A Denoising Autoencoder Stacked Deep Learning Method for Clinical Trial Enrichment and Design Applied to Alzheimer’s Disease

In this research, we first present some background on the sample size estimation for conducting clinical trials, discussing the necessity of a computational enrichment criterion. The Denoising Autoencoder Stacked Deep Learning (DASDL) design and development are directly motivated by the optimal enrichment design. Although there are many types of deep architectures in the literature, we focus our presentation using two of the most widely used models stacked denoising autoencoders and fully-supervised dropout. The ideas presented here are applied to any such architectures used for learning problems in the small-sample regime. In this work, we propose a novel, scalable, deep learning method that is applicable for learning problems in the small sample regime and obtains reliable performance. The results show via extensive analyses using imaging, cognitive, and other clinical data alongside a ROC curve analysis. When used as trial inclusion criteria, the new computational markers result in cost-efficient clinical Alzheimer’s disease trials with moderate sample sizes.

Keywords:

Pattern Analysis , Medical imaging , deep learning , Alzheimer’s disease , Signal processing