Related Experiment Video
Updated: Jul 17, 2025
08:13
A Guide to Build a Highly Inclined Swept Tile Microscope for Extended Field-of-view Single-molecule Imaging
Published on: April 8, 2019
17.5K
Synthetic whole-slide image tile generation with gene expression profile-infused deep generative models
Francisco Carrillo-Perez1,2, Marija Pizurica1,3, Michael G Ozawa4
1Stanford Center for Biomedical Informatics Research (BMIR), Stanford University, School of Medicine, 1265 Welch Road, Stanford, CA 94305-547, USA.
Cell Reports Methods
|September 6, 2023
Summary
Researchers developed RNA-GAN, a novel deep learning model that generates realistic whole-slide image tiles using gene expression data. This approach enhances tissue image generation and shows potential for data imputation in computational pathology.
Area of Science:
- Computational pathology
- Bioinformatics
- Deep learning
Background:
- Whole-slide images (WSIs) are crucial for digital pathology.
- Generating realistic WSIs with associated gene expression data is challenging.
- Deep generative models offer potential for synthetic data creation.
Purpose of the Study:
- To develop a novel deep generative model for synthesizing whole-slide image tiles.
- To integrate gene expression profiles into the image generation process.
- To improve the quality and efficiency of synthetic tissue image generation.
Main Methods:
- Training a variational autoencoder (VAE) to learn latent representations of gene expression profiles.
- Infusing generative adversarial networks (GANs) with VAE representations to create RNA-GAN.
- Generating lung and brain cortex tissue tiles using the RNA-GAN model.
Main Results:
- RNA-GAN generated synthetic tissue tiles that were preferred by expert pathologists over traditional GANs.
- RNA-GAN required fewer training epochs to produce high-quality tiles.
- The model demonstrated generalization capabilities, imputing gene expression profiles outside the training set.
Conclusions:
- RNA-GAN represents a significant advancement in generating biologically relevant synthetic histology images.
- The model's ability to integrate gene expression data enhances its utility in computational pathology research.
- RNA-GAN shows promise for data augmentation and imputation in digital pathology workflows.
