Recently, researchers at Sichuan University and collaborating institutions have developed a label-free virtual staining method that reveals richer histological information in thick tissue sections, which is published in PhotoniX Life . This study combines high-throughput multispectral priors with an optics-informed deep-learning framework to generate high-fidelity virtual H&E images of unstained 10μm-thick sections.
Histochemical staining is central to pathological examination, allowing clinicians to assess tissue architecture and cellular abnormalities. Conventional sections are typically around 4μm thick, partly because the limited depth-of-field of optical microscopy makes thicker specimens difficult to visualize clearly. However, thin sections may truncate cellular structures and leave depth-dependent pathological features incompletely represented. Thicker sections retain more information, but structures at different depths cannot all remain sharply focused.
The new approach addresses this challenge by exploiting two complementary sources of information within the spectrum. Spectral transmittance provides an optical fingerprint of tissue, with relationships across wavelengths offering distinguish different tissue components and hereby guide virtual stained representation generation. Meanwhile, optical dispersion links wavelength to axial focal position, allowing different spectral channels to capture sharply focused structures at different depths. Together, these spectral and spatial cues provide the physical basis for recovering detailed staining representations across a thicker section.
The team built a high-throughput microscopy system to acquire multispectral transmittance and phase-contrast images of unstained specimen. These complementary optical features are integrated within a modular staining network that combines computational refocusing with physically informed image generation. Supervised by histochemical stained reference images fused from multiple focal depths, the framework generates virtual H&E images of 10μm-thick specimen while preserving fine cellular detail.
Results showed that the 10μm section revealed more continuous cellular structures and richer depth-dependent histological features than corresponding 4μm section prepared from the same human liver tissue block. Furthermore, two experienced pathologists independently conducted blinded evaluations of virtual and histochemical stained images, assessing overall image quality and staining details. The high pathological evaluation scores and strong concordance between virtual stains and corresponding histochemical stains for both 10μm and 4μm sections demonstrated the staining fidelity and morphological preservation capability of proposed approach, while the consistently higher scores of 10μm sections across the evaluated criteria suggest the potential of thick-section virtual staining to provide enhanced structural representation.
“Moving forward, we plan to strengthen collaborations between engineering and medicine, exploring applications such as live-cell and whole-tissue virtual staining,” said Dr. Junfei Shen, corresponding author and associate professor at Sichuan University. “Ultimately, virtual staining can only realize its full potential by addressing the real clinical needs of pathologists.”
See the article:
Multispectral dispersive-refocused imaging enables label-free virtual staining of 10μm thick-tissue sections
https://doi.org/10.3724/PXLIFE.2026-0017
Experimental study
Human tissue samples
31-Oct-2026
The author(s) declare that they have no conflict of interest.