Case-Aware Medical Image Classification with Multimodal Knowledge Graphs and Reliability-Guided Refinement

Researchers proposed a case-aware medical image classification framework that retrieves similar cases and builds a multimodal knowledge graph.

The method propagates case knowledge into visual features with graph and cross-modal attention, then refines predictions using confidence and similarity.

Experiments on multiple medical imaging datasets show better performance than strong baselines.

The approach improves both diagnostic accuracy and interpretability by making use of external knowledge and similar examples.