--- title: "diegr" output: rmarkdown::html_vignette vignette: > %\VignetteIndexEntry{diegr} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8} --- ```{r, include = FALSE} knitr::opts_chunk$set( collapse = TRUE, eval = FALSE, comment = "#>" ) #> library(rgl) #> knitr::knit_hooks$set(webgl = hook_webgl) ``` ```{r setup} #| echo: false #| eval: true library(diegr) library(rlang) ``` The package `diegr` displays high-density electroencephalography (HD-EEG) data in different ways using interactive elements or animations for a comprehensive overview of data. ### Installation ```{r} #| eval: false # development version: # install.packages("devtools") devtools::install_github("gerslovaz/diegr") # install.packages("diegr") # after deploying to CRAN ``` ### Data The package contains built-in datasets. Due to the nature of the experiment and the size of the data, this is only a small section which serves as a demonstration of the implemented functions and it is not possible to reconstruct the real signal of a specific sensor or subject from it. Details of the experiment from which the original data comes can be found in Madetko-Alster and col., 2025. #### epochdata The brain signal originally measured by 256-channel HydroCel Geodesic Sensor Net with sampling frequency 250 Hz. Example dataset contains amplitude values from 204 selected channels at 50 time points (stimulus at time point 10) for 2 representative subjects (one healthy control, one patient). From 50 epochs per subject, 14 (or 15) epochs were selected. The data is organized into a data frame with the following columns: - `time`: Number of time point. Number 10 indicates the time of stimulus (0 ms), the interval between two time points corresponds to 4 ms (sampling frequency 250 Hz). - `signal`: HD-EEG signal amplitude (in microvolts). - `epoch`: Epoch number (there are 14 epochs for subject one, 15 epochs for subject two). - `sensor`: Sensor label (labeling corresponds to the Geodesic Sensor Net Technical Manual, 2024). - `subject`: Subject ID (1 - representative healthy control subject, 2 - representative patient subject). #### HCGSN256 The Cartesian coordinates of HD-EEG sensor positions in 3D space on the scalp surface and their projection into 2D space according to 256-channel HydroCel Geodesic Sensor Net average template montage. The data is organized into a list with the following elements: - `D2`: A tibble with coordinates and labels of sensors in 2D. - `D3`: A tibble with coordinates and labels of sensors in 3D. - `ROI`:A vector with the region of interest (ROI) assignment for each sensor ("central", "frontal", "occipital", "parietal" or "temporal"). - `pairedname`: Alternative sensor labels corresponding to left/right paired electrodes (with C for midline). Sensors are labeled according to Geodesic Sensor Net Technical Manual, 2024. Axis orientation (3D): - x: left (-) to right (+); - y: posterior (-) to anterior (+); - z: inferior (-) to superior (+). #### rtdata The response time values in individual experiment epochs for 2 representative subjects (one from patient, one from healthy control group). The epochs and subjects correspond to the sample dataset `epochdata`. The data is organized into a data frame with the following columns: - `subject`: Subject ID (1 - representative healthy control subject, 2 - representative patient subject). - `epoch`: Epoch number (there are 14 epochs for subject one, 15 epochs for subject two). - `RT`: Response time in ms. ### Tools useful in preprocessing The package assumes that the input data have already been preprocessed (artifact removal, independent component analysis - ICA, segmentation, ...), but may still contain outliers that should be excluded for specific statistical analyses. The useful tool in outliers detection is interactive plot with boxplots of amplitudes (`boxplot_epoch()`) or response times (`boxplot_rt()`) in individual epochs. The `plotly` output makes it easy to determine the epoch number from which outliers come and additionally allows editing of the output image layout. There is also an interactive graph of amplitudes in individual epochs with (or without) the average plotted (`interactive_waveforms()`). This function can also be used to compare signal for different sensor or subjects according to the `level` argument setting. ```{r} # in interactive boxplot of signal we choose subject, # channel and concrete time range, # in which the boxplots should be plotted # example for subject 1 and channel E34 in time points from 10 to 20 boxplot_epoch(epochdata, amplitude = "signal", subject = 1, channel = "E34", time_lim = 10:20) # customizing the output title and axis labels using |> operator p1 <- boxplot_epoch(epochdata, amplitude = "signal", subject = 1, channel = "E34", time_lim = 10:20) p1 |> plotly::layout(xaxis = list(title = ""), yaxis = list(title = "Amplitude"), title = list(text = "Outlier detection for subject 1 and channel E34", font = list(size = 11, color = "darkblue")) ) # note: the output is not plotted to reduce the vignette file size ``` Outlier epochs can be easily detected using the hover text in the `plotly` output. In the case above, the outlier is epoch 14. The results can be supported computationally by the `outliers_epoch()` function, which allows the evaluation of outliers using one of three criteria (interquartile range - IQR, percentile approach or Hampel filter). The output of this function is a table containing epoch IDs and the number of time points in which the epochs were evaluated as outliers (according to chosen criteria). ```{r} # outlier detection for subject 1 and channel E34 using IQR method outliers_epoch(epochdata, subject = 1, sensor = "E34", method = "iqr") ``` The results correspond with findings from `boxplot_epoch()` - epoch 14 was evaluated as an outlier in all 50 time points. Analogously to function `boxplot_epoch()`, function `boxplot_subject()` works to identify outliers at the subject level. ### Preparing signal for visualisation The main focus of the package is interactive visualization and animation of the HD-EEG data. This often requires first baseline correcting the data or averaging the signal across epochs, sensors/time points, or subjects. The baseline correction of the amplitude is enabled through the function `baseline_correction()`. In the first release of the package, only most commonly used absolute correction (`type = "absolute"`), which subtracts mean over `base_int`, is available. The calculation of the average signal can be performed using the function `compute_mean()`, which computes pointwise or jackknife (leave one out) mean of the signal at different levels. In addition to the mean itself, the function also calculates the boundaries of the point confidence interval (`ci_low` and `ci_up` in the output). The argument `group` determines whether the average is calculated within individual time points (for curves) - `group = "time"` or sensors (for topographic maps) - `group = "space"`. ```{r} #| eval: true # preparing data for later visualisation # filtering signal without outlier epochs 14 and 15 # and computing epoch level jackknife average from baseline corrected data edata <- epochdata |> dplyr::filter(subject == 2, epoch %in% 1:13) edata <- baseline_correction(edata, baseline_range = 1:10) s1 <- compute_mean(edata, amplitude = "signal_base", time = 10, level = "epoch", group = "space", type = "jack") ``` ### Temporal domain visualization The `interactive_waveforms()` and `plot_time_mean()` can be used to display the time course of the signal. While function `interactive_waveforms()` interactively displays individual waveforms at different levels via `plotly`, function `plot_time_mean()` is adapted directly to display the average signal including the pointwise confidence interval as a `ggplot` object. ```{r, fig.alt="An average amplitude time-series plot showing the brain's electrical activity (in microvolts) over time (in milliseconds), time-locked to a stimulus event at 0 ms. The red line represents the average amplitude, and the shaded red area represents the confidence interval."} #| eval: true # computing jackknife average from channel E65 s2 <- compute_mean(edata, amplitude = "signal_base", channel = "E65", level = "epoch", group = "time", type = "jack") # plotting the average, the zero time point (stimulus) is set to 10 plot_time_mean(s2, t0 = 10) ``` ### Spatial visualization #### Mesh Generation The original electrode grid is too coarse for graphical rendering. To achieve smoother and more visually appealing results in plots, it's necessary to generate a denser mesh of points. This refined mesh provides better spatial resolution and enhances the quality of visualizations. Creating a new mesh is available through `point_mesh()` function. This function computes the coordinates of mesh in 2D and/or 3D and enables to control a density (argument `n`) or the shape (argument `type`) of the result mesh. The default setting creates a polygonal mesh with starting number 10000 points and computes the coordinates in both dimensions, which are available in `$D2` and `$D3` parts of a result list (class `"mesh"`). For simple created mesh plotting use the function `plot_point_mesh()`, which results in two dimensional `ggplot` of point mesh or three dimensional `rgl` plot depending on the input dimension. Notes: - Creating and plotting a mesh for subset from the whole template montage is available through `sensor_select` parameter. For