Classification of Raman spectra of single cells with autofluorescence suppression by wavelength modulated excitation

Wavelength modulated Raman spectroscopy has recently been shown to suppress the fluorescence background generated by the sample and the substrate. Here we apply this technique to collect wavelength modulated Raman spectra from 697 individual cells for a model system of circulating tumour cells that consists of leukocytes from patient's blood, acute myeloid leukaemia cells (OCI-AML3), and breast tumour cells BT-20 and MCF-7. We study the classification behaviour of wavelength modulated Raman spectra in comparison to a common background correction method in chemometrics. Classifications using a support vector machine with a radial based kernel function were compared for classical Raman spectra, average Raman spectra of each cell and wavelength modulated Raman spectra. The dataset was divided into 80% training spectra and 20% independent validation spectra. The stability of the classification was tested by performing training and validation 200 times with randomly selected datasets. The results are displayed in box whisker plots. Cell identification based on wavelength modulated Raman spectra gives similar classification rates than classical and averaged Raman spectra with a tendency of reduced accuracies and increased modelling variations. Possible explanations and strategies to further improve the wavelength modulated Raman spectroscopy are discussed.