Detecting incipient cognitive dysfunction in Preclinical Alzheimer’s Disease subjects

Abstract

[en] The number of people with Alzheimer’s disease, a degenerative brain disorder, is projected to triple worldwide by 2060, with no current cure. There has been a paradigm shift in the diagnostic conceptualization of Alzheimer’s disease (AD) based on evidence suggesting that structural and biological changes start to occur during a preclinical phase beginning decennia prior to the emergence of symptoms. However diagnostic methods for this phase are invasive and costly, thus clinicians are searching for cognitive tools for screening the population before diagnosing them. The goal of this thesis is to support the clinicians in their search for these new cognitive tests for Preclinical AD (pre-AD) detection through machine learning. In particular to provide a tool for clinicians to validate if a test is sensitive enough to detect incipient cognitive dysfunction in pre-AD subjects. To achieve this we first investigated multiple classifiers and ensemble methods to find a suitable one for the datasets supplied by the clinicians. We incorporate data augmentation through Synthetic Minority Oversampling Technique (SMOTE) to deal with the imbalanced nature of the dataset. We also compute the importance for each individual feature using a technique that assigns a score to these features based on how useful they were during the classification. We found Random Forest to be the preferred choice among the tested algorithms. SMOTE proved to be a crucial step, improving both the AUC and most importantly the sensitivity. The traditional neuropsychological tests were not sensitive enough to detect incipient cognitive dysfunction in pre-AD subjects. While the new tapping tests were more sensitive. Our tool was also easily understandable for the clinicians thanks to the feature importance.