Knowledge-Based Discrimination in Alzheimer’s Disease

Abstract

Our goal is to propose a methodology to integrate clinical, cognitive, genetic and neuroimaging results, a disparate assembly of categorical, ordinal, and numerical data, within the context of Alzheimer’s disease (AD) research. We describe a knowledge-based, explicit decision model for the discrimination of AD subjects from normal controls (CTRL) based on a recent approach for diagnostic criteria for AD. We proceed by (a) creating a set of rules for each datum source; (b) integrating these disparate data into an information feature in the form of a binary string; and (c) using a machine learning approach based on the Hamming distance as an information similarity measure for the discrimination of probable AD from CTRL. When tested on 249 subjects from the multi-centric Alzheimer’s Disease Neuroimaging Initiative study, we reach up to 99.8% discriminative accuracy using a 9-bit information string in a series of 10-fold validation experiments. We expect this versatile approach to become useful in the identification of specific and sensitive phenotypes from large amounts of disparate data.