The utility of non-linear Raman scattering for the rapid, label-free hyperspectral imaging of biological systems has been demonstrated extensively in the literature over the past decade. Broadband coherent anti-Stokes Raman scattering (BCARS) is particularly suited for cases where the entire, biologically-relevant Raman spectral region is desired. A variant of BCARS, spectral interferometric polarisation CARS (SIPCARS), has been shown previously to provide the advantages of BCARS imaging, without the non-linearly mixed non-resonant background (NRB) that effects BCARS spectra, and with additional information on molecular symmetry encoded into the SIPCARS spectra. This thesis is concerned with three objectives; understanding the limitations of the reported SIPCARS process, exploring schemes to enhance to performance of SIPCARS, and developing tools for the analysis of hyperspectral Raman imaging data.
Limitations of SIPCARS. There are two prominent factors that are deemed essential to investigate to determine if there are fundamental limitations for the application of SIPCARS in certain contexts. The first of these is the noise characteristics of the SIPCARS process. It is shown that, in a shot-noise limited regime, the signal-to-noise ratio (SNR) of SIPCARS spectra will be at least a factor of √2 lower than for BCARS spectra. Moreover, in the current implementation of SIPCARS presented in this thesis, at low exposure times (< 50ms), this factor is even greater. Additionally, the use of SIPCARS in anisotropic systems is reported. SIPCARS uses elliptically polarised excitation fields and thus anisotropic systems can be expected to "scramble" the polarisations leading to the breakdown of the SIPCARS process. Using collagen fibrils to investigate this, it is shown that the SIPCARS scheme does indeed fail to hold for samples with highly-ordered molecules, though it is also demonstrated that the signals generated nonetheless provide information on molecular orientation.
Enhancement of SIPCARS. Two possible schemes are reported; homodyne amplification via an additional χNR signal and an implementation of a hybrid 2-colour/3-colour excitation scheme. Homodyne amplification via an additional χNR signal is shown to be possible, theoretically, by considering additional molecules at the focal volume that scatter with a low χR at the frequency range acquired. An experimental implementation of this scheme provides some confirmation of its ability to increase the signal intensity from a desired sample. The final scheme considered, a hybrid 2-colour/3-colour excitation scheme, follows from previous work in the literature demonstrating that this excitation scheme is a photon-efficient method for generating signal in the entire biologically-relevant spectral range. The theory for SIPCARS with a hybrid 2-colour/3-colour excitation scheme had been developed previously; the work reported in this thesis implements this experimentally and, in particular, reports on the unique dependence on the polarisation of the excitation fields for each component of the excitation scheme.
Hyperspectral analysis. A novel method for the spectral unmixing of hyperspectral Raman data is outlined that incorporates the spatial information present in the data. Borrowing from ideas previously reported in the literature, a spatially constrained clustering algorithm is used to reduce the number of spectra in the data, constrained by the spatial features present in the data, and dictionary learning (a matrix factorisation technique) enables the extraction of unique spectra present in the data, along with the concentrations of each spectra at each spatial pixel in the data. The method outlined is shown to perform well against the commonly used techniques in the field. Reported also is a method for automating the denoising of hyperspectral data (using a low rank approximation method) and a method for performing real-time detection of cosmic ray artefacts during hyperspectral Raman acquisition.
Overall, it is shown that there are several possible schemes that could improve the signal generation characteristics of SIPCARS that can aid in further decreasing acquisition times. However SIPCARS is fundamentally limited in that it cannot be applied to anisotropic systems, though in the context of biological imaging this is rarely encountered. The funda-mentally lower SNR compared to BCARS also illustrates the need for implementing some of the enhancement schemes, to remain competitive with the acquisition speeds of a state of the art BCARS system. For hyperspectral image analysis, a novel method for determining the unique spectra and their concentration is outlined and the broad range of applicability could prove useful for others in the field.
|Date of Award||1 Mar 2021|
|Supervisor||David Richards (Supervisor) & Amelle Zair (Supervisor)|