Modeling Alzheimer's disease progression via amalgamated magnitude-direction brain structure variation quantification and tensor multi-task learning

Machine learning (ML) techniques for predicting the progression of Alzheimer’s disease (AD) can greatly assist researchers and clinicians in establishing effective AD prevention and treatment strategies. The problems of monotonicity of data forms and scarcity of medical data are the main reasons that currently limit the performance of ML approaches. In this research, we propose a novel similarity-based quantification approach that simultaneously considers the magnitude and direction relationships of structural variations among brain biomarkers, and encodes quantified data as third-order tensors to solve problem of data form monotonicity, then combining tensor multi-tasking learning model to predict AD progression. In this model, the prediction of each patient is considered as a task, and each task shares a set of latent factors obtained by tensor decomposition, knowledge sharing between tasks can improve the generalization of the model and solve the problem of scarcity of medical data. The model can be utilised to efficiently predict the progression of AD integrating magnetic resonance imaging (MRI) data and cognitive scores of AD patients at different stages. To evaluate the effectiveness of the proposed approach, we conducted extensive experiments utilising MRI data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). The results reveal that the proposed model predicts AD progression more accurately and consistently than single-task and state-of-the-art multi-task regression approaches on various cognitive scores. The proposed approach can recognize brain structural variation in patients and apply it to reliably predict and diagnose AD progression.