Hyperspectral coherent anti-stokes Raman scattering (CARS) imaging

The optical analysis of microscopic samples has a long history, and many different microscopy techniques evolved through the years. While many of the optical microscopy methods provide destruction-free and noninvasive investigation of the samples, if not only morphological information is desired, its combination with optical spectroscopy techniques provide additional chemical information of the sample. Among the chemical specific optical spectroscopies, the most sensitive technique is based on fluorescence spectroscopy, which takes advantage of chemically specific labeling of the sample with fluorophores. In contrast, vibrational spectroscopies allow for its label-free investigation, among which Raman spectroscopy is one of the most versatile methods for the extraction of intrinsic molecular structure information about the sample. In addition to high chemical specificity, it provides high spatial resolution when performed in a confocal microscope. However, as the spontaneous Raman effect is weak, often long integration times are needed, and its signal may often be overwhelmed by fluorescence in various samples. Coherent anti-Stokes Raman scattering (CARS) circumvents these limitations by adding a second excitation field, coherently exciting the molecular vibrations, which are probed by a third field. The resulting CARS emission is generated on the anti-Stokes side of the excitation wavelengths, avoiding the impact of one-photon induced fluorescence background, when compared to spontaneous Raman detection. Moreover, its coherent amplification provides about three orders of magnitude higher detection sensitivity, and its multi-photon nature provides intrinsic three-dimensional sectioning capability in optical microscopy. The first implementation of CARS in a microscopy setup reported by Xie and coworkers in 1999 [1] has started a new interest in the coherent Raman technique as a powerful noninvasive optical tool for fast label-free chemical imaging based on the intrinsic vibrational contrast of the sample and providing insights of the molecular composition within microscopic samples. Beyond single-resonance CARS detection, an experimental realization of probing the contiguous Raman range over 4000 cm−1 for each image pixel is required. In this thesis, a novel concept of hyperspectral CARS imaging is introduced, which relies on the generation of broadband Stokes pulses in a photonic-crystal fiber (PCF) with a spectral width and energy density which are adequate for efficient CARS excitation. For the successful realization of this concept, first the following questions had to be answered: What PCF-generated supercontinuum (SC) pulse allows hyperspectral CARS generation with both high spectral resolution and high signal-to-noise ratio? Furthermore, what is the best experimental solution for the direct measurement of both amplitude and phase of a broadband field under tightly focused excitation laser beam conditions? As the hyperspectral CARS generation is dependent on the interaction of a single pair of a narrowband pump and a broadband Stokes pulse in the sample, it is important to investigate the generation of the broadband CARS field on a singlepulse level, which cannot be done experimentally. In chapter 5 the hyperspectral CARS generation with ps-seeded SC pulses is therefore simulated to identify the ideal Stokes pulse for a given PCF design. By simulating the phase and amplitude for the broadband PCF seeded SC, and consecutively for the broadband complex CARS field generated with these single SC fields, single-pulse broadband CARS spectra are characterized, which has not been reported in the literature yet. Seed pulse lengths in the fs- and ps-regime are investigated, obtaining non-compressible single SC spectra covering the full Raman shift range needed for subsequent hyperspectral CARS generation. Considering the spectral resolution, with a bandwidth matching that of a typical vibrational resonance, the covered spectral Raman-shift range, the CARS signal-to-noise ratio (SNR), and the simplest experimental implementation, the use of a single laser oscillator, providing 2:94-ps seed-pulses proved to be the best choice. Based on the simulation results, the concept of all-ps hyperspectral CARS imaging with 2:94-ps pulses was developed and an experimental setup was subsequently realized, which are described in detail in chapter 6. Picosecond Hyperspectral CARS imaging is then successfully demonstrated and characterized by investigating two exemplary samples, which both strongly benefit from avoiding the intrinsic fluorescence background. First, samples of primordial broth from a Miller-Urey (MU) experiment are studied in order to gain more insight into their complex and unknown molecular composition and chemical structure in a non-invasive way. Depending on the initial conditions in the MU experiment, molecules with a higher degree of aromaticity, or a higher amount of nitriles