Skip to contents

Aesthetic refinements of ggbrain plots

Michael Hallquist

3 Oct 2022

Source: vignettes/ggbrain_aesthetics.Rmd
ggbrain_aesthetics.Rmd

This is still in progress

The sections below are incomplete!

Images that will be used in the demo

The package includes a few images that can be used for demonstrating its capabilities. The locations of the images (and the code to load them) are listed below.

# MNI 2009c anatomical underlay
underlay_3mm <- system.file("extdata", "mni_template_2009c_3mm.nii.gz", package = "ggbrain")

# onset of decision phase for learning task
decision_onset_3mm <- system.file("extdata", "decision_onset_zstat_3mm.nii.gz", package = "ggbrain")

# Parametric modulator: entropy change following feedback in learning task
echange_overlay_3mm <- system.file("extdata", "echange_overall_zstat_3mm.nii.gz", package = "ggbrain")

# Schaefer 200-parcel atlas of cortex
schaefer200_atlas_3mm <- system.file("extdata", "Schaefer_200_7networks_2009c_3mm.nii.gz", package = "ggbrain")

print(c(underlay_3mm, decision_onset_3mm, echange_overlay_3mm, schaefer200_atlas_3mm))
## [1] "/home/runner/work/_temp/Library/ggbrain/extdata/mni_template_2009c_3mm.nii.gz"          
## [2] "/home/runner/work/_temp/Library/ggbrain/extdata/decision_onset_zstat_3mm.nii.gz"        
## [3] "/home/runner/work/_temp/Library/ggbrain/extdata/echange_overall_zstat_3mm.nii.gz"       
## [4] "/home/runner/work/_temp/Library/ggbrain/extdata/Schaefer_200_7networks_2009c_3mm.nii.gz"

Converting to a ggplot object and displaying the plot

For reasons that are somewhat boring (having to do with how R handles S3 operators like +), the brain-related addition steps for ggbrain object must come before generic ggplot2 modifications of the plot. More specifically, the use of the + operator for ggbrain commands like geom_brain() or images() create and modify an object that is internal to the

Adjustments to the appearance of image layers

blur_edge

remove_specks

trim_threads

fill_holes 

  blur_edge = NULL,
  fill_holes = NULL, remove_specks = NULL, trim_threads = NULL)

Cleaning small ‘specks’ from image slices

At times, when slicing a given image (esp. functional activations), it is possible that some clusters are very small, yielding small ‘specks’ on some rendered slices. These are visually unappealing and may merit removal, depending on your tastes. Of course, removing these specks can, at the extreme, misrepresent the data, which is undesirable. This concern notwithstanding, you can use the clean_specks argument of images to specify the voxel threshold used to remove clusters smaller than a certain size. For example, clean_specks = 30 would remove any clusters smaller than 30 voxels in size from each slice on the rendered image.

# TODO: make this in geom_brain
gg_obj <- gg_obj + images(echange=echange_overlay_3mm)

Filling small holes in image slices

Another aesthetic problem is that at times there will be small holes on the interior of a cluster. These may reflect voxels that fall below a statistical threshold (e.g., voxels that have a z-statistic value of 3.02 in a cluster defined by z > 3.1). These can be filled in using the fill_holes argument of images. This specifies the size of holes (in voxels) that should be filled on the rendered slices. For example, fill_holes = 30 would fill holes of 30 or fewer voxels on the interior of each cluster. The holes are filled by nearest neighbor imputation in which the nearest non-missing voxels are used to impute missing voxels. In integer-valued/categorical images, the mode of the neighboring voxels is used for imputation.

