東華大學圖書館 |

Topics on Functional Variable Selection with Application to EMG Data Analysis.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Topics on Functional Variable Selection with Application to EMG Data Analysis./
作者:	North, Rebecca Marie.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	131 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-07, Section: B.
Contained By:	Dissertations Abstracts International83-07B.
標題:	Electrodes. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28928210
ISBN:	9798762121569

Topics on Functional Variable Selection with Application to EMG Data Analysis.
North, Rebecca Marie.

Topics on Functional Variable Selection with Application to EMG Data Analysis. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 131 p.

Source: Dissertations Abstracts International, Volume: 83-07, Section: B.

Thesis (Ph.D.)--North Carolina State University, 2021.

This item must not be sold to any third party vendors.

Robotic hand prostheses require a prosthesis controller to translate electromyogram (EMG) signals into the user's desired movement. State-of-the-art controllers must undergo extensive training on data from a large number of EMG sensors, whereas a biomechanical model for a single movement degree-of-freedom shows that relatively few forearm muscles are needed to explain hand movement. A prosthesis controller based on such a biomechanical model should then require fewer EMG sensors to produce accurate predictions under a broad set of conditions. Using data collected by the North Carolina State University Department of Biomedical Engineering, Stallrich et al. [2020] proposed a dynamic scalar-on-function model to predict hand velocity with EMG signals and a multi-stage, adaptive penalized regression procedure to simultaneously perform sensor selection and EMG effect estimation. Although the method performed well overall, certain aspects of the data collection, processing, and model fitting raise important research questions.The EMG data were collected with surface electrodes that were positioned fairly close to one another. This close proximity produces highly correlated signals due to cross-talk between sensors, and strong multicollinearity is known to increase the variance of coefficient estimates. This leads to unstable estimates that may switch signs and be difficult to interpret, reduced statistical power, and difficulty specifying the correct model. Moreover, the data were transformed into a usable format using a "sliding window" process, where consecutive time windows of observations were stacked to form the data matrix. This process also results in autocorrelated responses which may also cause model selection issues if not properly accounted for. Stallrich et al. [2020] thinned the data to reduce this potential correlation, but this thinning reduces the number of observations and potentially loses important information. Since the proposed fitting procedure utilizes penalized linear regression, there are concerns as to how the EMG signals' multicollinearity and the response's potential autocorrelation might affect support recovery and estimation performance. These concerns motivated the first part of this dissertation that develops a number of diagnostic tools that can help practitioners determine the suitability of a given data set for analysis with two penalized regression methods, the lasso and the group lasso, in order to make reliable inference.The second component of this dissertation considers the form of the penalty used by Stallrich et al. [2020] as well as the chosen optimization algorithm. Specifically, the current penalty jointly imposes sparsity and smoothness on the EMG effects, which conflates the behaviors and interpretations of the corresponding tuning parameters. With the goal of reducing the number of stages required for accurate selection, estimation, and prediction, two alternative penalties that separately impose sparsity and smoothness are investigated. In order to efficiently fit the model with these penalties, an alternate algorithm is also utilized and shown to greatly reduce overall computational expense.The final part of this dissertation recognizes that the placement of the EMG sensors results in collection of redundant information. This is a result of electrodes detecting signals from neighboring muscles, not just their intended muscles. Thus, the muscle contractions are viewed as latent factors with the observed EMG signals being surrogate measurements.

ISBN: 9798762121569Subjects--Topical Terms:

629151
Electrodes.

Topics on Functional Variable Selection with Application to EMG Data Analysis.
LDR:04689nmm a2200337 4500 001 2351968
005 20221111113726.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798762121569
035 $a (MiAaPQ)AAI28928210
035 $a (MiAaPQ)NCState_Univ18402039110
035 $a AAI28928210
040 $a MiAaPQ $c MiAaPQ
100 1 $a North, Rebecca Marie. $3 3691571
245 1 0 $a Topics on Functional Variable Selection with Application to EMG Data Analysis.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 131 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-07, Section: B.
500 $a Advisor: Huang, He;Stefanski, Leonard;Xiao, Luo;Staicu, Ana-Maria;Stallrich, Jonathan.
502 $a Thesis (Ph.D.)--North Carolina State University, 2021.
506 $a This item must not be sold to any third party vendors.
520 $a Robotic hand prostheses require a prosthesis controller to translate electromyogram (EMG) signals into the user's desired movement. State-of-the-art controllers must undergo extensive training on data from a large number of EMG sensors, whereas a biomechanical model for a single movement degree-of-freedom shows that relatively few forearm muscles are needed to explain hand movement. A prosthesis controller based on such a biomechanical model should then require fewer EMG sensors to produce accurate predictions under a broad set of conditions. Using data collected by the North Carolina State University Department of Biomedical Engineering, Stallrich et al. [2020] proposed a dynamic scalar-on-function model to predict hand velocity with EMG signals and a multi-stage, adaptive penalized regression procedure to simultaneously perform sensor selection and EMG effect estimation. Although the method performed well overall, certain aspects of the data collection, processing, and model fitting raise important research questions.The EMG data were collected with surface electrodes that were positioned fairly close to one another. This close proximity produces highly correlated signals due to cross-talk between sensors, and strong multicollinearity is known to increase the variance of coefficient estimates. This leads to unstable estimates that may switch signs and be difficult to interpret, reduced statistical power, and difficulty specifying the correct model. Moreover, the data were transformed into a usable format using a "sliding window" process, where consecutive time windows of observations were stacked to form the data matrix. This process also results in autocorrelated responses which may also cause model selection issues if not properly accounted for. Stallrich et al. [2020] thinned the data to reduce this potential correlation, but this thinning reduces the number of observations and potentially loses important information. Since the proposed fitting procedure utilizes penalized linear regression, there are concerns as to how the EMG signals' multicollinearity and the response's potential autocorrelation might affect support recovery and estimation performance. These concerns motivated the first part of this dissertation that develops a number of diagnostic tools that can help practitioners determine the suitability of a given data set for analysis with two penalized regression methods, the lasso and the group lasso, in order to make reliable inference.The second component of this dissertation considers the form of the penalty used by Stallrich et al. [2020] as well as the chosen optimization algorithm. Specifically, the current penalty jointly imposes sparsity and smoothness on the EMG effects, which conflates the behaviors and interpretations of the corresponding tuning parameters. With the goal of reducing the number of stages required for accurate selection, estimation, and prediction, two alternative penalties that separately impose sparsity and smoothness are investigated. In order to efficiently fit the model with these penalties, an alternate algorithm is also utilized and shown to greatly reduce overall computational expense.The final part of this dissertation recognizes that the placement of the EMG sensors results in collection of redundant information. This is a result of electrodes detecting signals from neighboring muscles, not just their intended muscles. Thus, the muscle contractions are viewed as latent factors with the observed EMG signals being surrogate measurements.
590 $a School code: 0155.
650 4 $a Electrodes. $3 629151
650 4 $a Prostheses. $3 3683608
650 4 $a Amputation. $3 1066918
650 4 $a Sensors. $3 3549539
650 4 $a Biomechanics. $3 548685
650 4 $a Neurosciences. $3 588700
650 4 $a Biomedical engineering. $3 535387
650 4 $a Robotics. $3 519753
690 $a 0648
690 $a 0541
690 $a 0317
690 $a 0771
710 2 $a North Carolina State University. $3 1018772
773 0 $t Dissertations Abstracts International $g 83-07B.
790 $a 0155
791 $a Ph.D.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28928210