Virtual Staining for Nonlinear Imaging of Cancers.

Abstract

Nonlinear imaging such as third-harmonic generation (THG) microscopy can produce label-free images of tissue with histopathological quality, yet the interpretation of label-free images is challenging. Virtual staining can convert label-free images into histology-like images, but its application to nonlinear imaging remains limited even though it has been well established for fluorescence and bright-field imaging. In this review, we summarize the optical principles of several nonlinear imaging modalities, including multiphoton excitation fluorescence, stimulated Raman scattering, and THG microscopy. We then review the recent progression in virtual staining, major network architectures and metrics, and the main challenges for translation into the clinic. In particular, we emphasize the challenges of virtual staining in THG imaging. Unlike molecularly specific contrast mechanisms, THG contrast arises from spatial variations in refractive index and/or third-order nonlinear susceptibility. Consequently, the collected THG signals may reflect multiple structural sources, including water-lipid interfaces and chromatin-associated optical heterogeneity within the nucleus, making it complex and difficult to establish a direct correspondence between THG features and histological labels. By linking imaging to computational methods, this review highlights key directions for advancing THG-oriented virtual staining towards clinical translation.