Stabilization of Brain-Machine Interface Systems via Alignment to Baseline

Author: Porter, Tara S.

Year: 2021

Degree: Senior thesis (Major)

Advisor: Emami, Azita

Committee Member: None, None

Option: Electrical Engineering

DOI: 10.7907/y1jt-b620

Abstract

Research in the brain-machine interface has the potential to transform the lives of individuals with limited motor capabilities to allow for greater independence. By directly accessing signals in the brain, it is possible to train a decoder to identify intended motion and allow the user to control a prosthetic limb or computer cursor by simply thinking about the motion. However, neural data recorded from implanted electrodes is highly unstable over time and across multiple sessions, leading to a severe drop in decoding performance as the test data becomes more distant from the data on which the decoder was trained. Here, we investigate a method to stabilize neural spike data from human trials of a center-out cursor control task before it is passed to a linear decoder, using the techniques of factor analysis and Procrustes alignment. We find that for highly variable human neural data from experiment dates that are far apart, the method does not help the decoder better predict cursor kinematics. However, when factor analysis weights are averaged over multiple baseline days, the performance of the decoder significantly increases with Procrustes alignment, which gives a promising method to limit recalibration and retraining of neural decoders by prolonging their higher accuracy performance over time.

Files