Broadband CARS high-throughput single-cell imaging

. Broadband Coherent anti-Stokes Raman Scattering (BCARS) enables the whole vibrational spectrum of cytologically prepared samples to be obtained using a hyperspectral raster scan approach. This technique has the potential to enable high-throughput automated detection of cell abnormalities. Images are distorted by the non-resonant background which requires a treatment for proper analysis. Using statistical denoising and phase retrieval returns Raman spectra similar to that of a spontaneous Raman measurement. Here, we present our work using this method for single-cell imaging of PEO1 ovarian adenocarcinoma cells prepared with the ThinPrep processor which enables label-free Raman cytology.


Introduction
The Raman spectra of biological samples are rich with molecular information that can be used for identifying biochemical constituents.Conventional vibrational microscopy, however, lacks the sensitivity to allow cell imaging to be practical.Raman microspectroscopy is capable of performing cytological disease classification of bladder cancer [1], but the acquisition times are on the order of tens of seconds.Much stronger vibrational signals can be attained with broadband coherent anti-Stokes Raman scattering (BCARS).
BCARS is an established bioimaging technique, extensively applied to tissues and lipids.The application of BCARS to single cell samples has previously been limited by signal to noise constraints which are due to the nonresonant background which dominates the total measured signal.The results of bioimaging using BCARS have thus only made use of the high-wavenumber region, including the CH/OH-stretch vibrations, which have very high Raman intensities.However, most of the Raman information in biological samples is contained in the low-wavenumber fingerprint region where the NRB completely obscures the vibrational information.Here we present work on singlecell imaging using a combined denoising and NRB removal approach.

Current state of BCARS imaging
Current approaches to BCARS cell imaging attempts to avoid phase retrieval of the susceptibility using spectral unmixing on the raw BCARS signal [2].This produces concentration maps of spectrally distinct components based on endmember extraction such as multivariate curve resolution.Endmembers extracted through this process may not resemble the underlying Raman signal, and thus, the attribution of endmembers to cellular features must be based on assumptions about their morphology and spatial extent.Our approach uses Kramers-Kronig phase retrieval after denoising the data using the truncated singular value decomposition (SVD), as shown in [3].We apply this method to single cell imaging and show the result for a retrieved and denoised single pixel, thus demonstrating that BCARS hyperspectral images of single cells can be obtained using a combination of denoising and phase retrieval to obtain the useful Raman signal.

Spectral processing chain
The spectra were denoised using the truncated SVD method and the first 25 singular values (SVs).We then use the phase corrected Kramers-Kronig method [4] to retrieve the resonant susceptibility.This results in Ramanlike spectra, however the variance between pixels of the same cell is high, due to the use of a single NRB reference measurement.To account for this variance, we detrend all data-points using the extended multiplicative signal correction (EMSC) method.This results in a Ramanlike dataset, for which conventional spectral analysis can be performed.Finally we implement spectral unmixing using the least-squares method.This results in independent spectral signatures and spatial concentration maps of each endmember.

Results
In Figure 1 we show the retrieved and denoised BCARS spectrum of a single pixel from a hyperspectral image of PEO1 ovarian adenocarcinoma cells.The spectrum of the retrieved cell shows distinctive features such as the phenylalanine band at 1004 cm −1 and a lipid band at 1440 cm −1 .We note there is a lower intensity in the region > 1500 cm −1 , which is due to the laser spectrum having a low excitation power in this region.Also shown in Figure 1 is the retrieved spectrum of the glass slide, which returns a non-resonant noise signal.In Figure 2 we show the Raman map of the retrieved BCARS image, colored by the intensity of three different bands.A distinct spatial distribution of the three bands is present within each cell.In future work, we plan to extend our procedure to include the supervised classification of cells based on the BCARS spectrum, which would remove the barrier for Raman mapping for cytology.Figures 3 and 4 show four seperate endmember spectra extracted using the N-FINDR approach [5] and the concentration map associated with each endmember.

Figure 1 .
Figure 1.Retrieved and denoised BCARS spectrum of a PEO1 cell at the position shown in Figure 2.

Figure 2 .
Figure 2. Hyperspectral pseudo-color image of PEO1 ovarian adenocarcinoma cells prepared using the ThinPrep cytology standard.Colors based on three intensity bands.Red = 2992 cm −1 , Green = 1115 cm −1 , Blue = 1004 cm −1 (white X represents location of single pixel shown in Figure 1.)

Figure 4 .
Figure 4. Montage of four abundance maps created using spectral unmixing by the least-squares method.