consistency between created mesh and its plot use the same vector of sensor labels in both functions. - The input of `plot_point_mesh()` must be only the relevant part with coordinates - `$D2` for two dimensional and `$D3` for three dimensional plot. - If the number of mesh points is large, they are no longer visible in the plot and the mesh looks like a solid surface (as can be seen below in first plot). - Depending on the hardware used, the computational cost of default `n = 10000` can be high. In this case we recommend to reduce `n` and to reuse a precomputed mesh across plots for speed. ```{r, fig.show='hold', fig.alt="A schematic representations of an HD-EEG sensor net, showing the arrangement of 204 labeled channels distributed across the entire scalp area together with denser point mesh created for interpolation."} #| eval: true # creating a mesh with default settings sensors204 <- unique(epochdata$sensor) M1 <- point_mesh(template = "HCGSN256", sensor_select = sensors204) # plot output in 2D plot_point_mesh(M1$D2, sensor_select = sensors204) # creating a circular mesh, only 2D coordinates M2 <- point_mesh(dimension = 2, n = 3000, template = "HCGSN256", sensor_select = sensors204, type = "circle") # plotting a mesh - function allows different options of the result plot plot_point_mesh(M2$D2, sensor_select = sensors204, col_sensors = "purple", label_sensors = TRUE, cex = 0.1) ``` `diegr`'s plotting functions also allow users to input the custom mesh created in other way, it is only necessary to maintain the structure following a `"mesh"` object. If the `mesh` input is not specified, the graphic functions automatically use the mesh created by `point_mesh()` with default settings. #### Topographical maps in 2D projection The `topo_plot()` function is used to plot a topographic map of the brain signal, the amplitude values are colored using a topographic colour scale (unless users choose otherwise). An alternative to this function adapted directly to plot the mean along with the confidence interval bounds is `plot_topo_mean()`. In the default colour scale, amplitude = 0 is mapped to the blue–green boundary; positive values trend towards yellow/red, negative towards dark blue. Both functions allow setting some optionally arguments for editing the output appearance (with or without contours, sensor names, etc.) and return `ggplot` object for further customization. Notes: - The input values of amplitude (or also CI bounds in the case of the `plot_topo_mean()` function) outside the chosen `col_range` will cause "holes" in the resulting plot. - To compare results for different subjects or conditions, setting the same values of `col_range` and `col_scale` arguments in all cases is necessary. ```{r, fig.show='hold', fig.alt="Fig 1: A top-down topographic map of a high-density EEG amplitude in topographic colour scale with contours and black points on sensor locations. The amplitude legend is on the left side of the scalp projection. Fig 2: The same plot without contours and legend, but with sensor labels as text displayed together with black point on sensor positions. Fig 3: A ggplot divided into three panels with topographic maps of average signal (middle) and its lower (left) and upper (right) confidence interval bounds. The colour legend position is bottom of the facets."} #| eval: true # Plot average topographic map of baseline corrected signal for subject 1 # from the time point 10 (the time of the stimulus) # the outlier epoch 14 is extracted before computing average # preparing data edata <- epochdata |> dplyr::filter(subject == 1 & time %in% 1:10) data_base <- baseline_correction(edata, baseline_range = 1:10) data_mean <- compute_mean(data_base, amplitude = "signal_base", time = 10, type = "point", ex_epoch = 14, group = "space") # plotting the base topographic polygon map with contours and legend # interval (-30,15) is selected in consideration of the signal progress topo_plot(data = data_mean, amplitude = "average", col_range = c(-30, 15), contour = TRUE) # plotting the same map without contours and legend # but with sensor labels and adding the title g1 <- topo_plot(data = data_mean, amplitude = "average", col_range = c(-30, 15), label_sensors = TRUE, show_legend = FALSE) g1 + ggplot2::ggtitle("Subject 1, time of the stimulus") # plotting the average together with CI bounds using plot_topo_mean plot_topo_mean(data = data_mean, template = "HCGSN256", col_range = c(-30, 15)) ``` #### Topographical maps on the scalp The `scalp_plot()` function is created for plotting a scalp polygon map of the EEG signal amplitude using topographic colour scale (default scale is the same as in `topo_plot()`). The result plot is rendered using `rgl::shape3d` function and the