Initial centroids are determined by comparing data points from the quantified
spots and mapped pixels from the same coordinates.
First, mapping intensities counted from multi-phase pixels are removed from
the dataset (See "Multi-phase pixels" section).
Second, if the removal reduces sample size of some phases to less than 10,
then the mapping intensities are restored from signal intensities from spot
Third, for each phase,
median values of signal intensities are utilized
as initial centroids.
Forth, if there are missing values especially for intensities of electrons
(secondary, backscatter, ...), imputation is performed for them
(See "Impute missings" section).
find_centers( xmap, qnt, phase = everything(), element = everything(), saveas = "centers0.csv", epma = tidy_epma(qnt, xmap), fine_phase = NULL, ... )
An object generated by
An object generated by
File name to save result.
A value returned by
Deprecated as of qntmap > 0.4.0. Use
Arguments passed on to
X-ray mapping inevitably involves multi-phase pixels, especially for
fine-grained phases. When the multi-phase pixels are member of training
data set, initial centroids can be biased. This is why guessing
centroids is based on
median() instead of
mean(). However, if pixels
to guess centroids for cetrtain phase are comprising a large number of or
even a full of multi-phase pixels,
median() is still vulnerable. For
example, refer to a reprinted figure 5 from
Yasumoto et al. (2018), and
captions in original article. Phases such as Amp, Di, and Pl are outlying
the regression curves due to multi-phase pixels. For these phases,
median values of peak X-ray intensities from X-ray mapping are unreliable
values for initial centroids. Thus, data points from multi-phase pixels
should be regarded as outliers, and imputetions should be performed on them.
In qntmap, mapping intensities are simply substituted by spot intensities.
Outliers are identified by examining if confidence intervals data points overlap with predictive intervals or conditional whiskers of Tukey's choice.
When spot analysis on some phases are performed outside the mapping area, X-ray intensities under mapping conditions can be calculated from quantified peak intensities. However, intensities of secondary and backscatter electrons cannot be calculated because electrons are generally not analyzed in spot analysis. Thus, imputation is performed. First, calculate median values of mapping X-ray intensities. Second, find a pixel in the map which have closest value to the median values. Third, retrive electron intensities from that pixel and utilize as a centroid.
Note that missing values may remain if certain elements are analyzed in spot analysis but not in mapping analysis.
Yasumoto, A., Yoshida, K., Kuwatani, T., Nakamura, D., Svojtka, M., & Hirajima, T. (2018). A rapid and precise quantitative electron probe chemical mapping technique and its application to an ultrahigh-pressure eclogite from the Moldanubian Zone of the Bohemian Massif (Nové Dvory, Czech Republic). American Mineralogist, 103(10), 1690-1698, https://doi.org/10.2138/am-2018-6323CCBY.