Fast and label-free 3D virtual H&E histology via active modulation-assisted dynamic full-field OCT

TL;DR

This paper introduces a rapid, label-free 3D virtual H&E histology method using active modulation-assisted D-FFOCT combined with deep learning, enabling real-time intraoperative cancer diagnosis with improved image stability and contrast.

Contribution

The study presents the first integration of active phase modulation with D-FFOCT and deep learning to generate stable, high-contrast virtual H&E images in real-time for intraoperative use.

Findings

Achieved 3D virtual H&E images at 1 fps scanning rate.

Improved image stability and contrast for static tissue visualization.

Demonstrated effectiveness in human CNS and breast cancer diagnosis.

Abstract

Pathological features are the gold standard for tumor diagnosis, guiding treatment and prognosis. However, standard histopathological process is labor-intensive and time-consuming, while frozen sections have lower accuracy. Dynamic full-field optical coherence tomography (D-FFOCT) offers rapid histologic information by measuring the subcellular dynamics of fresh, unprocessed tissues. However, D-FFOCT images suffer from abrupt shifts in hue and brightness, which is confusing for pathologists and diminish their interpretability and reliability. Here, we present active phase modulation-assisted D-FFOCT (APMD-FFOCT) to improve the imaging stability and enhance the contrast of static tissues. This enables us to further employ an unsupervised deep learning to convert APMD-FFOCT images into virtual hematoxylin and eosin (H&E) stained images for the first time. Three-dimensional (3D) virtual H&E-stained images have been obtained at a scanning rate of 1 frame per second, as demonstrated in cancer diagnosis for human central nervous system and breast. The results prove that this new method will play a unique and important role in intraoperative histology.