Deep Feature Selection and Causal Inference for Alzheimer's Disease

• Activity Number: 173 - Recent Advances on Neuroimaging Analysis
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Yuanyuan Liu* and Qiyang Ge and Nan Lin and Wenjia Peng and Rong Jiao and Xuesen Wu and Momiao Xiong
• Companies: The University of Texas Health Science Center at Houston and Fudan University and The University of Texas Health Science Center at Houston and Bengbu Medical College and The University of Texas Health Science Center at Houston and Bengbu Medical College and The University of Texas Health Science Center at Houston
• Keywords: feature selection; deep learning; causal inference; Alzheimer's disease

Abstract

Image feature selection plays an important role in prediction, diagnosis, treatment and imaging-genomics data analysis. Image feature selection is aimed to find the image areas that are most relevant for the classification of the disease status. Convolutional neural networks (CNN) have achieved great success for image classification in medical research. However, it is also well known as a 'black box' due to its low interpretability to humans. To overcome this limitation, we develop a novel general framework that integrate deep leaning and causal inference for image feature selection. The new paradigm for image feature selection consists of two stages: (1) use CNNs to find image regions that are most distinctive for disease status and (2) the state-of-the-art causal inference tools to determine if the selected image regions are causal for AD. Two stage platforms for searching causal features was applied to MRI images from 454 subjects. We successfully identified several AD causal related brain subregions.