High-Speed Virtual Re-Staining via Near-Infrared Quantitative Phase Imaging

Abstract

To rapidly and cost-effectively re-stain faded histological specimens, we present a high-speed virtual re-staining pipeline that combines near-infrared quantitative phase imaging utilizing Fourier ptychography and a trained deep neural network. Our supervised virtual re-staining model comprises two stages: a raw-to-phase module that replaces iterative FP reconstruction, and a phase-to-color module that synthesizes histologic RGB images for virtual staining. Because hematoxylin and eosin absorption is weak in the near infrared (NIR), the NIR quantitative phase image serves as a stable intermediate representation that is largely invariant to stain fading, which enables registration-free dataset generation and supervised training. Our prototype virtually re-stains faded stomach slide of a 3.9 x 2.6mm(2) field of view in approximately 1 minute.