A Novel Approach to Multimodal Biomedical Data Integration via Hybrid Knowledge Graphs and Large Language Models

multimodal biomedical data integration: Heterogeneous multimodal medical data makes data interpretation, interoperability, and clinical decision support increasingly difficult.

unstructured biomedical text processing: Unstructured clinical text and medical literature need entity extraction, normalization, and relation identification to connect structured and unstructured biomedical data.

biomedical semantic inference: Biomedical informatics requires semantic inference and inferential reasoning over integrated multimodal biomedical data.

healthcare AI trustworthiness: The study addresses data heterogeneity, privacy protection, and explainability of AI-generated insights in biomedical AI systems.

multimodal biomedical data integration: Heterogeneous multimodal medical data makes data interpretation, interoperability, and clinical decision support increasingly difficult.

hybrid KG-LLM biomedical integration: The paper proposes a hybrid system combining large language models and biomedical knowledge graphs for multimodal information integration and semantic inference.

multimodal biomedical knowledge graph construction: The approach builds a knowledge graph that semantically links biomedical entities extracted from multimodal sources.

LLM-based clinical entity and relation extraction: The LLM component performs entity extraction, normalization, and relation identification from unstructured clinical text and medical literature.

GNN reasoning over biomedical KG: The paper designs graph neural network frameworks to perform inferential reasoning over the synthesized biomedical graph.

federated real-time KG extension: Future work will explore real-time knowledge graph updates and federated learning for wider medical deployment.

better than single-modality models:

improved diagnostic and treatment support:

actionable cross-modal clinical links: