21th AIAI 2025, 26 - 29 June 2025, Limassol, Cyprus

Early Detection and Prevention of Dementia: an AI-Driven Multimodal Approach

Huang Andy, Manias George, Sangiovanni Mirella, Borovits Nemania, Tamburri Damian, Ntanasi Eva, Scarmeas Nikolaos, van den Heuvel Willem-Jan

Abstract:

  Dementia is a progressive neurodegenerative condition affecting millions worldwide, highlighting the need for early and accurate detection. This study leverages the Aiginition Longitudinal Biomarker Investigation of Neurodegeneration (ALBION) dataset, integrating cognitive, psychological, physical, socio-demographic, among others, to enhance early diagnosis. In this direction, two approaches are proposed: the Always-Measured Model, which uses a limited set of consistently recorded features, and the Voting-Based Hybrid Model, which utilizesthe dataset’s full multimodal and longitudinal scope. While the AlwaysMeasured Model exhibited bias toward the Normal Cognition class (MCC = 0.64), the first-visit model within the ensemble achieved an MCC of 0.83. This demonstrates that initial-visit data alone can enable accurate detection. Additional data from follow-up visits did not improve the performance of the ensemble approach. However, the ensemble proved valuable in high-certainty cases (85.42% of instances), achieving an MCC of 0.94 and showcasing high robustness and accuracy.  

*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.