Fingerprint stimulated Raman scattering (SRS) produces label-free chemical maps of molecules in living systems with higher specificity compared to CH vibration region. However, due to the weak signal levels in the fingerprint window, it remains challenging for fingerprint SRS to study highly dynamic or large-scale samples. Here, we push the design space of SRS using deep learning, which can recover the signal-to-noise ratio to the levels comparable to measurements with 100 times longer integration time. Combined with an ultrafast 50-kHz delay-line tuner, we can generate real-time images of cholesterol, fatty acid, and proteins of living cells and large-area tissues including the whole brain.
|