A data mining approach for automated classification of Alzheimer's disease

This work presents the creation of classifiers able to automatically diagnose Alzheimer’s disease from structural magnetic resonance images. Measurements of regions inside the human brain and how they relate to a diagnosis of Alzheimer’s disease are investigated. A genetic algorithm is used to extract a subset of brain measurements which have some of the highest predictive power when compared to the rest of the measurements. A descriptive classifier is created which uses the size of each of the hippocampal volumes of a subject and compares it to the distribution of other subjects. Based on how the individual subject’s measurements compares to the distribution, we can determine whether it is an Alzheimer’s disease predictive volume. This classifier achieves an accuracy equivalent to state-of-the-art approaches. A second descriptive classifier is created using a regression model to predict a healthy subject’s age and applying this to Alzheimer’s disease positive subjects to generate an apparent brain age. The classification accuracy of when the apparent brain age is compared to the subject’s real age is comparable to the state-of-the-art methods in the literature.