Commit fe4deb20 authored by Julius Welzel's avatar Julius Welzel
Browse files

Replace 2_methods

parent a42c951a
\documentclass[../main.tex]{subfiles}
\begin{document}
%doi:10.1093/cercor/bhr325
% doi:10.1093/cercor/bhr325
% doi: 10.3389/fpsyg.2018.02289
\subsection{Participants}
Thirty-four healthy participants took part in this EEG-study. Numbers were equally distributed in the younger sample (n = 17, 10 = female, mean age: 24 = years, SD = 5.2 years) and the elderly sample (n = 17, 10 = female, mean age: 63 = years, SD = 8.4 years), with everyone being right-handed and showed no history of neurological problems (see Table \ref{table:screening} for detailed screening overview). Participants received a compensation of 10 \euro per hour and gave their written consent before any proceedings of the study were conducted. The study was approved by the local Medical Ethics Committee of the University of Oldenburg, Germany and followed the Declaration of Helsinki.
Two groups took part in the study, one consisting of older adults (n= 17, mean age: 67±6.7 years) and one of younger adults (n= 17, mean age: 25±3.6 years). All participants were right-handed, as indicated by the Edinburgh Handedness Inventory (EHI), and showed no past or present neurological problems. Participants received a compensation of 10 \euro \ per hour and gave their written consent before any proceedings of the study were conducted. The study was approved by the local Medical Ethics Committee of the University of Oldenburg, Germany and followed the Declaration of Helsinki.
% Table 1: Exclusion
\begin{table}[h!]
\renewcommand{\arraystretch}{1.4}
\centering
\begin{tabular}{l l}
\hline
\textbf{Exclusion Criteria} & \\ \hline
Handiness & Medication \\ \hline
Cognitive function & Sleep (\textless 6 hours) \\ \hline
\end{tabular}
\caption{Overview screening}
\label{table:screening}
\end{table}
\subsection{Experimental setup}
Previous to EEG recordings, participants completed a neuropsychological test battery (Table \ref{table:questionaires}). During the EEG recording, everyone followed a protocol according to \cite{Allami.2008} with a few adaptations.
\begin{table}[h!]
\renewcommand{\arraystretch}{1.4}
\centering
\begin{tabular}{llll}
\hline
\textbf{TIME} & 4s & self paced & 3s \\ \hline
\textbf{EEG} & default & ERD & ERS \\ \hline
\textbf{TASK} & rest & MI & rest \\ \hline
\textbf{PANE} & opaque & transparent & opaque \\ \hline
\end{tabular}
\caption{trial}
\label{table:trial}
\end{table}
\subsection{Neuropsychological assessment}
All participants underwent a testing of a neuropsychological battery
% Table 2: Questionaires
\begin{table}[h!]
\renewcommand{\arraystretch}{1.4}
\centering
\begin{tabular}{l l l l l}
\hline
\textbf{Group} & \multicolumn{4}{l}{\textbf{Questionaires}} \\ \hline
Elderly participants & FAL & KVIQ & EHI & MoCA \\ \hline
Younger participants & FAL & KVIQ & EHI & \\ \hline
\end{tabular}
\caption{Overview Questionaires per group}
\label{table:questionaires}
\end{table}
None of the participants had a history of neurological or psychiatric illness. Participants from the elderly group did not show significant impairment in the Mild Cognitive Impairment Assesment (MoCA).
\nomenclature{\textbf{MoCA}}{Mild Cognitive Impairment} % include in abbreviations
\subsection{Task}
Previous to EEG recordings, participants completed multiple questionnaires. The "Fragebogen zur Ausgangslage" (FAL) was used to evaluate the current condition of the volunteers and their clinical history relevant to the experiment. By completing the Kinesthetic and Visual Imagery Questionnaire \citep{malouin2007kinesthetic} the individual's ability of MI was assessed and the difference between visual and kinesthetic MI was introduced. To ensure older participants were able to successfully complete the experiment, they underwent a cognitive screening using the Montreal Cognitive Assessment \citep{nasreddine2005montreal}. None of the older participants scored below a critical cutoff \citep{carson2018re}, allowing to include everyone in the study.
The EEG part of the experiment consisted of three experiments, which will be defined in more detail in the following section. For data analysis only the EEG-data of the Visuo-Motor-Task (\Cref{MI-training}) was included.
\subsubsection{Visuo-Motor-Task} \label{MI-training}
The first task of the EEG-experiment was a complex visuo-motor task, which was reproduced and adopted from \citet{Allami.2008}. Participants were seated in front of a table with both hands resting palm down on either side of a window pane. A small object was located behind the window pane and could take different orientations on the table surface.
