Classification of Alzheimer’s Disease by Cascaded Convolutional Neural Networks Using PET Images

Abstract

Accurate and early diagnosis of Alzheimer’s disease (AD) plays important role for the patient care and development of future treatment. Positron Emission Tomography (PET) is a functional imaging modality which can help physicians to predict AD. In recent years, machine learning methods have been widely studied on analysis of PET brain images for quantitative evaluation and computer-aided-diagnosis (CAD) of AD. Most existing methods extract the hand-craft imaging features from images, and then train a classifier to distinguish AD from other groups. This paper proposes to construct cascaded 3D convolutional neural networks (3D-CNNs) to hierarchically learn the multi-level imaging features which are ensembled for classification of AD using PET brain images. First, multiple deep 3D-CNNs are constructed on different local image patches to transform the local image into more compact high-level features. Then, a deep 3D CNNs is learned to ensemble the high-level features for final classification. The proposed method can automatically learn the generic features from PET imaging data for classification. No image segmentation and rigid registration are required in preprocessing the PET images. Our method is evaluated on the PET images from 193 subjects including 93 AD patients and 100 normal controls (NC) from Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. Experimental results show that the proposed method achieves an accuracy of 92.2% for classification of AD vs. NC, demonstrating the promising classification performance.