Related Experiment Video
Updated: Jan 9, 2026
05:33
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
752
A Generative Foundation Model for Scalable Cytology Image Synthesis in AI-Powered Diagnostics
Ke Zheng1, Xueyi Zheng1, Jue Wang2,3
1Department of Pathology, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-sen University Cancer Center, Guangzhou, China.
Summary
This study introduces COIN, a controllable image generation model that creates realistic cytology images. COIN enhances artificial intelligence diagnostics and supports clinical applications, overcoming data limitations and privacy concerns in pathology.
Area of Science:
- Computational pathology
- Medical imaging
- Artificial intelligence in diagnostics
Background:
- Cytology is crucial for pathological diagnosis, but AI development is hindered by limited data and privacy regulations.
- Existing AI diagnostic tools require large, diverse datasets, which are difficult to obtain due to privacy concerns.
Purpose of the Study:
- To develop COIN, a controllable cytology image generation foundation model.
- To synthesize high-quality cytology images for enhancing AI diagnostics and supporting clinical applications.
- To address data scarcity and privacy challenges in AI-driven cytology.
Main Methods:
- Trained COIN on 112,226 cytology image-report pairs from 16 anatomical sites.
- Generated high-fidelity cytology images with morphologically and semantically coherent features using diagnostic textual reports.
- Assessed model utility through expert evaluation, data augmentation, AI model training, and content-based image retrieval.
Main Results:
- Expert cytologists confirmed the anatomical and diagnostic authenticity of COIN-generated images.
- COIN significantly improved AI model performance when used for data augmentation.
- Models trained on COIN images generalized effectively to real-world datasets, even under data-scarce conditions.
- COIN demonstrated utility in content-based image retrieval for clinical decision support.
Conclusions:
- COIN provides a robust, privacy-preserving framework for scalable cytology data generation.
- The model's ability to synthesize realistic images enhances AI diagnostics in computational pathology.
- COIN is a valuable tool for accelerating AI-based diagnostic solution development and implementation.
