Versatile and rapid virtual H&E staining of unstained, label-free bright-field z-stack microscopy images using deep learning techniques

In the realm of histopathological analysis, hematoxylin and eosin (H&E) staining remains the gold standard for differentiating cellular and tissue structures, underpinning diagnostic workflows in oncology and pathology. However, the conventional protocol for frozen tissue sections, which is characterized by labor-intensive sample preparation, stringent technical requirements, and inherent variability, poses challenges including suboptimal color consistency, potential tissue trauma, and limited scalability for high-throughput applications. To surmount these technical bottlenecks, we present a deep learning-optimized computational framework tailored for label-free virtual H&E staining from grayscale photomicrographs, which are captured through bright-field microscopy configured with white-light illumination optics. Using human glioma tissue specimens, the BFZMN framework is trained on a dataset of 24,901 grayscale-H&E image pairs and tested on 19,268 independent samples, achieving superior performance with a maximum mean SSIM/PCC index of 0.7744/0.7338 at a speed of 9.01 mm 2 /s, which quantifies a faithful reproduction of cellular morphology and chromatic consistency with gold-standard H&E and can support real-time histopathological analysis. To mitigate the multi-focal plane ambiguity arising from coverslip-free tissue preparations, eleven datasets were systematically generated via meticulous modulation of z-axis focal depths to quantify how optical sectioning quality influences the histological fidelity of virtual H&E staining. Findings showed that consistent spectral transmittance within physiological z-ranges enables more reliable histologically faithful reconstructions with preserved nuclear-cytoplasmic contrast across typical tissue thicknesses. Additionally, we introduce a label-free virtual staining platform integrating adaptive sampling and real-time colorization, which operates on low-cost white-light microscopes with cross-platform compatibility, user-friendly operation, and reagent-free histology, and is tailored for resource-constrained settings and intraoperative diagnostics of glioma.