Towards multimodal longitudinal analysis for predicting cognitive decline

Understanding and predicting cognitive decline in Alzheimer's disease (AD) is crucial for timely intervention and management of disease symptoms and progression. While neuroimaging biomarkers and clinical assessments are valuable individually, their combined predictive power and interaction with demographic and cognitive variables remain underexplored. Our study lays the groundwork for comprehensive longitudinal analyses by integrating neuroimaging markers and clinical data to predict cognitive changes over time. Using data from the Alzheimer's Disease Neuroimaging Initiative (ADNI), we applied feature-driven supervised machine learning techniques for assessing cognitive decline predictability. We hypothesized that combining neuroimaging biomarkers with demographic and clinical assessment variables significantly improves the prediction of cognitive decline in AD. Our results show that while imaging biomarkers alone offer moderate predictive capabilities, utilizing key clinical assessment and demographic variables in conjunction with imaging biomarkers significantly improves the model performance. Furthermore, our results indicate that non-imaging variables alone can serve as effective and cost-efficient predictors of cognitive decline. We also introduce the Neuroscience-Longitudinaland-Multimodal-Analysis-System (NeuroLAMA), an open and extensive data engineering and machine learning system, to support continued investigation into prediction of cognitive decline using non-imaging variables by the community. Our study underscores the need for integrating multi-dimensional data in future longitudinal research to capture time-dependent patterns in cognitive decline and guide the development of targeted intervention strategies.