Multimodal Integration of Alzheimer’s Plasma Biomarkers, MRI, and Genetic Risk for Individual Prediction of Cerebral Amyloid Burden

Alzheimer’s disease (AD), the most prevalent neurodegenerative disorder, is marked by the accumulation of amyloid-β (Aβ) plaques. Although cerebral Aβ positron emission tomography (Aβ-PET) remains the gold standard for assessing cerebral Aβ burden, its clinical utility is hindered by cost, radiation exposure, and limited availability. Plasma biomarkers serve as promising non-invasive predictors of cerebral Aβ burden, but reliance on a single marker often leads to suboptimal predictive performance. To address this, we proposed a multimodal machine learning strategy that integrates readily accessible and non-invasive features—such as plasma biomarkers, structural magnetic resonance imaging (sMRI)-derived atrophy measures, diffusion tensor imaging (DTI)-based structural connectomes (SCs), and genetic risk profiles—to predict cerebral Aβ burden and evaluate the relative contribution of each modality to predictive performance. Specifically, a random forest regressor was trained using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI; n = 150) and evaluated with leave-one-out cross-validation. Our results showed that integrating multimodal features improves the predictive power on cerebral amyloid burden: while the baseline model using plasma and clinical variables alone achieved an R² of 0.52, adding neuroimaging and apolipoprotein E (APOE) genotype features improved performance (R² = 0.617), and replacing APOE with polygenic risk scores (PRS) further enhanced accuracy (R² = 0.637). The predictive value of multimodal integration was also replicated in an independent cohort (SILCODE; n = 101). Moreover, a multiclass classifier trained with the same multimodal features achieved high accuracy in distinguishing clinical stages of Aβ burden—normal controls (NC), mild cognitive impairment (MCI), and Alzheimer’s disease (AD)—with area under the curve (AUC) values of 0.86, 0.77, and 0.93, respectively. These findings highlight the value of combining plasma, imaging, and genetic data to non-invasively estimate cerebral Aβ burden, offering a potential alternative to PET imaging for early AD risk assessment.