Virtual 3D H&E Staining from Phase-contrast Back-illumination Interference Tomography

2026-05-22 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Advanced, long

Summary

HistoBIT3D introduces the first voxel-wise paired 3D Back-illumination Interference Tomography (BIT) and fluorescence-labeled nuclei dataset, alongside a novel virtual staining framework designed to transform unprocessed tissue volumes into realistic 3D H&E images. This addresses challenges in 3D histopathology, particularly the shift-variant contrast of BIT images and the absence of quantitative validation benchmarks for unsupervised 3D virtual staining. The framework leverages bidirectional multiscale content consistency and cross-domain style reuse to enhance structural fidelity and perceptual realism. Evaluated against ground-truth nuclear distributions, the method achieves leading realism metrics, with an FID of 60.69 and KID of 0.0417. It also significantly improves 3D nuclei segmentation accuracy, yielding a 3D Dice score of 0.594 and a 95th percentile Hausdorff Distance (HD95) of 4.04 µm, with nuclei volumes consistent at 408.3 µm³ versus 405.6 µm³ ground truth. The HistoBIT3D dataset comprises approximately 5,000 images of size 512x512 across various tissue types.

Key takeaway

For AI Scientists or Machine Learning Engineers developing 3D histopathology solutions, you should consider this HistoBIT3D framework to overcome limitations of traditional staining. Its validated approach for generating realistic 3D H&E images from label-free BIT data significantly improves structural fidelity and nuclei segmentation accuracy. This enables faster, non-destructive tissue analysis, crucial for real-time intraoperative assessment and in-vivo imaging applications.

Key insights

A novel dataset and GAN framework enable accurate 3D virtual H&E staining from label-free Back-illumination Interference Tomography.

Principles

• Voxel-wise paired datasets enable quantitative 3D validation.
• Multiscale feature alignment improves structural fidelity.
• Cross-domain style injection enhances perceptual realism.

Method

A CycleGAN-based framework with a ViT generator uses bidirectional multiscale content consistency loss and AdaIN-based cross-domain style fusion for virtual staining.

In practice

• Perform slide-free, volumetric computational histopathology.
• Support real-time intraoperative and in-vivo imaging.
• Improve 3D nuclei segmentation accuracy for pathology.

Topics

• 3D Histopathology
• Virtual Staining
• Back-illumination Interference Tomography
• Generative Adversarial Networks
• HistoBIT3D Dataset
• Image-to-Image Translation

Code references

• aasong113/HistoBIT3D

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Generative Deep Learning for Computational Destaining and Restaining of Unregistered Digital Pathology Images — Preprocessing-based domain adaptation enables cGAN models to generalize for computational pathology across institutions without retraining.
• Atlas H&E-TME: Scalable AI-Based Tissue Profiling at Expert Pathologist-Level Accuracy — Atlas H&E-TME provides scalable, expert-level AI-based tissue profiling from H&E slides, validated by a robust dual framework.
• Benchmarks and methods for 3D medical image retrieval — The 3D-MIR benchmark and methods advance medical image retrieval by providing standardized evaluation and multi-modal search strategies.
• Three-Dimensional Retinal Microvasculature Restoration in OCT Angiography — A deep learning algorithm accurately restores 3D retinal microvasculature from single OCTA volumes, overcoming imaging artifacts.
• Transformer-Guided Graph Attention for Direct Cardiac Mesh Reconstruction: A Structural Digital Twin Framework — The framework directly generates simulation-ready cardiac meshes from medical images, bypassing traditional multi-stage, manual post-processing.
• Virtual Scanning for NSCLC Histology: Investigating the Discriminatory Power of Synthetic PET — Synthetic PET data can significantly enhance NSCLC histological classification when combined with CT scans.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.