Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Comprehensive multi-omic dissection and AI-prioritized target identification in inverted papilloma-associated sinonasal squamous cell carcinoma.

NPJ precision oncology·2026
Same author

Generalizable and explainable deep learning for brain MRI: a multi-cohort evaluation of 3D architectures for age and sex prediction.

Brain informatics·2026
Same author

Erratum for: Associations of MRI-derived Paraspinal IMAT and LMM with Cardiometabolic Risk Factors: Results from a German Cohort.

Radiology·2026
Same author

Counterfactual Diffusion Models Provide Interpretable Explanations of Artificial Intelligence Models in Pathology.

Cancer research·2026
Same author

Towards autonomous medical artificial intelligence agents.

Nature·2026
Same author

Spatial biomarker discovery via interpretable semantic learning in histopathology.

Cancer cell·2026

Related Experiment Video

Updated: Jun 27, 2025

Reconstruction of 3-Dimensional Histology Volume and its Application to Study Mouse Mammary Glands
10:59

Reconstruction of 3-Dimensional Histology Volume and its Application to Study Mouse Mammary Glands

Published on: July 26, 2014

14.4K

Using histopathology latent diffusion models as privacy-preserving dataset augmenters improves downstream

Jan M Niehues1, Gustav Müller-Franzes2, Yoni Schirris3

  • 1Else Kroener Fresenius Center for Digital Health, Technical University Dresden, Dresden, Germany.

Computers in Biology and Medicine
|April 28, 2024
PubMed
Summary

Latent diffusion models (LDMs) generate high-quality histopathology images, outperforming Generative Adversarial Networks (GANs). The KL-autoencoder LDM (KLF8-DM) excels in complex tissue classes, improving classifier accuracy through data augmentation.

Keywords:
Artificial intelligenceColorectal cancerComputational pathologyDiffusion modelsGenerative adversarial networksGenerative models

More Related Videos

Enhancing Tumor Content through Tumor Macrodissection
10:04

Enhancing Tumor Content through Tumor Macrodissection

Published on: February 12, 2022

10.0K
Intact Histological Characterization of Brain-implanted Microdevices and Surrounding Tissue
11:31

Intact Histological Characterization of Brain-implanted Microdevices and Surrounding Tissue

Published on: February 11, 2013

16.9K

Related Experiment Videos

Last Updated: Jun 27, 2025

Reconstruction of 3-Dimensional Histology Volume and its Application to Study Mouse Mammary Glands
10:59

Reconstruction of 3-Dimensional Histology Volume and its Application to Study Mouse Mammary Glands

Published on: July 26, 2014

14.4K
Enhancing Tumor Content through Tumor Macrodissection
10:04

Enhancing Tumor Content through Tumor Macrodissection

Published on: February 12, 2022

10.0K
Intact Histological Characterization of Brain-implanted Microdevices and Surrounding Tissue
11:31

Intact Histological Characterization of Brain-implanted Microdevices and Surrounding Tissue

Published on: February 11, 2013

16.9K

Area of Science:

  • Computational pathology
  • Medical image analysis
  • Artificial intelligence in healthcare

Background:

  • Latent diffusion models (LDMs) represent a significant advancement in image generation, surpassing Generative Adversarial Networks (GANs) in stability and quality.
  • Generative models are crucial in computational pathology for secure data sharing and augmenting limited datasets.
  • The comparative performance of LDMs versus GANs in histopathology tasks remains under-explored.

Purpose of the Study:

  • To systematically evaluate the impact of LDM-generated histopathology images on classification tasks compared to GANs.
  • To assess the image quality and memorization potential of different LDM architectures and a styleGAN2 model.
  • To determine the utility of LDM-generated data for enhancing histopathology classifier performance.

Main Methods:

  • Trained three LDMs (Stable Diffusion v1.4 fine-tune, KLF8-DM, VQF8-DM) and a styleGAN2 model on colorectal cancer (CRC) histology tiles across nine tissue classes.
  • Assessed image quality using expert ratings, dimensionality reduction, and distribution similarity metrics (e.g., Frechet Inception Distance - FID).
  • Investigated image memorization and evaluated the impact of generated images on training a multiclass tissue classifier.

Main Results:

  • All generative models produced high-quality images; KLF8-DM achieved superior FID and expert scores for complex CRC tissue classes.
  • VQF8-DM and styleGAN2 demonstrated better performance on simpler tissue classes.
  • Image memorization was negligible for both styleGAN2 and KLF8-DM.
  • Classifiers trained with a blend of KLF8-DM generated and real images showed a 4% increase in classification accuracy.

Conclusions:

  • KLF8-DM emerges as a leading LDM for generating high-fidelity histopathology images with minimal memorization risk.
  • The integration of LDM-generated data significantly enhances the performance of histopathology classifiers, demonstrating its value for dataset augmentation.
  • LDMs offer a promising avenue for improving data availability and model robustness in computational pathology.