Leveraging Knowledge Graphs and Advanced Al for Medical Interpretation

Abstract

Knowledge graphs are graph-based structures that represent entities and relationships as triples, enabling machines to perform semantic reasoning. They are widely used in fields such as healthcare. This project aims to build a medical diagnostic system based on a knowledge graph. The system integrates explainable deep learning models and leverages knowledge graph embeddings to enhance interpretability. The design follows multiple stages: collecting and preprocessing data, transforming it into RDF format, and constructing a knowledge graph that captures patients, symptoms, risk factors, and diseases. Embeddings are generated using Node2Vec, converting graph nodes into numerical vectors suitable for deep learning models. A Graph Attention Network (GAT) is trained for diagnosis, and SHAP (SHapley Additive explanations) and LIME (Local Interpretable Model-agnostic Explanations) are used to interpret model results both globally and locally, identifying important features. The study is conducted on lung cancer diagnosis, and the obtained results are interpreted using explainable AI techniques. The embeddings and model are evaluated using accuracy, F1-score, precision, and recall. The combination of GAT with Node2Vec outperforms all other configurations and is selected for final deployment. The proposed solution demonstrates strong predictive performance and high interpretability, aligning AI decisions with medical reasoning. Key words: Knowledge Graph, Node2Vec, Graph Embedding, GAT (Graph Attention Network), Ex- plainable AI (XAI), SHAP, LIME