Multimodal diagnosis of Alzheimers disease through causal imaging markers and risk factors

ObjectivesStage-sensitive markers can aid in early diagnosis of Alzheimers disease (AD) and can improve sensitivity, performance and interpretability. In this study, causal markers from longitudinal imaging data were extracted and integrated with risk factors to improve diagnostic models. Data DescriptionOASIS-3, a longitudinal dataset consisting of 613 controls and 214 cases with very mild to moderate Alzheimers disease is used for this study. A meta model was built using a predisposition model built from risk factors, a stage-sensitization model built from MRI markers at various stages of atrophy and a confirmatory model built using PET markers. The meta model achieved good diagnostic performance (accuracy = 93%, sensitivity = 80%, specificity = 95%). Exclusion of PET data achieved comparable performance (accuracy = 91%, sensitivity = 85%, specificity = 92%). The results demonstrate that integrating causal pathological markers with risk factors improves diagnosis and aids in elucidating stage-specific patterns of AD.