JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Clinical-Grade Validation of an Autofluorescence Virtual Staining System With Human Experts and a Deep Learning System for Prostate Cancer

  • 0Verily Life Sciences LLC, San Francisco, California.
Modern Pathology : an Official Journal of the United States and Canadian Academy of Pathology, Inc +

|

July 28, 2024

|

July 28, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

Researchers developed an AI-powered virtual staining system for prostate cancer diagnosis. This automated technology creates high-quality digital images from unstained tissue, aiding genitourinary pathologists in accurate Gleason grading.

Area Of Science

  • Digital Pathology
  • Artificial Intelligence in Medicine
  • Computational Pathology

Background

  • Prostate cancer diagnosis relies on histopathology, including Gleason grading of hematoxylin and eosin (H&E) stains and immunohistochemistry (IHC) markers.
  • Traditional staining methods are time-consuming and require specialized reagents and expertise.
  • Virtual staining offers a potential solution to digitize and standardize histopathological analysis.

Purpose Of The Study

  • To develop and validate an automated system for generating virtual H&E and prostatic intraepithelial neoplasia-4 (PIN4) IHC stains from unstained prostate tissue.
  • To assess the diagnostic quality and clinical utility of the generated virtual stains for prostate cancer detection and grading.
  • To extend previous work on virtual staining using autofluorescence and establish rigorous evaluation standards for digital pathology.

Main Methods

  • Utilized a high-throughput hyperspectral fluorescence microscope to capture data from unstained prostate tissue.
  • Applied artificial intelligence (AI) and machine learning (ML) algorithms to generate virtual H&E and PIN4 (CK5/6, P63, AMACR) stains.
  • Validated the virtual stains through extensive human review by genitourinary pathologists, computational analysis, and comparison with a validated Gleason scoring model on a large dataset.

Main Results

  • The AI-powered virtual stainer successfully produced high-quality virtual H&E and PIN4 stains from unstained prostate tissue.
  • The generated virtual images were deemed suitable for diagnostic purposes by expert genitourinary pathologists.
  • Extensive validation confirmed the system's reliability and accuracy, demonstrating its clinical utility for prostate cancer pathology.

Conclusions

  • The developed automated virtual staining system demonstrates significant potential for revolutionizing prostate cancer diagnosis.
  • This technology offers a high-throughput, AI-driven approach to digital pathology, enhancing diagnostic efficiency and accuracy.
  • The study sets a benchmark for evaluating digital pathology tools, highlighting the clinical applicability of virtual staining in cancer diagnostics.

Keywords: