In a groundbreaking study published in BME Frontiersresearchers from the University of California, Los Angeles (UCLA), in collaboration with international partners, have developed a virtual multiplex immunostaining (mIHC) method based on deep learning. This novel approach enables the simultaneous generation of ERG, PanCK and H&E images from label-free tissue sections, greatly enhancing the accuracy and efficiency of the assessment of vascular infiltration in thyroid cancer.
Traditional immunohistochemistry (IHC) techniques, which are instrumental in the diagnosis of various cancers, require separate tissue sections for each stain, leading to increased cost, labor and potential tissue loss. Additionally, these methods exhibit slice-to-slice variability, compromising diagnostic accuracy. Multiplex IHC (mIHC) technologies, although capable of simultaneous staining with multiple antibodies, are complex and not widely available in routine pathology laboratories.
The research team, led by Aydogan Ozcan and Nir Pillar from UCLA, addressed these challenges by introducing a virtual mIHC framework that leverages deep learning algorithms. This technique uses autofluorescence microscopy images of unstained tissue sections to generate virtual stains that closely match their histochemically stained counterparts. Virtual stains include ERG for endothelial cells, PanCK for epithelial cells, and H&E for general tissue morphology.
The virtual mIHC method was trained and validated using a dataset including pairs of autofluorescence and histochemically stained images from thyroid tissue microarrays. Using conditional generative adversarial networks (cGANs) and a digital coloring matrix, the framework successfully converted unlabeled images into virtual colored ones, achieving high agreement with traditional coloring methods.
Blinded evaluations by board-certified pathologists confirmed the efficacy of mock mIHC staining, with strong concordance in staining patterns, intensity, and cellular localization. Virtual stains accurately highlighted epithelial and endothelial cells, making it easier to identify and localize vascular invasion—a critical step in cancer metastasis.
The virtual mIHC technique represents a significant advance in histopathological evaluation, offering a cost-effective, efficient and accurate alternative to conventional IHC and mIHC methods. By streamlining the diagnostic workflow and preserving valuable tissue samples, this innovation has the potential to transform clinical practice, improving patient outcomes in thyroid cancer and beyond. The research team’s future work will focus on further validating the technology in various tissue types and multi-site cohorts, paving the way for wider clinical adoption.