% Figure: object
\begin{figure}
\vspace*{2em} %add vertical space
\centering
\includegraphics[width=\textwidth]{images/2_methods/block.jpg}
\caption{Object in Visuo-Motor-Task}
\label{fig:object_MI}
\end{figure}
As soon as the window pane turned transparent, participants were asked to grab the object with the right thumb and index finger and transport it to a end position as fast a possible, located in front of the window pane, before returning their hand to the initial starting position. A button press of the participant with the index finger indicated the end of the trial. To increase task difficulty, a marble was placed on a slight hole on the objects upper surface. A trial was counted as invalid, if the marble was dropped.
The experiment consisted of four blocks in total. The first consisted of eight trials of motor execution (ME) to initially get familiar with the movement. The second and third block included 40 trials of MI. The final block consisted of again eight trials of ME. Between all blocks participants did a 1-2 minute break and filled out a questionnaire about their status of fatigue and motivation. As the two groups were recorded in different studies, the experimental design differed slightly in the number of MI blocks. (cf. \cref{table:table_exp} for further details). As trials were self paced, the duration for the whole task varied between 35 and 50 minutes.
% Table: Trials in groups
\begin{table}[ht]
\vspace*{2em} %add vertical space
\centering
\begin{threeparttable}
\caption{Procedures EEG Experiment 1}
\label{table:table_exp}
\begin{tabular}{l l l l l}
\toprule
\textbf{Group} & \multicolumn{4}{l}{\textbf{Number of trials per block}} \\ \hline
Old & 8 trials ME & 40 trials MI & 40 trials MI & 8 trials ME \\ \hline
Young & 8 trials ME & 40 trials MI & 5x40 trials MI \tabfnm{*} & 8 trials ME \\ \bottomrule
\end{tabular}
\begin{tablenotes}[para,flushleft]
{\small
\tabfnt{*} This sample did six blocks of 40 MI trials in total.
}
\end{tablenotes}
\end{threeparttable}
\end{table}
\subsubsection{Limb Lateralisation Task} \label{LLT}
The second task of the EEG experiment was a mental rotation task of extremities. We used custom made 3D hand and feet models. All stimuli were rotated over three axes. \citep{ter2010mental}. A total of 96 different stimuli resulted from all possible combinations (view [palm, back] x extremities [hand, foot] x side [left, right] x in-depth rotation [0°, 60°, 300°] x in-plane rotation [60°, 120°, 240°, 300°]). The task consisted of one training block (10 trials) and three experimental blocks (3x32 stimuli). Stimuli were presented once in random order using custom developed software in Presentation (Neurobehavioral systems, Albany, USA) on a 15" LCD monitor at an eye screen distance of 80 cm. In the training block, ten stimuli were presented, randomly drawn from the full set, to familiarise participants with the task. Their exercise was to determine as fast as possible if the stimuli presented belonged to the right or left side of the body. The verbally formulated answer was recorded using Audacity 2.3 (Audacity Software, GPL) and additionally coded via a keyboard by the experimenter.
\subsection{Signal recordings}
The EEG and EMG data recordings were performed using three 32-channel BrainAmp amplifiers (BrainProducts, Gilching, Germany) and AGg/AgCl ring electrodes. Data were obtained with an amplitude resolution of 0.1 $\mu$V and a sampling rate of 500 $Hz$ with online analogue filter settings of 0.016 to 250 $Hz$. EEG data were recorded from 64 equidistant scalp sites using a central frontopolar site as ground and a nose-tip reference (\Cref{fig:cha_lay}). Two electrodes were placed below the eyes to capture eye artifacts. Surface EMG signals were recorded with two bipolar channels on both arms, by placing electrodes at proximal end of the opponens pollicis and over the muscle belly of the flexor digitorum superficialis on either side. The ground electrode for the EMG recording was placed on the left clavicle. Electrode impedances were tried to keep below 10 k$\Omega$ for the EEG and below 100 k$\Omega$ for the EMG data acquisition.
% Figure: chanl layout
\begin{figure}[ht]
\vspace*{2em} %add vertical space
\centering
\includegraphics[width=\textwidth]{images/2_methods/chan_lay.png}
\caption{Channel layout}
\label{fig:cha_lay}
\end{figure}
\vspace{1em}
In the following all relevant steps of the data analysis will be outlined, as defined by the COBIDAS initiative \citep{pernet2018best}. Data analysis was done with EEGLAB 14.1.2b \citep{delorme2004eeglab} and self written scripts in the MATLAB 2018b environment (Mathworks). Scripts for full data analysis are publicly available at GitLab \citep{Welzel2019}.
