Enhancing Chemical Contrast in Hyperspectral CARS Microscopy Using a Shallow Learning Approach

Coherent anti-Stokes Raman scattering is a powerful label-free contrast mechanism for chemical imaging that is hindered by a pervasive non-resonant background (NRB) that degrades the underlying spatial-spectral chemical information. Various NRB removal techniques have previously been developed, with a flurry of ever-complex deep-learning models garnering recent attention. Here, we introduce a “shallow learning” approach to NRB removal using gradient-boosted decision trees, enabling improved chemical contrast in hyperspectral CARS microscopy, on par with the best deep-learning models.