Characterization of familial Alzheimer’s disease using unsupervised learningalgorithmsA data analysis of Alzheimer’s disease cases

Abstract: Familial Alzheimer's Disease (FAD) is characterized by an early onset, cognitive impairment and dementia. Besides the symptomatic characterization, FAD shows an abundance of amyloid beta and tau protein in the patients' brains. In this thesis, a selected data set of closely related FAD cases was analyzed using two different unsupervised learning algorithms. These clusters were then compared with the severity of medical symptoms. A simple K-means clustering analysis, and a more sophisticated approach using self-organizingmaps (SOM) were used to determine different clusters within the data. The SOM analysisshowed the selected amyloid beta data is distributed quite uniformly. This not only explained a rather bad performance of K-means, but offered an opportunity to investigate cases that are especially different from each other in regard to the amounts of cerebral amyloid angiopathy, diffuse and cored plaques. These cases showed correlations within clinical data, and for some features showed significant differences. Due to the lack of data this thesis is constructed as a proof of concept. Further investigation with a larger sample size or different feature set should be used to establish the relations between pathology and clinical symptoms presented in this thesis in a statistically sound manner.