Ensemble of 3D Densely Connected Convolutional Network for Diagnosis of Mild Cognitive Impairment and Alzheimer’s Disease

Abstract

Automatic diagnosis of Alzheimer’s disease (AD) and mild cognitive impairment (MCI) from 3D brain magnetic resonance (MR) images play an important role in the early treatment of dementia disease. Deep learning architectures can extract potential features of dementia disease and capture brain anatomical changes from MRI scans. Given the high dimension and complex features of the 3D medical images, computer-aided diagnosis is still confronted with challenges. Firstly, compared with the number of learnable parameters, the number of training samples is very limited, which can cause overfitting problems. Secondly, the deepening of the network layer makes gradient information gradually weaken and even disappears in the process of transmission, resulting in mode collapse. This chapter proposed an ensemble of 3D densely connected convolutional networks for AD and MCI diagnosis from 3D MRIs. Dense connections were introduced to maximize the information flow, where each layer connects with all subsequent layers directly. Bottleneck layers and transition layers are also employed to reduce parameters and lead to more compact models. Then the probability-based fusion method was employed to combine 3D-DenseNets with different architectures. Extensive experiments were conducted to analyze the performance of 3D-DenseNet with different hyperparameters and architectures. Superior performance of the proposed model was demonstrated on ADNI dataset.