Hyperspectral unmixing of Raman micro-images for assessment of morphological and chemical parameters in non-dried brain tumor specimens

Abstract

Hyperspectral unmixing is an unsupervised algorithm to calculate a bilinear model of spectral endmembers and abundances of components from Raman images. Thirty-nine Raman images were collected from six glioma brain tumor specimens. The tumor grades ranged from astrocytoma WHO II to glioblastoma multiforme WHO IV. The abundance plots of the cell nuclei were processed by an image segmentation procedure to determine the average nuclei size, the number of nuclei, and the fraction of nuclei area. The latter two morphological parameters correlated with the malignancy. A combination of spectral unmixing and non-negativity constrained linear least squares fitting is introduced to assess chemical parameters. First, endmembers of the most abundant and most dissimilar components were defined that represent all data sets. Second, the content of the obtained components' proteins, nucleic acids, lipids, and lipid to protein ratios were determined in all Raman images. Except for the protein content, all chemical parameters correlated with the malignancy. We conclude that the morphological and chemical information offer new ways to develop Raman-based classification approaches that can complement diagnosis of brain tumors. The role of non-linear Raman modalities to speed-up image acquisition is discussed.