Landmark Detection and Alzheimer’s Disease Prediction using two Stage CNN and Landmark Features
Alzheimer's disease is an irreversible and gradual brain disorder that ruins memory and thinking abilities slowly, and perhaps the ability to perform the simpler task. Magnetic resonance imaging (MRI) has been shown to be an effective diagnostic tool for Alzheimer's disease (AD).In this paper, a landmark-based feature extraction method for AD diagnosis using MR images is used, which requires no nonlinear registration or tissue segmentation. The discriminative landmarks are automatically discovered from the whole brain. Here landmarks are detected by using Two-Stage Task-Oriented Deep Neural Networks. This method consists of two Convolutional Neural Networks(CNN). In the first CNN, image patches are taken as input in order to learn inherent associations between local image patches and target landmarks. In the second CNN, the entire image is considered as input. These two-stage CNN models jointly predict the actual location of the landmark. In the next level, HOG features and contextual longitudinal features are extracted based on those detected landmarks. The extracted features are then classified by using a commonly used classifier SVM (Support Vector Machine) which classify the AD subject or mild cognitive impairment (MCI) patients from healthy controls(HCs).
Keywords - Convolutional Neural Networks, Support Vector Machine, Anatomical landmark detection, Alzheimer’s disease, landmark-based feature extraction.