Novel approaches to broadband coherent Raman imaging

We present innovative approaches to broadband coherent Raman scattering microscopy, both in the stimulated Raman scattering (SRS) and broadband coherent anti-Stokes Raman scattering (CARS) modalities. A convolutional neural network removes the unwanted non-resonant background from broadband CARS spectra. The deep-learning model processes experimental data in 100microseconds and correctly retrieves all the relevant vibrational peaks without any user intervention or independent background measurement. A multi-channel lock-in amplifier, in combination with in-line balanced detection and a broadband optical parametric oscillator, allows sensitive measurement of the SRS spectrum at 32 frequencies in parallel, with pixel dwell times as short as 40 microseconds.