Accurate and Robust Prediction of Amyloid-β Brain Deposition from Plasma Biomarkers and Clinical Information Using Machine Learning

<b>Methods:<br/></b>We analyzed 1,043 patients from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset and validated our models on 127 patients from the Center for Neurodegeneration and Translational Neuroscience (CNTN) dataset. Brain Aβ status was determined using plasma biomarkers (Aβ42, Aβ40, phosphorylated tau (pTau) 181, neurofilament light chain (NfL)), Apolipoprotein E (APOE) genotype, and clinical information (Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), age, years of education, and gender). Decision tree, random forest, support vector machine, and multilayer perceptron (MLP) machine learning methods were used to integrate all this information. We introduced a feature selection method to balance performance and the number of features, and implemented a feature matching technique to enable testing on the external dataset without retraining.<br/><br/><b>Results:<br/></b>Our system achieved an Area Under the ROC Curve (AUC) of 0.95 on the ADNI dataset (n = 340) using the full set of 11 features. When tested on the external CNTN dataset (n = 127), it achieved an AUC of 0.90. Using only five features (pTau 181, Aβ42/40 ratio, Aβ42, APOE ε4 count, and MMSE) in 341 ADNI patients, the MLP method achieved an AUC of 0.87.<br/><br/><b>Conclusion:<br/></b>Random forest, support vector machine, and multilayer perceptron methods can accurately predict brain Aβ status using plasma biomarkers, genotype, and clinical information. The model generalizes well to independent datasets and maintains strong performance with a reduced feature set, offering a cost-effective alternative to PET imaging.<br/><br/><br/>