MaskImages: Image masking

In jbergenstrahle/STUtility: Analysis and visualization of Spatial Transcriptomics data

Image masking

Description

Masks the background of a tissue section in HE images

Usage

MaskImages(
  object,
  thresholding = FALSE,
  iso.blur = 2,
  channels.use = NULL,
  compactness = 1,
  add.contrast = NULL,
  verbose = FALSE,
  custom.msk.fkn = NULL
)

## S3 method for class 'Staffli'
MaskImages(
  object,
  thresholding = TRUE,
  iso.blur = 2,
  channels.use = NULL,
  compactness = 1,
  add.contrast = NULL,
  verbose = FALSE,
  custom.msk.fkn = NULL
)

## S3 method for class 'Seurat'
MaskImages(
  object,
  thresholding = TRUE,
  iso.blur = 2,
  channels.use = NULL,
  compactness = 1,
  add.contrast = NULL,
  verbose = FALSE,
  custom.msk.fkn = NULL
)

Arguments

object	Seurat or Staffli object o pre-procesing step [default: TRUE]
thresholding	Applies thresholding step
iso.blur	Sets the level of smoothing in the pre-procesing step [default: 2]
channels.use	Select channel to use for masking [default: 1 for '1k' and '2k' platforms and 1:3 for 'Visium' platform]
compactness	Scales the number of super-pixels [default: 1]
add.contrast	Add contrast to pre-procesing step [default: TRUE platforms and 1:3 for FALSE for 'Visium' platform]
verbose	Print messages
custom.msk.fkn	Custom masking function that takes an image of class "cimg" as input and returns a mask of class "pixset" outlining the tissue area.

Details

The masking method provided uses the SLIC method; Simple Linear Iterative Clustering (for more info, see https://www.r-bloggers.com/superpixels-in-imager/). This algorithm generates superpixels by clustering pixels based on their color similarity and proximity in the image plane. Before running the SLIC method on the HE images we have noticed that some pre-processing can improve the masking significantly. First of all, a thresholding is applied to the image using a variant of Otsu's method (see threshold. Thereafter, you can select which color channels to use for the masking. it can be beneficial to remove 1 or two color channels from the image using channels.use. Specifying channels.use = 1 will keep only the first channel before running SLIC. The next step is to apply some blurring efect on the image to "smooth" out speckles in the image. This "smoothing" effect can be adjusted with iso.blur, where a higher iso.blur leads to more smoothing. The compactness will adjust the number of superpixels to compute. If you double this number you will get twice as many superpixels. This can sometimes be helpful to get a finer sensitivity of the masking.

Value

A Seurat or Staffli object with masked HE images

A Staffli object

A Seurat object

Custom mask function

The 'custom.msk.function' gives you the option to use your own masking function specifically designed for your images. The custom function has to take a "cimg" class image as input and return an object of class "pxset".

Troubleshooting

Masking HE images is a non trivial problem as the tissue morphology comes in many different shapes and colors. The default masking algorithm can fail or perform poorly for a number of different reasons and below is a couple of examples of common problems.

• Bubbles and dirt - If you have air bubbles or other no tissue residue in you images, there is a risk that this will be picked up as tissue by the 'MaskImages' function. The function uses the spotfile coordinates to define what region is outside or inside tissue, but if these are not provided, any part of the image picked up by the masking algorithm will be interpreted as relevant. If there is no way around this, you can mask the images manually using an image editing software before loading them into your Seurat object.

• Tissue gets masked instead of background - If this happens, it probably means that your pixel coordinates provided in the spotfiles do not match the HE images provided. For example, if you've run the ST spot detector on a set of HE images, you have to provide these exact HE images when running 'InputFromTable' or otherwise the pixel coordinates will be incorrect.

• Parts of tissue gets masked - The masking algorithm relies on differencies in color intensity to segment out the tissue area. A common problem is that parts of your tissue is more similar to the background than the rest of the tissue, whcih typically happens for cell sparse tissue regions such as adipose tisse or connective tissue. If this happens you can try out different settings for the 'thresholding', 'ìso.blur', 'channels.use', 'compactness' or 'add.contrast' options or you might have to write your own masking function and pass it using the 'custom.msk.fkn' option.

Examples

## Not run: 
# Create a new Staffli object, mask and plot images
st.obj <- CreateStaffliObject(imgs, meta.data)
st.obj <- LoadImages(st.obj, verbose = TRUE) %>% MaskImages()
plot(st.obj)

## End(Not run)

## Not run: 
# Load images into a Seurat objectm, mask and plot images
se <- LoadImages(se, verbose = TRUE) %>% MaskImages()
ImagePlot(se)

## End(Not run)

jbergenstrahle/STUtility documentation built on March 14, 2023, 7:15 a.m.