# TODO: make this in geom_Brain
gg_obj <- gg_obj + images(echange=echange_overlay_3mm)

Labeling regions 

a <- ggbrain(bg_color = "gray60", text_color = "black") +
  images(c(underlay = "template_brain.nii.gz")) +
  images(c(dan_clust = dan_clust), fill_holes = TRUE, clean_specks = 20, labels = dan_labels) +
  # slices(paste0("x = ", seq(10, 90, 10), "%")) +
  slices(montage("axial", 10, min = 0.1, max = 0.8)) +
  # add_slice("y=50%", title = "hello", xlab = "testx", border_size = 1, border_color = "blue") +
  geom_brain(definition = "underlay", fill_scale = scale_fill_gradient(low = "grey8", high = "grey62"), show_legend = FALSE) +
  geom_brain(definition = "dan_clust", mapping = aes(fill = label), fill_scale = scale_fill_brewer("DAN 17 label", palette = "Set1"), show_legend = TRUE) +
  geom_outline(definition = "dan_clust", mapping = aes(group = roi_label), show_legend = TRUE, outline = "black") + # fill_scale = scale_fill_discrete("hello"),
  # geom_outline(definition = "dan_clust", show_legend = TRUE, mapping=aes(group=label), size=2, outline="cyan") + #fill_scale = scale_fill_discrete("hello"),

  # geom_outline(definition = "dan_clust", show_legend = TRUE, outline = "cyan") + #fill_scale = scale_fill_discrete("hello"),

  geom_region_label_repel(image = "dan_clust", label_column = "roi_label", min.segment.length = 0, size = 1.5, color = "black", force_pull = 0.2, max.overlaps = Inf) +
  render() + plot_layout(guides = "collect")

Saving ggbrain plots to file

Aesthetic mappings for

Combining ggbrain plots

[Principles of patchwork]

Other considerations

Resampling images to have the same resolution

Oftentimes, higher-resolution images (e.g., 1mm) yield better graphics since they potentially provide more detail. But fMRI data are often of lower resolution than T1-weighted images. If you wish to render your fMRI data on a higher-resolution anatomical template in ggbrain, you can first use AFNI’s 3dresample command to resample your functional data to match the anatomical template (underlay).

Here is an example of resampling a 3mm statistic image to a 1mm underlay using nearest neighbor interpolation. Note that the image to be resampled should be in the same sterotaxic space as the master/template image.

3dresample -input echange_overall_zstat_3mm.nii.gz \
  -prefix echange_overall_zstat_1mm.nii.gz \
  -master mni_template_2009c_1mm.nii.gz \
  -rmode NN

Overview of ggbrain

The ggbrain package uses the ggplot2 package to create 2D volume renderings of NIfTI files containing MRI data. It seeks to follow the principles of a graphical grammar in which plots are built from datasets that are mapped onto the visual space via a set of aesthetic mappings. Likewise, ggbrain follows the approach of creating layers in which different geometric marks (e.g., square pixels) are overlaid on the plot, such that many images and marks can be combined.

The hierarchy of a ggbrain plot

Every ggbrain plot is composed of one or more panels that have layers representing different images that have been overlaid. Each panel contains a slice along one of the three axes of the volume: x (Left-Right), y (Anterior-Posterior), or z (Superior-Inferior). Panels can also contain annotations that reflect one or more marks to aid in the interpretation of the plot, as well as region labels in which text is positioned at certain markers within the displayed brain slice.

  • Plot
    • Panels (aka Slices)
      • Layers: each pixel in the 2D slice contains a value such as a test statistic from a voxelwise analysis. Pixel values can also be categorical, such as levels of a factor (e.g., lobe).
      • Annotations: custom text or graphical marks overlaid on the panel using the ggplot2 annotate function (e.g., coordinate label)
      • Region Labels: a variant of ggplot2 geom_text, region labels represent text that should be overlaid at one or more spatial positions across the panel/slice (e.g., anatomical labels)

Question of how to get entire background to match in color

Use bg argument

ggsave("test.png", gg_obj$render(), width=8, height=6, bg="blue")

pdf("test_dev.pdf", width=8, height=6, bg="blue")
plot(gg_obj)
dev.off()