Abstract
Mild Cognitive Impairment (MCI) serves as a prodromal stage of Alzheimer'sDisease (AD), where early identification and intervention can effectively slowthe progression to dementia. However, diagnosing AD remains a significantchallenge in neurology due to the confounders caused mainly by the selectionbias of multimodal data and the complex relationships between variables. Toaddress these issues, we propose a novel visual-language causal interventionframework named Alzheimer's Disease Prediction with Cross-modal CausalIntervention (ADPC) for diagnostic assistance. Our ADPC employs large languagemodel (LLM) to summarize clinical data under strict templates, maintainingstructured text outputs even with incomplete or unevenly distributed datasets.The ADPC model utilizes Magnetic Resonance Imaging (MRI), functional MRI (fMRI)images and textual data generated by LLM to classify participants intoCognitively Normal (CN), MCI, and AD categories. Because of the presence ofconfounders, such as neuroimaging artifacts and age-related biomarkers,non-causal models are likely to capture spurious input-output correlations,generating less reliable results. Our framework implicitly eliminatesconfounders through causal intervention. Experimental results demonstrate theoutstanding performance of our method in distinguishing CN/MCI/AD cases,achieving state-of-the-art (SOTA) metrics across most evaluation metrics. Thestudy showcases the potential of integrating causal reasoning with multi-modallearning for neurological disease diagnosis.