End-to-end generation and evaluation of nuclei-aware histology patches using diffusion models

TMLR Paper4589 Authors

Abstract: In recent years, computational pathology has witnessed remarkable progress, particularly through the adoption of deep learning techniques in segmentation and classification tasks that enhance diagnostic and prognostic workflows. Despite its importance, training effective deep learning models for these applications remains a significant challenge due to the need for large-scale annotated datasets. We present a nuclei-aware semantic tissue generation framework leveraging advancements in conditional diffusion modeling. Our framework generates high-quality synthetic tissue patches that are inherently annotated with instances of six distinct nuclei types. We demonstrate the efficacy of generated samples through extensive quantitative and expert evaluation.

Submission Length: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Efstratios_Gavves1

Submission Number: 4589