WCLC 2025 - Posters & ePosters-P2.06.29 Bridging Histopathology and Molecular Profiling: AI-Powered Virtual Multiplex Immunofluorescence Staining for LUAD Prognosis

P2.06.29 Bridging Histopathology and Molecular Profiling: AI-Powered Virtual Multiplex Immunofluorescence Staining for LUAD Prognosis

Back to course

Pdf Summary

This study presents HEMIFuse, an AI-driven platform that generates virtual multiplex immunofluorescence (mIF) images from standard hematoxylin and eosin (H&E) stained tissue sections to aid prognosis in lung adenocarcinoma (LUAD). Multiplex immunofluorescence allows simultaneous visualization of multiple biomarkers but suffers from high cost, time consumption, and limited histopathological context.<br /><br />Using the pix2pixHD generative adversarial network, HEMIFuse was trained on 133 stage-I LUAD patient samples, creating paired datasets from the same formalin-fixed paraffin-embedded tumor sections with co-registered H&E and mIF images at the cell level. This setup enabled the AI to learn to synthesize high-quality virtual mIF images purely from conventional H&E images.<br /><br />Comparative evaluation showed that virtual mIF images closely matched chemically stained ground-truth images in both visual quality and fluorescence signal accuracy. Importantly, biomarker expression levels extracted from virtual mIF images allowed effective prognostic stratification of LUAD patients, with statistical significance (p < 0.05), performing comparably to real mIF data.<br /><br />HEMIFuse offers a rapid, cost-effective alternative to traditional multiplex staining by bridging routine histopathology with multiplex molecular profiling, potentially streamlining workflows in clinical pathology. The study acknowledges the need for prospective, multicenter validation to confirm the robustness and generalizability of the virtual staining model.<br /><br />Supported by multiple Chinese scientific foundations, this work demonstrates a promising step toward integrating AI-assisted virtual staining to enhance precision oncology, enabling comprehensive tumor profiling without added experimental burden. Future research is planned to expand validation and refine the methodology for broader clinical application.

Asset Subtitle

Yu Lei

Meta Tag

Speaker Yu Lei

Topic Pathology and Biomarkers

Keywords

HEMIFuse

virtual multiplex immunofluorescence

multiplex immunofluorescence

hematoxylin and eosin staining

lung adenocarcinoma

pix2pixHD generative adversarial network

AI-driven image synthesis

prognostic stratification

digital pathology

precision oncology