PriorPath: Coarse-To-Fine Approach for Controlled De-Novo Pathology Semantic Masks Generation

Nati Daniel , May Nathan , Eden Azeroual

Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference +

|

December 3, 2025

View abstract on PubMed

Summary

PriorPath generates realistic synthetic histopathological images with controlled cellular features for artificial intelligence (AI) in digital pathology. This method overcomes data bias by creating diverse, high-fidelity images for improved AI model training and diagnostics.

Area Of Science

Digital pathology
Computational pathology
Artificial intelligence in medicine

Background

Digital pathology leverages artificial intelligence (AI) for enhanced image analysis and diagnostics.
Biased datasets due to tissue diversity and labeling limitations hinder AI applicability.
Existing methods for synthetic histopathological image generation suffer from mode collapse, failing to capture data diversity.

Purpose Of The Study

To introduce PriorPath, a novel pipeline for generating controllable, realistic semantic masks and synthetic histopathological images.
To address the limitations of existing generative models in capturing data diversity and controlling cellular characteristics.

Main Methods

PriorPath generates detailed semantic masks from coarse-grained tissue images, enabling control over spatial distribution.
The pipeline uses these masks as priors for conditional generative approaches to create photorealistic synthetic images.
The method was validated across skin, prostate, and lung cancer datasets.

Main Results

PriorPath effectively covers the semantic mask space, outperforming previous methods in similarity to real masks.
The approach allows for precise control over tissue distributions in generated synthetic images.
Demonstrated efficacy across three distinct cancer types (skin, prostate, lung).

Conclusions

PriorPath offers a state-of-the-art, controllable solution for generating synthetic histopathological images.
This facilitates the creation of robust, unbiased AI models for computational pathology.
Enables advancements in clinical decision support, diagnostics, and early cancer detection.