Deep neural network application to multi-modal brain imaging data for
individualized predictive modeling in AD 
   Lead Investigator:    Jiook Cha
   Institution      :    Columbia University Medical Center
   E-Mail           :    Jiook.cha@nyspi.columbia.edu
   Proposal ID      :    1089



   Proposal Description:
   Here we propose to build and validate deep neural networks using multi-modal brain imaging data for accurate and reliable prediction of key clinical outcomes of AD (e.g., risk, disease progression). Dr. Cha and colleagues at Columbia University and Brookhaven National Laboratory have established high-throughput brain phenotyping pipelines using structure, diffusion, and functional MRI and machine learning algorithms for early prediction of AD risk and disease progression (i.e., MCI to AD) using a moderate size of clinical cohort (N=200) and a replication dataset (ADNI-2 N=200). This study shows the feasibility of the data-driven computational modeling for accurate prediction of disease risk, state, and trajectory in a given individual. Using the large NACC data we will, firstly, perform rigorous computational imaging analysis to estimate quantitative and robust metrics indexing complementary and multi-dimensional brain features (e.g., macroscopic and microscopic morphometry, and structural and functional connectomes) and, secondly, we will build and validate deep neural networks using the brain phenotype data to predict key clinical variables, such as diagnosis, symptom severity, cognitive scales, and disease progression. Rigorous pattern recognition enabled by the deep neural network will permit to extract salient information about brain health/disease status from complex, high-dimensional, multimodal imaging data, and use it for highly accurate clinical prediction.