and aliphatic C-H compounds, were identified by broadband CARS spectroscopy, which could not be detected by conventional Raman spectroscopy because of the presence of a strong fluorescence background. The assignment of the characteristic peaks to aromatic CxNy ring structures indicates the presence of nucleic acids, which find supporting evidence in the corresponding UVVIS absorption spectra. In the second demonstration of the novel concept, the all-ps hyperspectral CARS imaging is then experimentally applied to the 2D and 3D mapping of chemical and structural properties of molecules inside a single pollen grain. Here, the advantage of the coherent enhancement of the samples intrinsic Raman response in hyperspectral CARS is exploited, performing fast and label-free imaging of a biologically relevant example of a plant cell with unknown composition. The spectral fingerprints of the various biological constituents within their sub-cellular structures are revealed within a sub-micron focus spot. Simultaneous 3D hyperspectral CARS and 2PF volumetric and quantitative imaging of an entire single daisy pollen grain was successfully demonstrated, where in total 950400 spectra were acquired within 127 minutes. The multivariate analysis of the recorded hyperspectral data set has enabled an evaluation without a priori knowledge of the sample. As a result, the exine, the pollen plasma, and both pollen nuclei, were visualized based on their characteristic spectral Raman signatures of phenolic biopolymers, proteins, and nucleic acids, respectively. Hyperspectral CARS imaging is a coherent technique where a coherent enhancement of the CARS field generated in the sample is possible by interference with an external local oscillator (LO) field. In chapter 7, a new approach to interferometric hyperspectral CARS imaging is realized, where broadband phase-correlated signal and idler photon pairs are produced in a PCF by picosecond-seeded broadband spontaneous FWM, which provide the LO field and the Stokes fields, respectively. The temporal delay between the LO field and the CARS field from the sample enables the full control over their phase relation, and hence provides the most general approach to interferometric hyperspectral CARS imaging. This direct experimental access to the phase and amplitude of the CARS field allows the extraction of the complex vibrational response of the samples third order nonlinear susceptibility χ(3)(ν), which is in contrast to the standard hyperspectral CARS imaging approach, where the phase needs to be estimated posterior to the measurements. The proof of this novel concept of interferometric hyperspectral CARS imaging is experimentally demonstrated for a single microscopic droplet of benzaldehyde. After performing hyperspectral CARS imaging exclusively in the frequency domain, in chapter 8 the time domain is added. By exploiting the characteristic of CARS being a Raman pump-probe technique, where the probe pulse is delayed with respect to the pair of temporally overlapped pump and Stokes pulses, the implementation of time-resolved multiplex (2D) CARS enables the measurement of the dephasing times of all spectrally resolved Raman coherences simultaneously. Here, 2D-CARS microspectroscopy with a ps- and a near transform-limited few-cycle fs-pulse has been successfully applied to a neat toluene test sample within a sub-femtoliter probe volume. The Raman free induction decay (RFID) constants of various vibrational modes of toluene were extracted from the analysis of their simultaneously recorded CARS intensity time-profiles. As such, for the first time we were able to spectrally separate the 1000 cm−1 and the 1019 cm−1 modes in time resolved 2D-CARS microspectroscopy. For the application of CARS microscopy in the life sciences, it is often not possible to measure in transmission due to scattering and absorption of the excitation fields inside a thick sample. Therefore it is often indispensable to collect the generated CARS in a epi-detection geometry. In most previous studies on epi-detected CARS microscopy, the back-reflection of forward scattered CARS (F-CARS) from the sample is giving the strongest epi-detected CARS signal contribution. For a complete experimental characterization and better understanding of intrinsic episcattered CARS (E-CARS), the back-reflection of the F-CARS from the sample needs to be separated or better eliminated. In chapter 9, a very thin PMMA polymer sample of continuously varying thickness was introduced for the first systematic and quantitative experimental study of intrinsic broadband E-CARS emission. Pure broadband E-CARS spectra were measured, providing simultaneously both resonant and non-resonant E-CARS from a microscopic sample, which could not be measured by conventional F-CARS microspectroscopy. The fact, that the full broadband ECARS spectrum is measured allows the evaluation of the origin of different E-CARS signal contributions and therefore their comparison with previous simulations. For