Enhancing Translation of H&E to IHC with Robust Tumor-Infiltrating Lymphocytes Quantification Using Deep Generative...

Area of Science:

• Computational pathology
• Digital pathology
• Artificial intelligence in oncology

Background:

• Breast cancer treatment response prediction is crucial, with tumor-infiltrating lymphocytes (TILs) as key prognostic biomarkers.
• Traditional visual TILs quantification from H&E slides is subjective and prone to interobserver variability.
• Virtual H&E-to-IHC stain translation offers a promising, less labor-intensive alternative to conventional IHC.

Purpose of the Study:

• To develop a deep generative framework for synthesizing virtual IHC patches from H&E images.
• To create an automated pipeline for accurate TILs quantification using virtual IHC.
• To evaluate the performance and accuracy of the developed framework and pipeline.

Main Methods:

• A deep generative framework with spatial attention and chromogen loss was developed for H&E-to-virtual IHC synthesis.
• H&E slides were destained and restained for IHC, followed by image registration for alignment.
• An automated TILs quantification pipeline was designed, accounting for cells in overlapping regions.

Main Results:

• The framework significantly improved virtual IHC synthesis, reducing mean absolute error (MAE) for DAB and AP chromogens.
• The automated TILs quantification pipeline demonstrated moderate to high agreement with pathologists' counts (ICC: 0.565-0.852).
• Synthesized chromogens were visually comparable to ground truth IHC, and relative density ranking exceeded inter-reader consistency.

Conclusions:

• The developed deep generative framework and automated pipeline offer a feasible and accurate method for TILs quantification.
• Virtual H&E-to-IHC translation can overcome limitations of manual quantification, enhancing prognostic biomarker assessment.
• This approach holds potential for improving breast cancer treatment response prediction and patient outcomes.