Anna Pritchard
BME PhD Proposal Presentation

Date: 2025-10-07
Time: 4:00 pm
Location / Meeting Link: HSRBII N600 & Zoom:https://emory.zoom.us/j/97349350711

Committee Members:
Chethan Pandarinath, PhD (Advisor); Nicholas Au Yong, MD/PhD; Lena Ting, PhD; Garrett Stanley, PhD; Vikash Gilja, PhD 


Title: State decoding for seamless selection between multiple iBCI functions

Abstract:
Intracortical brain-computer interfaces (iBCIs) have shown remarkable progress in restoring motor and communication functions for people with paralysis by translating brain activity into a user’s intended action such as moving their hand or speaking. Previous demonstrations have included computer cursor use, typing, robotic hand control, and speech decoding, but most studies focused on a single one of these output functions. For iBCIs to be versatile and clinically viable, we must understand how to integrate multiple iBCI functions– specifically, how to infer which function the user intends to operate at a given time and how to enable seamless switching between them. As part of the BrainGate2 clinical trial (NCT00912041), we have integrated cursor and speech iBCI functions for computer control using constant cursor activation and an on-screen button to enable speech decoding. To scale this approach to a broader repertoire of functions and increase system usability, we are developing methods to gate relevant outputs (e.g. cursor velocities) according to the user’s functional intent. In this work, I will investigate the feasibility of decoding functional intent from motor cortex activity across diverse hand and speech iBCI functions. First, I have implemented real-time intent decoding that allows iBCI users to naturally switch between cursor and speech control for computer use. Next, I will characterize how different hand-related functions (cursor, typing, robotic hand, and handwriting) are represented in the motor cortex, specifically isolating neural features that separate these functions for intent decoding. Finally, I will apply large-scale deep-learning models to real-time functional intent decoding, investigating data-efficient pretraining and fine-tuning strategies to minimize function- and user-specific data collection requirements. Together, this work will establish a foundation for iBCIs that allow natural, flexible control of multiple functions, moving the field beyond single-function restoration towards practical iBCI systems for daily life.