\vspace{1em}
\subsection{EMG data processing} \label{EMG analysis}
EMG data were offline high-pass filtered at 25$Hz$ (Hamming windowed, finite impulse response (FIR), filter order: 264) and segmented into 96 epochs. Each epoch corresponded to the respective trial described in \cref{MI-training}. Each epoch began with the window turning fully transparent and ended with the participant pressing the button, indicating the end of the trial. For the purpose of cleaning movement contaminated EEG epochs, movement needed to be reliably detected in the EMG epochs. Traditionally this can be done by visual inspection. However, for individual most reliable movement detection we used an automated approach with an implemented Evolution Strategy \citep{kramer2017genetic}. In a first step we substracted some samples $BP$ (Button Press) from the end of the epoch to remove actual movement of the button press. In a second step, we applied a moving SD (Standard Deviation) filter (window size: 125 samples) to every epoch. If the moving SD value exceeded a multiple $N$ of the SD of the same trial, it was marked as an artefact. This method proved to be sufficient to detect movement, as long as participants were comparable in muscle tonus and arbitrary movement. However, as our participants were a heterogeneous group, the threshold $N$ and the number of samples $BP$ had to be adopted individually for more comparable results. To do this, we applied an ES to learn the individual most reliable $BP$ time to substract at the end of the epoch and multiplier $N$ of the threshold.
\subsection{EEG data processing} \label{eeg_data_pro}
To identify artefacts in the EEG data, we combined data from the first (\Cref{MI-training}) and the second EEG experiment (\Cref{LLT}) as artefacts were expected to be comparable. First we deleted data between the experimental blocks, because participants might have induced artefacts we would not expect during the trials of the experiment. In a following step data was high-pass filtered at 1$Hz$ with (Hamming windowed, FIR, filter order automated with EEGLAB) and downsampled to 250. Subsequently we segmented the data into consecutive fragments of 1$s$, whereafter an independent component analysis (ICA) based on the extended Infomax \citep{bell1995information} was applied. The resulting unmixing
weights were used to linearly decompose the original raw data and attenuate typical artifacts. A fully automated method to identify eye-movement related EEG components was used \citep{bigdely2013eyecatch}, the remaining artefactual components were identified by visual inspection.\\
As three parallel analyses of the EEG data were conducted, from now on relevant values for the broad band analysis will first be defined, further values for the $\mu$ and $\beta$ band will be in parenthesis.
EEG data were offline high-pass filtered 8$Hz$ [8$Hz$, 15$Hz$] with (Hamming windowed, FIR, filter order automated with EEGLAB), downsampled to 100 and low-pass filtered at 30$Hz$ [12$Hz$, 30$Hz$] with (Hamming windowed, FIR, filter order automated with EEGLAB). Bad channels were detected using EEGLAB inbuilt functions \citep{delorme2004eeglab} and interpolated if removed. Data were then re-referenced to an average reference.
The ICA cleaned and filtered data was segmented in epochs of individual length, starting 2.5 $s$ before the window pane turned transparent to when the participant pressed the button. Epochs marked as artefacts by the EMG analysis (\cref{EMG analysis}) were discarded from further analysis. In a next step, the task relative Event Related Desynchronisation (ERD) was calculated for every epoch for every channel at timepoint $t$ in accordance to \citet{pfurtscheller1999event}:
\begin{equation}
\vspace{0em}
{ERD(t)} = \frac{{Power(t)} - avg. Power_{BL}}{avg. Power_{BL}} \times 100
\vspace*{1em} %add vertical space
\end{equation}
The baseline (BL) for the ERD calculation was chosen to be 2 $s$ to 0.5 $s$ before the window turned transparent. An exemplary illustration of the time course can be seen in \cref{fig:overview_EP}. All samples of the MI period of each trial were averaged for every channel, for every participant, resulting in a 34 x 64 x 96 matrix [Subjects x channel x trials].
% Figure: epoch
\begin{figure}[ht]
\centering
\includegraphics[width=\textwidth]{images/2_methods/epoch.pdf}
\caption{Illustration of an epoch. State of the window during the trial presented in top row, corresponding times in bottom row}
\label{fig:overview_EP}
\end{figure}
\subsection{Analysis of spatial pattern}
In this section, a detailed explanation on how to derive the difference in spatial distribution of the topographies for the MI part of the experiment will be given.
The ensuing list should serve as an inital overview. Corresponding matrix size will be given in brackets.
