Related Experiment Video
Updated: May 7, 2025
11:00
Visualization, Quantification, and Mapping of Immune Cell Populations in the Tumor Microenvironment
Published on: March 25, 2020
16.9K
Aligning knowledge concepts to whole slide images for precise histopathology image analysis
Weiqin Zhao1, Ziyu Guo1, Yinshuang Fan1
1School of Computing and Data Science, The University of Hong Kong, Hong Kong SAR, China.
NPJ Digital Medicine
|December 31, 2024
Summary
This study introduces ConcepPath, a novel framework for analyzing Whole Slide Images (WSIs) by integrating human expert knowledge with machine learning. ConcepPath enhances histopathology analysis by combining learned patterns with expert concepts for improved diagnostic accuracy.
Area of Science:
- Digital Pathology
- Computational Pathology
- Artificial Intelligence in Medicine
Background:
- Whole Slide Images (WSIs) analysis is complex due to large size and limited annotations, often treated as Multiple Instance Learning (MIL) problems.
- Existing MIL methods primarily learn from training data, neglecting the crucial role of human expert knowledge in clinical reasoning.
Purpose of the Study:
- To introduce ConcepPath, a novel knowledge concept-based MIL framework.
- To bridge the gap between data-driven learning and human expert reasoning in WSI analysis.
- To improve the accuracy and interpretability of computational pathology by incorporating linguistic knowledge concepts.
Main Methods:
- ConcepPath utilizes GPT-4 to extract disease-specific expert concepts from medical literature.
- It integrates these extracted concepts with learnable concepts for comprehensive knowledge extraction.
- WSIs are aligned to linguistic knowledge concepts using a pathology vision-language model.
Main Results:
- ConcepPath significantly outperformed state-of-the-art methods in lung cancer subtyping.
- It demonstrated superior performance in breast cancer HER2 scoring.
- The framework achieved high accuracy in gastric cancer immunotherapy-sensitive subtyping.
Conclusions:
- Knowledge concept-based MIL frameworks like ConcepPath can effectively integrate human expert knowledge into WSI analysis.
- Incorporating expert concepts alongside learned features enhances diagnostic performance.
- ConcepPath offers a promising direction for advancing computational pathology by leveraging both data and human expertise.
