Fig 1.
The workflow of Raman measurements and the Raman spectroscopy setup.
A) HE-stained slides were used for histopathological tumor identification and consecutive deparaffined slides for Raman spectroscopic measurements in the identified tumor regions. B) The Raman spectroscopy setup consists of a 785 nm laser for excitation. Elastic scattered light was removed by a filter that is impermeable for light with a wavelength 785 nm and smaller.
Table 1.
Anonymized patient information.
Fig 2.
Histological properties of ACC and PA samples.
Example for HE-stained A) ACC and B) PA tissue samples, imaged in 20-fold magnification. C) shows a representative image as seen in the Raman spectroscope integrated digital microscope with a 60-fold magnification. Single measuring points are indicated.
Fig 3.
Entity dependent mean spectra: Mean Raman spectra of PA and ACC samples: The spectral intensity of ACC samples is mostly higher compared to PA samples.
Only in the wavenumber area 935–888 cm-1 the intensity measured in PA is higher.
Table 2.
Entity dependent Raman intensities and biochemical assignments of the visually analyzed mean spectra: The Raman spectra detected in ACC are mostly assigned to nucleic acids, lipids, and amides.
Comparably higher intensities in PA were measured for peaks coding for collagens, and saccharides (based on [14, 15]).
Fig 4.
Scores plots of the PCA: The scores of the A) PC-1 + PC-2, B) PC-3 + PC-4, C) PC-5 + PC-6 and D) PC-6 +PC-7 are illustrated.
The best data separation dependent on the tumor entity are seen by the PC-2 and PC-6.
Fig 5.
A) shows the scores plot of the PCA in the spectral range from 600–1700 cm-1. The tumor entities PA and ACC are well-separated by PC-2 and PC-6. B) and C) show the loadings plots of PC-2 and PC-6, respectively. Red arrows mark the peaks that have been chosen for recalculation of PCA.
Table 3.
Raman wavenumber ranges (based on PCA) that are relevant to discriminate the tumor entity and major biochemical assignments (based on [14, 15]).
Table 4.
Confusion matrix of PCA-LDA model with an overall accuracy of 90%.