signal interpolation between sensor locations is based on thin-plate spline interpolation model. Correct mesh triangulation is required for appropriate colour rendering. This triangulation is constructed by `make_triangulation()`, which creates a Delaunay type-I triangulation (Lai & Schumaker, 2007) with consistently oriented edges (Schneider & Eberly, 2003). The function allows setting the `col_range` argument, which is necessary to obtain comparable results between different subjects or conditions. Note: The input values of amplitude outside the chosen `col_range` will cause "holes" in the resulting plot. ```{r} # plotting the scalp polygon map of previously prepared signal s1 #open3d() scalp_plot(s1, amplitude = "average", col_range = c(-30, 15)) # note: the output is not plotted to reduce the vignette file size ``` The `rgl` output enables to zoom and rotate the image with default view of the back of the head. ### Animations The package includes the `animate_topo()` and `animate_topo_mean()` function for 2D topographic animation in time and `animate_scalp()` function for 3D scalp animation in time. Both of these functions allow to export the output, a gif animation using `gifski` package for 2D case and a .mp4 video format using `av` package for 3D case. The topographic map animation uses `gganimate` and and accepts additional parameters passed to `gganimate::animate()` via `...`. The most of input parameters of `animate_topo()` are the same as in `topo_plot()` function, parameters related to animation properties and export are added and also the `t_lim` parameter for setting the length of the timeline displayed below the animation. The `animate_topo_mean()` function is customized for topographic maps of average signal together with pointwise confidence interval bounds displayed as topographic maps on the sides of the average map. Usage is almost identical to the `animate_topo()` function, only input data must contain columns `average`, `ci_up` and `ci_low` (like the output from the `compute_mean` function). Example: The animation of time course of the EEG signal for subject 1 and epoch 5 (from `epochdata`) between time points 10 and 20 with the whole range of time 0:50 is created as follows (the output is not rendered to reduce the vignette file size): ```{r} # filtering the example data s1e05 <- epochdata |> dplyr::filter(subject == 1 & epoch == 5 & time %in% 10:20) # Plot animation with setting the range of time: animate_topo(s1e05, t_lim = c(0,50)) # Export the gif animation - this code will create the animation inside your working directory. # If you want to export it elsewhere, set the whole path in output_path parameter. animate_topo(s1e05, t_lim = c(0,50), output_path = "example_animation.gif") ``` The scalp animation enables to export a video in `.mp4` format using the `av` package or save a sequence of frames into chosen directory setting the `frames_dir` parameter. The animation speed can be controlled by the `sec` parameter in the viewer or by the `framerate` parameter for the exported video. The video resolution quality can be affected using `par3d` function, for example `rgl::par3d(windowRect = c(100, 100, 800, 800))`. Example of scalp animation with the same data as above (the output is not rendered to reduce the vignette file size): ```{r} # Plot animation with default mesh and triangulation: animate_scalp(s1e05, amplitude = "signal") # export the video - the .mp4 extension is required for correct functioning rgl::par3d(windowRect = c(100, 100, 800, 800)) animate_scalp(s1e05, amplitude = "signal", frames_dir = "your_frames_dir_path", output_path = "your_created_video_path.mp4") ``` If you are interested in animation of rotating scalp visualization in fixed time point, you can easily achieve this by using the combination of `rgl::play3d` and `rgl::spin3d` functions according to the following code. ```{r} #open3d() scalp_plot(s1, amplitude = "average", col_range = c(-30,15)) rgl::play3d(rgl::spin3d(axis = c(0, 0, 1), rpm = 10), duration = 10) # note: the output is not plotted to reduce the vignette file size ``` ### Step-by-step example: Comparing Topographic Maps of Two Subjects In this section we present a sample code for analyzing topographic maps across subjects: 1. Outliers detection (for each subject separately) ```{r} # using IQR method for selected sensor from central region outliers_epoch(epochdata, subject = 1, sensor = "E59", method = "iqr") outliers_epoch(epochdata, subject = 2, sensor = "E59", method = "iqr") # verify results by visual inspection interactive_waveforms(epochdata, amplitude = "signal", subject = 1, channel= "E59", level = "epoch", t0 = 10) interactive_waveforms(epochdata, amplitude = "signal", subject = 2, channel= "E59", level = "epoch", t0 = 10) ``` Subject 1: epoch 14 is an obvious outlier. Subject 2: epoch 15 is an obvious outlier, epoch 14 was denoted as outlier for 37 time points - we exclude both of them from further analysis. 