Supervised 3D Virtual Fluorescence and Pseudo-H&E Staining of Fresh Bulk Tissues via Multimodal qOBM and Fluorescence Microscopy

Abstract

We introduce a multimodal microscope that combines quantitative oblique back-illumination microscopy (qOBM) with wide-field epi-fluorescence deconvolution microscopy to acquire intrinsically co-registered phase and fluorescence volumes of fresh, bulk tissue. Using paired qOBM phase and DAPI fluorescence images of fresh ex vivo mouse brain tissue, we train a supervised conditional generative adversarial network (cGAN) to predict DAPI-like nuclear contrast from the structural information encoded in the phase images. Results not only demonstrate high-fidelity image translation but also show improved imaging depth and more homogeneous staining. Additionally, we combine the qOBM phase contrast with captured or virtual fluorescence nuclear images to produce volumetric pseudo-H&E representations of fresh tissue, enabling a supervised approach for achieving robust virtual histological staining. This work paves the way for virtual fluorescence imaging of many more structural targets and supervised virtual histological staining of thick, unaltered, bulk tissues ex vivo and potentially in vivo. Such capabilities could have significant implications for biological research as well as for point-of-care, intraoperative, and other clinical applications.