\renewcommand{\labelenumii}{\roman{enumii}}
\begin{enumerate}[label=\textbf{S.\arabic*}]
\setlength\itemsep{0em}
\item Inital ERD data matrix [34x64x96]
\item Average ERD data for all MI trials [34x64]
\item \label{z_score_l} $z$-Score MI trials [34x64]
\item \label{interpol_l} Interpolate $z$-scored ERD values [34x40.000]
\item Plot 3D representation of interpolated trial ERD map [34x40.000]
\item \label{intersect_l} Intersect map from maximal to minimal $z$ value $N$ times [34x40.000x$N$]
\item \label{estimate_l} Estimate area of intersections [34x$N$]
\end{enumerate}
After averaging the MI trials for every participant, the data was $z$-scored (\cref{z_score_l}) to obtain a zero-mean for every channel over all participants. We did this to make single trial maps comparable across participants. In a next step the data of 64 scalp site channels was interpolated using Cubic Delaunay Triangulation \citep{barber1996quickhull,watson1992contouring} to a 200x200 grid which resulted in higher spatial resolution (\cref{interpol_l}). The interpolated data was then plotted in a 3D representation. Accordingly, there was one plot for each participant. These plots were intersected horizontally $N$ times in plane with the z-score axis. The number of intersections $N$ was derived from the spatial Nyquist frequency of our EEG system (\cref{intersect_l}). Considering the distance between electrodes as $d$, the spatial frequency which can detect changes in space reliably is calculated as follows \citep{srinivasan1998estimating}:
\begin{equation}
{f_{spat}} = \frac{d_{max}}{d_{min}} \times 2
\vspace*{1em} %add vertical space
\end{equation}
This spatial resolution is far lower than what one can get with fMRI, but half of what is maximally possible with EEG \citep{srinivasan1998estimating}. Concluding, we calculated $N$ = 24 intersection areas from the maximal to the minimal $z$-Score of every map. This was achieved by dividing $z_{max}-z_{min}$ in 24 equal parts, which gave us 24 $z$-values. We could then generate contour lines of the intersections, at those 24 $z$-levels from the 3D plots and estimate the area enclosed by the contour lines by simply using Pythagoras' theorem (\cref{estimate_l}). A scheme of the procedure (with only 6 intersctions for illustrative purposes) can be seen in \cref{fig:3D_topo}. Summarizing, we ended up with a 34x24 matrix, which contained the 24 intersection area values of all participants averaged MI ERD maps.
% Figure: ES setup
\begin{figure}[ht]
\centering
\includegraphics[width=\textwidth]{images/2_methods/3D_top_young_broad.png}
\caption{3D representation of interpolated topography. Contour lines indicate area of intersection between 3D plot and planes.}
\label{fig:3D_topo}
\end{figure}
\subsection{Statistical Analysis}
All statistical analysis were performed using Matlab (Mathworks). To see if the EMG analysis was comparable between the two groups, two-tailed $t$-tests were conducted. The number of excluded trials were compared as well as the individual $BP$-times and the threshold of the SD estimated my the ES.
To determine if the area of the ERD for the MI task differed between the young and the old participants, a distribution from the number of intersections was estimated. The sum of all intersection areas at the full width at half maximum ($FWHM$) was used, to make a fair comparison between the groups total ERD areas. To determine if there is a statistical difference, between the groups, a two-tailed $t$-test were conducted on the areas and on the parameters of the estimated underlying normal distributions.
The subjects individual sum of areas was correlated to their difference between trial time of MI \& ME.
A Spearman-rank correlation was conducted on the individual sum of areas with the age of the participant.
\subsection{EEG Recordings}
The technical setup can be seen in Figure \ref{fig:setup_MI} which was used for the main MI experiment.
\begin{figure}
\centering
\includegraphics[width=\textwidth]{images/setupMI}
\caption{Technical setup of the MI Experiment}
\label{fig:setup_MI}
\end{figure}
\end{document}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% ARCHIVE
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Statistical Analysis}
% Timing and synchronisation
% The ensure the correct timing and synchronisation of the full experimental setup, the table had an adruino attached to control the opacity of the window pane and the send trigger to the network to record all data streams simultaneously using Lab Stream Recorder.
\end{document}
\ No newline at end of file
% Figure: MI setup
% \begin{figure}
% \vspace*{2em} %add vertical space
% \centering
% \includegraphics[width=\textwidth]{images/2_methods/setup_MI.png}
% \caption{Technical setup of the MI Experiment}
% \label{fig:setup_MI}
% \end{figure}
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment