In the recent years there have been a number of studies that applied deeplearning algorithms to neuroimaging data. Pipelines used in those studiesmostly require multiple processing steps for feature extraction, althoughmodern advancements in deep learning for image classification can provide apowerful framework for automatic feature generation and more straightforwardanalysis. In this paper, we show how similar performance can be achievedskipping these feature extraction steps with the residual and plain 3Dconvolutional neural network architectures. We demonstrate the performance ofthe proposed approach for classification of Alzheimer's disease versus mildcognitive impairment and normal controls on the Alzheimer's Disease NationalInitiative (ADNI) dataset of 3D structural MRI brain scans.
展开▼
