In a groundbreaking study published in BME FrontierResearchers at the University of California, Los Angeles (UCLA), in collaboration with international partners, have developed a deep learning-based virtual multiplexed immunostaining (mIHC) method. This new approach enables the simultaneous generation of ERG, PanCK, and H&E images from label-free tissue sections, greatly improving the accuracy and efficiency of vascular invasion assessment in thyroid cancer.
Traditional immunohistochemistry (IHC) techniques, which are crucial in the diagnosis of various cancers, require separate tissue sections for each staining, leading to increased cost, labor, and potential tissue loss. Additionally, these methods have section-to-section variability, which compromises diagnostic accuracy. Multiplex IHC (mIHC) technology allows simultaneous staining with multiple antibodies, but is complex and not widely available in routine pathology laboratories.
A research team led by UCLA’s Aydogan Ozcan and Nir Pillar addressed these challenges by introducing a virtual mIHC framework that leverages deep learning algorithms. This technique utilizes autofluorescent microscopy images of unstained tissue sections to generate virtual stains that closely match their histochemically stained counterparts. Virtual staining includes ERG for endothelial cells, PanCK for epithelial cells, and H&E for common tissue morphology.
The virtual mIHC method was trained and validated using a dataset containing paired autofluorescence and histochemical staining images from a thyroid tissue microarray. By employing conditional generative adversarial networks (cGAN) and digital staining matrices, this framework successfully transformed label-free images into virtually stained images and achieved high agreement with traditional staining methods.
Blinded evaluation by a board-certified pathologist confirmed the validity of virtual mIHC staining, with strong agreement in staining pattern, intensity, and cellular localization. Virtual staining accurately highlighted epithelial and endothelial cells and facilitated the identification and localization of vascular invasion, a critical step in cancer metastasis.
Virtual mIHC technology represents a significant advance in histopathological evaluation and provides a cost-effective, efficient, and accurate alternative to traditional IHC and mIHC methods. This innovation has the potential to transform clinical practice and improve patient outcomes in thyroid cancer and other diseases by streamlining diagnostic workflows and preserving valuable tissue samples. The research team’s future work will focus on further validating the technology across diverse tissue types and multi-site cohorts, paving the way for broader clinical adoption.
sauce:
Reference magazines:
Zhang Yu others. (2026). Virtual multiplex immunostaining of label-free tissues using deep learning. BME Frontier. DOI: 10.34133/bmef.0226. https://spj.science.org/doi/10.34133/bmef.0226
