> Yahia Ali
BME PhD Proposal Presentation

Date: 2022-09-29
Time: 2:00 pm (ET)
Location / Meeting Link: https://emory.zoom.us/j/98886145265?pwd=SkxCemVZNGhBNG85N0Y0R3pOZGZ5Zz09

Committee Members:
Chethan Pandarinath, PhD (Advisor) Garrett Stanley, PhD Eva Dyer, PhD Michael Borich, DPT, PhD David Brandman, MD, PhD


Title: Robust and responsive brain-computer interfaces enabled by neural dynamics modeling

Abstract:
Brain-computer interfaces (BCIs) may one day become long-term solutions for restoring movement to people with paralysis. For BCIs to be practical for long-term use, they must reliably decode a user's intended behavior from their neural activity. Decoding from neural activity is challenging because electrodes can only sample from a small, noisy subset of the brain’s activity. This activity changes as electrodes shift in position and glial cells form barriers between electrodes and neurons, so decoders often fail to perform for more than a few hours before needing recalibration. In past work, features extracted from modeling neural activity as a dynamical system have improved decoding performance over models that decode from binned spiking data directly. In this project, I aim to use dynamical systems modeling of the neural population to increase the overall performance and stability of BCI control. This approach will pair existing dynamical systems models with custom software for running real-time neural network inference in BCI experiments. In Aim 1, I will develop software for running low-latency BCI control experiments with support for neural network inference. In Aim 2, I aim to use dynamical systems models to increase BCI control performance. In Aim 3, I will apply dynamical systems models to the stabilization of BCI control across multiple days without decoder recalibration. Together, the goal of these aims is to advance BCIs toward more practical use that will be accurate, responsive, and stable for several days while addressing key barriers to the translation of deep learning models into BCI control experiments.

]]> <![CDATA[Zoom]]> 221981 1788 102851