Machine-Learning Aided Diagnosis of Alzheimer's Disease

Alzheimer’s disease is a neurodegenerative disorder characterized by the accumulation of amyloid-beta proteins in the brain, leading to loss of neuronal function and eventual death. Though the incidence of Alzheimer’s has risen in recent years, in no small part due to increasing lifespans, there has been little progress in the diagnosis and prevention of the disease. Diagnosis premortem is possible, but mainly through costly imaging or invasive brain biopsies, the latter of which is not recommended due to the possibility of further brain damage in the AD patient. Furthermore, AD treatments are difficult to study due to the difficulty of identifying patients as well as the diseases’ stubborn progression. Thus, there is an area of opportunity in accurately identifying these patients for both diagnostic and therapeutic purposes. There are many biomarkers correlated with the presence of AD, whether that be noticeable brain damage via scanning, the biomarkers of neuron cell death, or latent biomarkers which may cooccur in the progression of the disease. Given that these are non-linear relationships, computer-aided diagnosis may help in elucidating the diagnosis of AD. Random Forest models, given their ability to generate human-understandable trees and decision surfaces, are primed to assist medical professionals with the diagnosis of AD. This thesis analyzes several such models and evaluates their accuracies, as well as providing an overview of the state of the computer-aided medical diagnostics field.