AD-Reasoning: Multimodal Guideline-Guided Reasoning for Alzheimer’s Disease Diagnosis

Qiuhui Chen, Xuancheng YAO, Xinyue Hu, Yi Hong

Keywords: Alzheimer's Disease, Multimodal Learning, Large Language Model

Abstract: Accurate diagnosis of Alzheimer’s disease (AD) requires combining multimodal data with established clinical guidelines. However, most deep learning models operate as black boxes, offering limited interpretability and weak alignment with medical standards. We propose AD-Reasoning, a framework for multimodal AD diagnosis that integrates structural MRIs and diverse clinical data with guideline-guided reasoning. A rule engine ensures NIA-AA diagnostic criteria, while reinforcement fine-tuning with domain-informed rewards promotes clinically consistent and transparent decision-making. Evaluated on the AD-MultiSense dataset, AD-Reasoning achieves state-of-the-art diagnostic accuracy and demonstrates improved interpretability compared with recent baselines. This work highlights a clinically grounded solution that connects large language models with medical expertise, advancing interpretable and guideline-compliant AD diagnosis.

Primary Area: interpretability and explainable AI

Submission Number: 9325