Raman Spectroscopy for Early Cancer Detection, Diagnosis and Elucidation of Disease-Specific Biochemical Changes - Emerging Raman Applications and Techniques in Biomedical and Pharmaceutical Fields - page 325

Fig. 13.8. Left shows an example of a core biopsy specimen (H&E stained) with
high grade DCIS (marked by a consultant histopathologist) and right a white light
image of the selected duct as imaged on Raman microspectrometer
Wave number calibration, cosmic ray removal and system energy sensitiv-
ity correction were performed following spectral acquisition. Each spectrum
acquired in the Raman map can be represented by a pixel in an image us-
ing principal components to construct pseudocolour score images. Figure 13.9
shows an example duct PC score image and associated mean spectra calcu-
lated from the regions of interest selected on the score image.
Further analysis was achieved by utilising the spectra selected from the
specific regions of each duct. These spectra were averaged according to their
pathological classification to enable visualisation of the mean spectral differ-
ences (Fig. 13.10). The individual spectra were then entered into a principal
component-fed linear discriminant classification model (as described earlier).
This used a total of 5,661 spectra, which were acquired from 22 selected breast
ducts covering the four DCIS groups as defined by the Holland classification
outlined above. There were four low grade, four intermediate grade, seven
high grade non-comedo and seven high grade comedo necrosis.
The classification model was cross-validated using leave one spectrum out
to test the robustness of the model. By doing so the individual spectrum is
Fig. 13.9. Pseudocolour map and selection of regions of the duct and their corre-
sponding spectra