LMP-TX: An AI-driven Integrated Longitudinal Multi-modal Platform for Early Prognosis of Late Onset Alzheimer’s Disease

Abstract

Alzheimer's Disease (AD) is the 7th leading cause of death worldwide. 95% of AD cases are late-onset Alzheimer's disease (LOAD), which often takes decades to evolve and become symptomatic. Early prognosis of LOAD is critical for timely intervention before irreversible brain damage. This study proposes an Artificial Intelligence (AI)-driven longitudinal multi-modal platform with time-series transformer (LMP-TX) for the early prognosis of LOAD. It has two versions: LMP-TX utilizes full multi-modal data to provide more accurate prediction, while a lightweight version, LMP-TX-CL, only uses simple multi-modal and cognitive-linguistic (CL) data. Results on prognosis accuracy based on the AUC scores for subjects progressing from normal control (NC) to early mild cognitive impairment (eMCI) andeMCI to late MCI (lMCI) is respectively 89% maximum (predicted by LMP-TX) and 81% maximum (predicted by LMP-TX-CL). Moreover, results on the top biomarkers predicting different states of LOAD onsets have revealed key multi-modal (including CL-based) biomarkers indicative of early-stage LOAD progressions. Future work will develop a more fine-grained LMP-TX based on disease progression scores and identify the key multi-modal and CL-based biomarkers predictive of fast AD progression rates at early stages.