Organ transplantation offers life-saving treatment for patients with end-stage organ failure, restoring function and vastly improving quality of life for thousands each year. Yet, transplant rejection remains a leading cause of morbidity in lung and heart recipients, with up to 29% of lung and 25% of heart transplant patients experiencing acute rejection within the first year. The clinical imperative to detect rejection as early and accurately as possible places a heavy demand on pathology workflows, which hinge on laborious histochemical staining of minute biopsy fragments. The conventional chemical staining process of multiple stains not only adds days to diagnostic turnaround-delaying critical treatment decisions-but also incurs high reagent and labor costs. Moreover, chemical staining is susceptible to tissue handling artifacts, uneven dye uptake, and batch-to-batch color variability, all of which can obscure subtle tissue changes associated with transplant rejection and complicate pathologist interpretation.
To address these problems, a research team led by Professor Aydogan Ozcan at the University of California, Los Angeles (UCLA), in collaboration with histopathologists from the University of Southern California (USC) and University of California, Davis, recently published an article in BME Frontiers (AAAS), demonstrating a panel of deep neural networks that virtually generate Hematoxylin & Eosin (H&E), Masson’s Trichrome (MT), and Verhoeff-Van Gieson (EVG) stains for label-free lung tissue, as well as H&E and MT stains for label-free heart tissue. By feeding autofluorescence microscopic images of unstained biopsy sections through these AI models, researchers digitally produce high-fidelity virtual slides, faithfully replicating multiple chemical stains and highlighting transplant rejection features without using any reagents.
“Our virtual staining platform not only delivers diagnostic-quality images but also preserves precious biopsy tissue for subsequent molecular analyses,” said the study’s senior author, Dr. Ozcan. “By eliminating chemical staining procedures, we can save labor, shorten turnaround times, reduce costs, and eliminate the structural mismatches that arise when staining adjacent tissue sections separately.”
In a blinded study involving four board-certified pathologists, the virtual stains achieved concordance rates of 82.4% for lung biopsies and 91.7% for heart biopsies in diagnosing transplant rejection, compared with conventional chemical staining methods. Quantitative assessment of the staining quality of nuclear, cytoplasmic, and extracellular features demonstrated non-inferiority of the virtual slides-and in some cases, virtual H&E outperformed standard stains, especially when histochemical artifacts were present. Beyond staining speed and accuracy, the virtual tissue staining approach also ensures consistent color uniformity across all slides, reducing inter-batch variability-a key advantage for downstream AI-based automatic detection and diagnostic workflows. Moreover, by virtually generating multiple stains from a single unstained tissue section, the framework eradicates the structural mismatches inherent to staining adjacent sections and streamlines pathologist review by obviating the need to align serial images manually.
Overall, this work lays the foundation for scalable, cost-effective digital pathology workflows in transplant medicine and paves the way for downstream AI-driven diagnostic tools that depend on standardized image inputs. Future efforts will extend the platform to additional organ types and disease stages, with the ultimate aim of delivering faster, reliable care to transplant recipients worldwide.
See the article:
Yuzhu Li, Nir Pillar, Tairan Liu, Guangdong Ma, Yuxuan Qi, Kevin de Haan, Yijie Zhang, Xilin Yang, Adrian J. Correa, Guangqian Xiao, Kuang-Yu Jen, Kenneth A. Iczkowski, Yulun Wu, William Dean Wallace, & Aydogan Ozcan (2025). Label-free evaluation of lung and heart transplant biopsies using tissue autofluorescence-based virtual staining. BME Frontiers, (AAAS).