Abstract

Recent years saw continual advances in coherent anti-Stokes Raman scattering (CARS) microscopy as a powerful tool to deliver chemical images of unstained cells and biological tissues at high speed. Narrowband CARS combines two picosecond pump pulses, namely pump and Stokes, whose frequency difference is resonant with a vibrational mode of the scrutinized sample. However, this configuration enables acquiring a single vibrational mode at a time. This calls for broadband approaches, able to merge the fast acquisition of coherent processes with the high chemical specificity of spontaneous Raman. Broadband CARS (BCARS)[1] employs the combination of a narrowband pump beam with a broadband Stokes beam to detect many vibrational modes at once. However, CARS suffers from a chemically unspecific background signal, known as non-resonant background (NRB), that arises from four wave-mixing processes. While, in the narrowband configuration, NRB limits the sensitivity and specificity of the technique and cannot be easily cancelled, in the broadband approach, it can be exploited to amplify the weak Raman modes of the fingerprint region (400-1800 cm^-1). The line shape distortions can be removed in post-processing, exploiting either numerical methods or deep-learning approaches. Despite many improvements in the last decade, BCARS microscopes struggle to work in the fingerprint spectral region, because of a weaker Raman response than the high-frequency CH-stretching region (2800-3100 cm^-1).

© 2023 IEEE