Optimal control enhances contrast in coherent anti-Stokes Raman scattering images

Coherent anti-Stokes Raman scattering (CARS) has become an established microscopy tool in biophysics, biomedicine, and chemistry for rapid imaging of living cells.¹ Unlike fluorescence, white-light, and phase-contrast microscopies, for example, CARS provides chemical contrast by looking at the intrinsic vibrational structure of analytes (molecules of interest). Consequently, it works on unstained, unlabeled—i.e., native—samples with imaging speeds of up to video rates.

A particularly desirable area of application is visualizing drugs or metabolites (byproducts of drugs) in complex samples such as mammalian and plant tissue. But in this context CARS microscopy remains challenging because the technique is insufficiently sensitive to detect low concentrations of target materials as with drugs.² Solving this problem is key to more widespread use of CARS, which is currently largely restricted to detecting lipid (fat) content in biological samples due to the high density of carbon-hydrogen groups contained in these molecules.³

Here, we report our recent work on combining spatial-light modulators for controlling incident light fields in microfluidic chips. The latter serve as the basis for performing microspectroscopic experiments on controlled and reproducible samples. Consequently, the focus of this study is on developing and validating methods. Our experiments aim at establishing the detection limits of model analytes in water-based solutions using CARS image contrast.⁴ Unlike previous efforts, the experiments reported here use an image-contrast parameter for direct feedback in a self-learning algorithm.

CARS-imaging systems are typically combined with a micro-channel flow system to quantify the relative contributions of resonant and nonresonant scattering at the CARS frequency. The resonant contribution originates from the analyte to be investigated, i.e., it is the part of the signal that accounts for the chemical sensitivity of the method. The nonresonant signal stems from the presence of any molecules and reduces the chemical image contrast. The two-channel microfluidic chip design, which uses deuterated isotopomers as internal standards (i.e., as a baseline for the image), allows fast and quantitative detection of analytes by CARS microscopy. Our experimental design departs from conventional CARS imaging by simultaneously measuring the chemically relevant Raman-resonant signal and the non-Raman-resonant background, and forms the basis for incorporating pulse-shaper-assisted optimal control of image contrast.

In a proof-of-concept experiment we used a two-channel microfluidic chip for quantitative CARS detection. We filled one of the channels of the chip with our model compound, deuterated toluene (C₇D₈). The other channel contained toluene (C₇H₈) as the reference sample. The droplet-based microfluidic device allows straightforward concentration-dependent measurements of binary C₇D₈/C₇H₈ mixtures to determine the detection limit of C₇D₈ (see Figure 1). When C₇D₈ is diluted in C₇H₈, the contribution of the C-D vibration to the overall CARS signal (which we selected to be in resonance with the lasers used in the experiment) decreases. When the detection limit is reached, the vibrationally resonant C-D signal becomes indistinguishable from the intrinsic background.

Figure 1. Coherent anti-Stokes Raman scattering (CARS) intensity profiles for a two-channel chip. High intensities indicate higher concentration of deuterated toluene. C₇D₈: Deuterated toluene. C₇H₈: Toluene.

The results of these CARS microscopic concentration-series experiments showed that a detection limit for deuterated toluene dissolved in toluene is reached at x (C₇D₈)=0.05, which translates into a molar concentration of about 470mM. This value relates to the ratio of vibrationally resonant CARS signal to vibrationally nonresonant background in a homodyne detection scheme.⁵

Finally, we combined strategies of optimal control, which are well known from spectroscopic experiments, with CARS microscopy. The setup uses an evolutionary algorithmic strategy and closed-loop feedback together with a liquid-crystal spatial modulator to control the spectrum of the Stokes pulse within a CARS scheme to improve the vibrational contrast of the images. We optimized the CARS excitation spectrum for image contrast at a predetermined wavenumber position. The feedback uses an experimentally imposed fitness parameter generated from the image itself. This strategy (which was used for the images shown in Figure 1) enhances image contrast by a factor of up to 2.6.

In summary, furthering the use of CARS imaging as an analytical tool requires developing methods of quantifying and enhancing image contrast. Thus far, our results in combining microfluidics with strategies of optimal control have shown great promise. Our next steps will focus on detecting analytes in more complex media and on optimizing images using more realistic sample geometries.

This research was supported by the European Regional Development Fund (project 578-06001). G.B. and S.S. acknowledge financial support from the Sonderforschungsbereich 630, and S.S. is grateful for a Heisenberg fellowship. B.D. and J.P. thank the Fonds der Chemischen Industrie for financial support. We would also like to thank the European Commission Sixth Framework Programme funded project ACCORD (034041) for supplying us with the acousto-optic programmable dispersive filter (DAZZLER) from Fastlite through the experimental component exchange program.