Medical Image Classification Using Multimodal Knowledge Graphs
A new framework for medical image classification uses multimodal knowledge graphs to simulate clinical diagnostic reasoning. The method retrieves similar cases adaptively, constructs a knowledge graph, and injects case-based features into visual representations via graph attention networks and cross-modal attention. The approach aims to improve explainability and accuracy by leveraging external knowledge and historical cases, addressing the limitations of isolated visual evidence in deep learning models.
Key facts
- The framework is called case-aware reasoning using multimodal knowledge graphs.
- It retrieves similar cases adaptively to construct a multimodal knowledge graph.
- An image-centric Graph Attention Network propagates knowledge semantics.
- Bidirectional cross-modal attention injects case-based features into visual representations.
- The method aims to simulate clinical diagnostic processes.
- It addresses limitations of existing deep learning methods that rely on isolated visual evidence.
- The approach is designed for explainable medical image diagnosis.
- The paper is available on arXiv with ID 2605.22547.
Entities
—