Automatic detection and classification of Alzheimer's Disease from MRI scans using principal component analysis and artificial neural networks

Early detection of Alzheimer's Disease (AD) is important so that preventative measures can be taken. Current techniques for detecting AD rely on cognitive impairment testing which unfortunately does not yield accurate diagnoses until the patient has progressed beyond a moderate AD. In this project, we develop a new approach based on mathematical and image processing techniques for better classification of AD. The most popular current technique analyzes MRI scans using properties of diffeomorphism which generates a mapping from one MRI to another. Since MRIs are very high dimensional vector spaces, the existing technique reduces it to three dimensions and then clusters the images according to presence or lack of AD. However, reducing a high dimensional vector space to three dimensions compromises the information in the data and thus results in some loss of accuracy. We propose to reduce the high dimensional MRI image vector space to 150 dimensions using Principal Component Analysis. In order to categorize the reduced dimensions from PCA for progression of AD, we employ a multiclass neural network. The neural network is trained initially on 230 diagnosed MRIs obtained from OASIS MRI database. We then test our trained neural network on the entire set of 457 MRIs provided by OASIS dataset to confirm the accuracy of diagnosis by our system. Our results produce nearly 90% accuracy in AD diagnosis and classification.