Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DiffusionQC: Artifact Detection in Histopathology via Diffusion Model

Zhenzhen Wang, Zhongliang Zhou, Zhuoyu Wen, Jeong Hwan Kook, John B Wojcik, John Kang

TL;DR

DiffusionQC addresses artifact detection in histopathology by learning the distribution of artifact-free images with a latent diffusion framework and flagging deviations as outliers. The method uses a VAE to map images to a latent space and a ViT based denoiser to predict noise in the forward and reverse diffusion steps; an optional contrastive adaptor is added to enlarge separation between clean and artifact images in latent space. It demonstrates competitive performance to supervised methods like GrandQC with far less annotated data, and enhanced learning improves discrimination; cross stain tests on IHC images indicate promising generalization to unseen artifact types. This annotation efficient, generalizable QC approach can be integrated into digital pathology pipelines to improve reliability with reduced labeling burden.

Abstract

Digital pathology plays a vital role across modern medicine, offering critical insights for disease diagnosis, prognosis, and treatment. However, histopathology images often contain artifacts introduced during slide preparation and digitization. Detecting and excluding them is essential to ensure reliable downstream analysis. Traditional supervised models typically require large annotated datasets, which is resource-intensive and not generalizable to novel artifact types. To address this, we propose DiffusionQC, which detects artifacts as outliers among clean images using a diffusion model. It requires only a set of clean images for training rather than pixel-level artifact annotations and predefined artifact types. Furthermore, we introduce a contrastive learning module to explicitly enlarge the distribution separation between artifact and clean images, yielding an enhanced version of our method. Empirical results demonstrate superior performance to state-of-the-art and offer cross-stain generalization capacity, with significantly less data and annotations.

DiffusionQC: Artifact Detection in Histopathology via Diffusion Model

TL;DR

DiffusionQC addresses artifact detection in histopathology by learning the distribution of artifact-free images with a latent diffusion framework and flagging deviations as outliers. The method uses a VAE to map images to a latent space and a ViT based denoiser to predict noise in the forward and reverse diffusion steps; an optional contrastive adaptor is added to enlarge separation between clean and artifact images in latent space. It demonstrates competitive performance to supervised methods like GrandQC with far less annotated data, and enhanced learning improves discrimination; cross stain tests on IHC images indicate promising generalization to unseen artifact types. This annotation efficient, generalizable QC approach can be integrated into digital pathology pipelines to improve reliability with reduced labeling burden.

Abstract

Digital pathology plays a vital role across modern medicine, offering critical insights for disease diagnosis, prognosis, and treatment. However, histopathology images often contain artifacts introduced during slide preparation and digitization. Detecting and excluding them is essential to ensure reliable downstream analysis. Traditional supervised models typically require large annotated datasets, which is resource-intensive and not generalizable to novel artifact types. To address this, we propose DiffusionQC, which detects artifacts as outliers among clean images using a diffusion model. It requires only a set of clean images for training rather than pixel-level artifact annotations and predefined artifact types. Furthermore, we introduce a contrastive learning module to explicitly enlarge the distribution separation between artifact and clean images, yielding an enhanced version of our method. Empirical results demonstrate superior performance to state-of-the-art and offer cross-stain generalization capacity, with significantly less data and annotations.
Paper Structure (10 sections, 7 equations, 4 figures, 1 table)

This paper contains 10 sections, 7 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Latent diffusion model for in-distribution learning
  4. Contrastive learning to enhance distribution shift
  5. Experiments and Results
  6. Model architecture:
  7. Dataset:
  8. Post-processing:
  9. Results:
  10. Discussion and Conclusions

Figures (4)

  • Figure 1: Conceptual comparison between supervised model and outlier detection in artifact detection.
  • Figure 2: Illustration of DiffusionQC. Cyan arrows indicate the base framework, while red arrows denote the auxiliary contrastive learning components introduced in the enhanced version.
  • Figure 3: Artifact detection on H&E-stained images. Columns: (1) WSI thumbnail; (2) annotations (red: OOF, orange: pen marking, lime: folding); (3–4) DiffusionQC basic and enhanced (white: artifact); (5) GrandQC (same color scheme as 2).
  • Figure 4: Cross-stain generalization on IHC-stained images. Left: GrandQC results, with artifact types overlaid (lime: dark spot; purple: out-of-focus). Right: DiffusionQC-enhanced results, with artifact boundaries marked in lime. Representative regions are boxed and magnified below.