AI and Missingness in Diagnostics for Alzheimer’s Disease

Research Project, 2020 – 2021

Alzheimer's disease (AD) is a chronic neurodegenerative disease estimated to be the root cause of up to 70% of dementia cases. Due to the high lethality and severe impact on quality of life, early detection and possible treatments are the focus of many active research projects world-wide. A central hypothesis has been that formation of plaques in the brain is both a disease marker and causal mechanism for Alzheimer's. However, only a weak link has been established between plaque formation and the degree of dementia.

An ongoing CHAIR-SU thesis project, AI4CDAD, demonstrated the feasibility of predicting Alzheimer’s progression from readily available clinical variables using machine learning applied to the ADNI dataset. It also identified significant challenges in dealing with missing values in collected data. This project intends to advance the handling of real data with missing values, in an application to AD progression modelling, through three main aims: 1) Exploring the limits for statistical imputation of clinical time- series data using the ADNI AD data set. 2) Developing expert-in-the-loop methods for identification of proxy relations between features. 3) Applying temporal latent space models to model disease dynamics.