Title: Synthesizing Computational PIN-4 Immunohistochemistry Stains from Prostate Biopsy H&E Images via Deep Learning

Abstract

Immunohistochemistry (IHC) serves as a critical tool for clarifying diagnostically uncertain findings in prostate cancer biopsies stained with hematoxylin and eosin (H&E). However, the standard protocol for PIN-4 IHC involves processing adjacent tissue sections, which prevents direct spatial correlation between the original H&E morphology and the resulting immunophenotypic signal. To address this limitation, this study constructed a paired, registered dataset of H&E and PIN-4 images derived from routine clinical prostate biopsy whole-slide images (WSIs). A conditional generative adversarial network (cGAN) was subsequently trained to generate PIN-4 staining patterns directly from native H&E image patches.

The final dataset included 172 paired WSIs sourced from 93 patients, comprising 27,298 registered patch pairs of size 1024x1024. These samples covered both adenocarcinoma-positive and benign cases, ensuring diverse representation across various age, racial, and ethnic demographics. Model performance was assessed on a held-out test set consisting of 1,814 patch pairs from 17 WSIs. The system achieved a mean peak signal-to-noise ratio (PSNR) of 21.88 dB, a structural similarity index measure (SSIM) of 0.667, a Pearson correlation coefficient (PCC) of 0.684, and a learned perceptual image patch similarity (LPIPS) of 0.417.

Qualitative evaluation by a board-certified pathologist confirmed that the generated images accurately captured diagnostically significant PIN-4 staining features, such as basal-cell-associated staining and AMACR/racemase expression, while maintaining spatial alignment with the source H&E morphology. While synthesis accuracy fluctuated in morphologically complex areas, such as high-grade and intraductal carcinoma, the findings demonstrate the feasibility of supervised PIN-4 synthesis from standard brightfield H&E prostate biopsy images. This methodology facilitates the direct interpretation of predicted PIN-4 marker patterns within the context of the original prostate H&E architecture, effectively overcoming the spatial constraints inherent in conventional adjacent-section IHC techniques.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC