PIE-Med: Predicting, Interpreting and Explaining Medical Recommendations

Antonio Romano, Giuseppe Riccio, Marco Postiglione, Vincenzo Moscato

Abstract: We introduce PIE-Med, a novel Clinical Decision Support System (CDSS) that integrates Graph Convolutional Networks (GCNs) and Large Language Models (LLMs) to deliver interpretable medical recommendations. PIE-Med leverages GCNs to generate recommendations based on patients’ health data and validated medical knowledge, ensuring clinical relevance and robustness. Interpretability algorithms evaluate the model’s reasoning, while LLM agents translate these insights into natural language explanations for clear, context-aware recommendations. By using LLMs as auxiliary reasoning agents rather than primary decision-makers, PIE-Med mitigates risks like hallucination and biased reasoning common in LLM-driven systems. Our code is publicly available on GitHub: https://github.com/picuslab/PIE-Med.

External IDs:dblp:conf/ecir/RomanoRPM25