2. Baseline correction (for each subject separately) We exclude outlier epochs before baseline correction, otherwise they would affect the baseline average computing. Note: `baseline_correction()` works with data of a specific subject. If there were multiple subjects in the input data, the baseline calculation would be done across all of them, which we want to avoid. ```{r} sub1clear <- epochdata |> filter(subject == 1, !epoch == 14) sub2clear <- epochdata |> filter(subject == 2, !epoch %in% c(14, 15)) sub1base <- baseline_correction(sub1clear, baseline_range = 1:10) sub2base <- baseline_correction(sub2clear, baseline_range = 1:10) ``` 3. Average computing We compute average topographic map in time point 35 (100 ms after stimulus) for both subjects. We use pointwise average with bootstrap confidence intervals. Note: `compute_mean()` allows for complex data at the input, from which we select using the parameter `subject`, but we have prepared separated data in previous steps, so we use them. ```{r} sub1mean <- compute_mean(sub1base, amplitude = "signal_base", time = 35, group = "space", level = "epoch", R = 1000) sub2mean <- compute_mean(sub2base, amplitude = "signal_base", time = 35, group = "space", level = "epoch", R = 1000) ``` 4. Average plotting To maintain consistency between both subjects, we first construct a mesh and colour scale that we will use in both figures. For plotting, we will use the `plot_topo_mean()` function, which is directly adapted for rendering topographic map of the average EEG amplitude together with its lower and upper confidence interval (CI) bounds. ```{r} # prepare a mesh for plotting M <- point_mesh(dimension = 2, n = 4000, template = "HCGSN256", sensor_select = unique(sub1mean$sensor), type = "polygon") # compute consistent col_range across subjects # a) find the range of average amplitude c(min(sub1mean$average, sub2mean$average), max(sub1mean$average, sub2mean$average)) # -15.634288 8.609518 # b) make the range symmetric cr_subjects <- c(-16,16) # plot the average topographic maps with the same color range for both subjects plot_topo_mean(sub1mean, M, template = "HCGSN256", col_range = cr_subjects, contour = TRUE) plot_topo_mean(sub2mean, M, template = "HCGSN256", col_range = cr_subjects, contour = TRUE) ``` #### References Lai M-J, Schumaker LL. *Spline functions on triangulations.* Cambridge University Press; 2007. Schneider PJ, Eberly DH. *Geometric Tools for Computer Graphics.* The Morgan Kaufmann Series in Computer Graphics. San Francisco: Morgan Kaufmann, 2003. Madetko-Alster N., Alster P., Lamoš M., Šmahovská L., Boušek T., Rektor I. and Bočková M. The role of the somatosensory cortex in self-paced movement impairment in Parkinson’s disease. *Clinical Neurophysiology.* 2025, vol. 171, 11-17. Wickham H, Bryan J. *R Packages (2e).* O'Reilly Media; 2023. Wickham H. *ggplot2: Elegant Graphics for Data Analysis.* Springer; 2016. Electrical Geodesics, Inc.: *Geodesic Sensor Net Technical Manual*. 2024. #### Reproducibility & System Requirements > **This document was compiled on `r Sys.Date()` using R version `r getRversion()` and the following package versions:** > > * **Primary Packages:** > * `diegr` (version `r packageVersion("diegr")`) > * `dplyr` (version `r packageVersion("dplyr")`) > * `rgl` (version `r packageVersion("rgl")`) > * `plotly` (version `r packageVersion("plotly")`) > * `gganimate` (version `r packageVersion("gganimate")`) > * `gifski` (version `r packageVersion("gifski")`) > * `av` (version `r packageVersion("av")`) > > * **rgl on Headless Systems:** > The `rgl` package requires specific setup for graphics on systems without a display (headless servers, like those often used for package checks). If you are reproducing this vignette on a headless system, ensure the following command is run **before** loading `rgl`: > > ```r > options(rgl.useNULL = TRUE) > # Alternatively, set the environment variable RGL_USE_NULL="true" > ``` > > * **Media Requirements:** > Interactive plots (from **plotly**) and animations (from **gganimate** and **gifski** / **av**) require the respective R packages and, for local rendering or specific output formats, may require external libraries or codecs. If media fails to render, check the documentation for `plotly`, `gganimate`, `gifski`, and `av` regarding system dependencies for video/GIF encoding. > > * **Full Session Details:** > For complete details on the software environment, please run the following command in your R session: > > ```r > sessionInfo() > ```