Digital staining of pathological tissue specimens with PCA-based feature extraction and linear mapping of spectral transmittance

Color is an important parameter in pathological diagnosis and it is the very reason why tissue samples are stained, since the morphological structure of the tissue components can only be vividly observed when they are colored differently. Common for routine staining is the Hematoxylin and Eosin(HE) dyes, however to assess diseases related to the condition of the fibrosis, Masson-trichrome (MT) dyes are used instead. The digital transformation of an HE-stained image to its MT-stained (digital staining) equivalent has already been proposed, where the information derived from the 16-band images of the stained specimens were utilized In this paper we addressed the possible reduction of the spectral dimension requirement to implement the proposed digital staining procedure. To find for the effective spectral dimension, from the classification point of view, principal component analysis (PCA) was applied independently to the five statistical descriptors of the tissue components transmittance spectra, i.e. mean, maximum, minimum, range and standard deviation. In our initial experiments with liver tissue specimens, it was found that ten principal components can be effective to implement the digital staining scheme.