Deep-learning assisted stimulated Raman histology of human biopsy

Rapid diagnosis of biopsies, including core-needle biopsy and gastroscopic biopsy, is crucial for clinical decision makings. We applied stimulated Raman scattering (SRS) microscopy on fresh biopsy specimens without fixation, sectioning or staining. We further combined SRS with various deep neural networks for fast histological imaging and automated diagnosis. These include the integration of U-Net for femtosecond-SRS histology, as well as the use of convolutional neural network for histological classifications and gradings. Our results indicated that SRS histology integrated with deep learning algorithm provides potentials for delivering rapid diagnosis that could aid the surgical management of cancers.