Research on Deep Statistical Learning Theory and Methods for Brain Imaging Data

Neuroimaging techniques provide diverse forms of data (e.g., f/sMRI and DTI data), capturing neural characteristics of the brain from various perspectives. These techniques hold significant potential in
exploring the neural mechanisms underlying psychological and behavioral processes, diagnosing diseases such as Alzheimer's disease, Parkinson's disease, depression, anxiety, schizophrenia, bipolar disorder, autism spectrum disorders, brain tumor as well as stroke, and predicting short- and long-term cognitive and behavioral outcomes. Developing new statistical models and learning methods for brain imaging data analysis based on spatial structural information will help improve and advance applications in these areas.

This project aims to develop a series of new theories and methods for neuroimaging data, proposing new techniques for dimension reduction of imaging big data and deep statistical learning theory and
methods, including functional data learning and modeling of structural magnetic resonance imaging data, tensor decomposition models, high-dimensional heteroscedastic regression, and Neyman-Pearson classification; variability modeling and functional data analysis methods for functional magnetic resonance imaging; as well as graph models and connectivity matrix estimation for diffusion
tensor magnetic resonance data. This project will systematically investigate these models and theoretical methods, design and optimize algorithms, and develop software. All of these will be tested and
applied in clinical diagnosis and prediction, treatment and evaluation, and community public health services, contributing to the advancement of the fusion of artificial intelligence and health
technology.