Manual reading of tissue slides by pathologists serves both as a foundation for clinical decision-making and as a source of ground truth for training artificial intelligence (AI) models. However, challenges such as inter-observer variability, limited tissue availability, and complex annotation tasks often compromise reliability and scalability. This study exemplifies a broader trend in pathology: leveraging virtual staining and other AI-based methodologies to address these challenges. We applied virtual stain multiplexing to a challenging annotation task - macrophage identification in non-small cell lung cancer tissue PD-L1 IHC stains, demonstrating its ability to improve pathologist performance and inter-observer agreement. In six challenging regions selected from 49 curated whole slide images, virtual staining significantly increased macrophage detection consistency, with Fleiss' kappa improving from -0.1 to 0.62, and enhanced overall accuracy, with the F1 score increasing from 0.13 to 0.65. These results highlight the potential use of AI-based virtual staining to assist pathologists reading slides, thereby improving consistency, enhancing accuracy, and alleviating the dependence on additional costly staining. Virtual stain multiplexing demonstrates a generalizable approach to improving pathologist performance through measurement-based AI tools, addressing broader needs for reproducibility and efficiency in diagnostic pathology.
Keywords: Immunohistochemistry, Virtual Stain, Multiplexing, Deep Learning, Macrophages, NSCLC, PD-L1, AI, Agreement, Pathology
TL;DR: Evaluation of Virtual Stain Multiplexing, demonstrating improved agreement and performance of VSM CD68 assisted pathologiststs detecting macrophages in PD-L1 IHC stained slides
Abstract:
Primary Subject Area: Application: Histopathology
Secondary Subject Area: Image Synthesis
Paper Type: Validation or Application
Registration Requirement: Yes
Visa & Travel: Yes
Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Created a single midl25_NNN.zip file with midl25_NNN.tex, midl25_NNN.bib, all necessary figures and files., Includes \documentclass{midl}, \jmlryear{2025}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package, Did not use the times package., All authors and co-authors are correctly listed with proper spelling and avoid Unicode characters., Author and institution details are de-anonymized where needed. All author names, affiliations, and paper title are correctly spelled and capitalized in the biography section., References must use the .bib file. Did not override the bibliographystyle defined in midl.cls. Did not use \begin{thebibliography} directly to insert references., Tables and figures do not overflow margins; avoid using \scalebox; used \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., Appendices and supplementary material are included in the same PDF after references., Main paper does not exceed 9 pages; acknowledgements, references, and appendix start on page 10 or later.
Latex Code: zip
Copyright Form: pdf
Submission Number: 103
Loading