Introduction

Reconstruction of the intended motion from the neural spike data is a very important problem. Given a reliable reconstruction we can build alternative interfaces for the disabled people that can act as true substitutes for their lost limbs. In this study we are limited to dealing with the motor-cortex spike data for the hand motion. The data was collected from a monkey with an implanted array of sensors. My goal is to study and compare the different ways we can achieve reconstruction of the motion based solely on the neural data with no sensory feedback mechanism. In particular I will study the performance of two models, Linear Filter model and Kalman Filter model, that can be used to reconstruct the intended motion. In addition I would like to see if the dimensionality reduction via PCA can help the performance of the filters. If the nature of data is such that the dimensional ty reduction is appropriate, applying it may yield better performance by reducing noise in the data.

Full paper with discussion and results can be found herein Acrobat PDF format.