Research On Critical Techniques And Neural Response Mechanisms Towards Setero-electroencephlography Based Hand Motor Brain-computer Interface

Brain-computer interface(BCI)is a technology that enables the human to communicate directly with the outside world by using neural signals from the brain.The advent of this technology has provided a brand new solution to help the patients,who lose their motor function due to the damage of the pathway transmitting nerve impulses from the brain to various muscles,regain their ability to interact with the external world.During the past 30 years,the BCI has made significant progress,where the invasive BCI has advantages in the quality of the obtained brain neural signals because it records directly inside the skull,and thus making it possible to decode fine motion directly.However,current invasive BCI is still suffering from the problems such as limited decoding types and poor stability.One of the important reasons is that the current understanding of the neural mechanism from the brain on motion control is still very limited.Therefore,in order to better serve the BCI,this thesis mainly conducts the research on the mechanism of the brain neural response during human hand movements by using stereoencephalography(SEEG).Then,based on current research results,we also conduct another research on testing the possibility of using SEEG signals to decode human hand movements and control a prosthetic hand.In order to carry out mechanism and application research based on SEEG,this thesis first solve three critical fundamental problems that exist in SEEG technology:(1)Since SEEG belongs to the intracranial EEG(electroencephalography),where a software package is necessary for the SEEG electrode localization and visualization.To solve the problems of lacking necessary functions and complicated operation that existed in current localization software,this article has developed a new Matlab toolbox that can be used for SEEG electrode localization and visualization.To address the problem of lacking necessary functions in the past,the software in this article provides enriched functions including 4 major functions,where these features can fulfill the basic needs for most of the SEEG research.For the problem of complex operations,this software integrates multiple functions into one Matlab interface and provides graph user interface-based operation,which effectively simplifies the operation for users.This toolbox also provides an basic instrument for the following research;(2)In order to answer the unclear question that which kind of spatial filtering methods should be used to maximize the signal quality during the pre-processing of SEEG signals,this thesis recorded SEEG data from multiple subjects,applies 6 different spatial filtering methods separately,and then utilizes three signal quality matrices to evaluate these methods comprehensively.The analysis results show that the Laplacian method produces the most significant effect in improving the overall signal quality.These findings successfully suggest an optimal spatial filtering method for SEEG analysis and answer the question regarding the method for signal quality improving in SEEG analysis;(3)Aiming at solving the problem that the characteristics of neural signals below the cortex are largely unknown,this thesis investigates the characteristics of neural signals in white matter using SEEG recordings.By localizing the white matter channel accurately using the proximal tissue density(PTD)and conducting signal analysis,this work finds that a wide range of white matter in the brain present activation that similar to gray matter but with a significantly lower amplitude than gray matter under the task.Furthermore,this thesis explores the possible source of signals that detected in white matter.This work reveals the characteristics of signals in white matter,proves that task-related neural information can also be recorded in white matter and expands the range of usable neural signals within the human brain.Subsequently,in order to address the problem that the neural mechanism of the human brain in relation to motion control is still not clear enough,this dissertation conducts investigation on the neural response characteristics of the human brain during visually-cued hand movements,which mainly answer three critical questions(3W)relating to the neural mechanism.Using SEEG data from multiple subjects and highfrequency-based analysis method,this work determines the responsive regions under the task and their corresponding correlation value with the task across a large area of the human brain(Where).By developing improved single-trial neural activation detection algorithm,this work reveals the large-scale temporal activation sequence of main responsive brain regions during the task(When).According to the neural activation pattern presented in each channel,this thesis evaluates the functional attributes of the main brain regions during task processing(What).Finally,based on the research findings above,this thesis describes the possible neural mechanism of the human brain under the task,reveals several important brain regions involving in the motion control,enhances the understanding of human brain and also provides clues for improving BCI performance and creating new BCI paradigms.Finally,aiming at addressing the problem that SEEG-based motor BCIs are short of study,this paper conducts the research on human hand motion decoding and prosthetic hand control using SEEG recordings based on the previous research findings.This thesis builds an integrated software-hardware control platform for SEEG-based BCI,extracts the frequency band power of the SEEG signal as features,selects the key reactive channels using discrimination index and adopts dual classifiers including linear discriminant analysis and support vector machine to decode the controller’s motion state and three different gestures.The results demonstrate that subjects can successfully control the prosthetic hand to make different gestures asynchronously using their own brain signals from only a few channels with a decoding accuracy of 78.7% in average(chance level 25.0%),verifying the possibility of implementing brain-machine motion control using SEEG recordings and enriching the research achievements for SEEG-based BCI.