Learning Longitudinal MRI Patterns by SICE and Deep Learning: Assessing the Alzheimer’s Disease Progression

Automatic knowledge extraction from medical images constitutes a key point in the construction of computer aided diagnosis tools (CAD). This takes a special relevance in the case of neurodegenerative diseases such as the Alzheimer's disease (AD), where an early diagnosis makes the treatments easier and more effective. Moreover, the study of the evolution of the illness results crucial to differentiate the neurodegenerative process associated to the disease from the natural degeneration due to the ageing process. In this paper we present a method to construct longitudinal models from subjects using a series of MRI images. Specifically, the method presented here aims to model Gray matter (GM) variation at different brain areas of a subject across subsequent examinations, being possible to relate those regions which degenerate jointly. Hence, it allows determining variation patterns that differentiate controls from AD patients. Additionally, White matter (WM) density is also incorporated to the longitudinal model to complement the information provided by GM. The results obtained demonstrated the effectiveness of the method in the extraction of these patterns, that can be used to classify between Controls (CN) and AD subjects with 94% of accuracy, outperforming other previous methods.