Prediction of Alzheimer's Disease in the Pre-Clinical Phase using Machine Learning

Abstract

Recent advancements in neuroimaging methods, such as diffusion tensor imaging (DTI), have become a valuable resource for structural brain research, allowing for the detection of changes associated with severe neurodegenerative illnesses like Alzheimer's disease (AD). Simultaneously, computational tools based on machine learning for early diagnosis and the decision tree method are used to identify hidden patterns in data for phenotypic classification and identification of pathological scenarios. In this paper, we present a unique method for automatically discriminating between healthy controls and Alzheimer's patients using DTI values as predictive features. We demonstrate that this approach enhances classification performance (accuracy of 98%) compared to the comprehensive strategy of concatenating global features. Lastly, this type of method may be used in the feature selection phase of similar classification problems, allowing one to exploit the information richness of data while reducing the size of the feature space and, consequently, the computational effort required.