SPUTNIK (version 1.1)

refAndROIimages: Compute the reference image and the ROI mask.

Description

refAndROIimages returns the reference image, calculated using the refMethod, and the ROI binary mask, calculated using roiMethod. These images represent the basic measures for the filters in SPUTNIK.

Usage

refAndROIimages(msiData, refMethod = "sum", roiMethod = "otsu",
  mzQueryRef = numeric(), mzTolerance = numeric(),
  useFullMZRef = TRUE, smoothRef = FALSE, smoothSigma = 2,
  invertRef = FALSE, numClusters = 4, sizeKernel = 5, numCores = 1,
  verbose = TRUE)

Arguments

msiData

msiDataset object..

refMethod

string (default = "sum"). Method used to calculate the reference image. Valid values are:

• "sum": peak intensities sum

• "mean": average peak intensities (without zeros)

• "median": median peak intensities (without zeros)

• "pca": first principal component scores.

roiMethod

string (default = "otsu"). Method used to extract the ROI binary mask. Valid values are:

• "otsu": the reference image is binarized using Otsu's thresholding

• "kmeans": msiData is partitioned in 2 clusters using k-means

• "kmeans2": k-means is applied with a user-defined number of clusters (see Details)

• "supervised": supervised segmentation based on user-defined areas corresponding to off-sample and sample regions.

mzQueryRef

numeric. Values of m/z used to calculate the reference image. 2 values are interpreted as interval, multiple or single values are searched in the m/z vector. It should be left unset when using useFullMZRef = TRUE.

mzTolerance

numeric. Tolerance in PPM to match the mzQueryRef values in the m/z vector. Only valid when useFullMZ = FALSE.

useFullMZRef

logical (default = TRUE). Whether all the peaks should be used to calculate the reference image.

smoothRef

logical (default = FALSE). Whether the reference image should be smoothed before binarizing. Only valid for roiMethod = "otsu".

smoothSigma

numeric (default = 2). Standard deviation of Gaussian kernel.

invertRef

logical (default = FALSE). Whether the reference image colors should be inverted. This can be necessary when the signal is more intense outside the ROI.

numClusters

numeric (default = 4). Only for 'kmeans2' method. Number of clusters.

sizeKernel

4-D numeric array or numeric (default = 5). Only for 'kmeans2'. Each element of the 4-D array represents the size of the corners square kernels used to determine the off-tissue clusters. The element order is clockwise: top-left, top-right, bottom-left, bottom-right. If negative, the corresponding corner is skipped. If only a single value is passed, the same kernel size is used for the 4 corners.

numCores

numeric (default = 1). Only for 'kmeans2' method. Number of CPU cores for parallel k-means. It must be smaller than the number of available cores.

verbose

logical (default = TRUE). Additional output text.

Details

Function to extract the reference image from a msi.dataset-class object. Two references images are returned, a continuous-valued and a binary-valued. Multiple methods can be used to extract both the continuous and the binary reference images, which afterwards can be used as argument for the globalPeaksFilter filter. When 'kmeans2' is applied, the ROI is obtained by merging the sample-related clusters. The user can set a larger number of cluster than 2 (like in 'kmeans'), in such a way a finer segmentation of the sample-related area can be generated. Currently, the off-sample clusters are identified by looking at the most frequent (statistical mode) labels in the corners of the image.

See Also

msiDataset, binOtsu, binKmeans

Examples

# NOT RUN {
## Load package
library("SPUTNIK")

## Mass spectrometry intensity matrix
X <- matrix(rnorm(200), 20, 40)

## Print original dimensions
print(dim(X))

## m/z vector
mzVector <- seq(600, 900, by = (900 - 600) / 39)

## Read the image size
imSize <- c(5, 4)

## Construct the ms.dataset object
msiX <- msiDataset(X, mzVector, imSize[1], imSize[2])

## Calculate the reference and ROI images from the ms.dataset-class object msiX.
## The reference is calculated as the first principal component scores scaled
## in [0, 1]; the binary ROI is calculated applying k-means on the entire dataset.
## Use only m/z values in the range of [700, 900]. The interval extremal values
## are matched within a tolerance of 50 ppm.

ref.roi <- refAndROIimages(msiData = msiX, refMethod = "sum",
                           roiMethod = "otsu", useFullMZRef = TRUE)

## Plot the reference and region of interest ROI
## plot(ref.roi$Reference)
## plot(ref.roi$ROI)
# }