The paper “Statistical Inferences for Complex Dependence of Multimodal Imaging Data”, co-authored by Prof. Jinyuan Chang, Prof. Jing He, and doctoral student Mingcong Wu from the Joint Laboratory of Data Science and Business Intelligence at Southwestern University of Finance and Economics (SWUFE), in collaboration with Prof. Jian Kang from the University of Michigan, has been officially accepted by Journal of the American Statistical Association, one of the most prestigious journals in the field of statistics.
Abstract
Taking the multi task fMRI data analysis of the Human Connectome Project (HCP) as an example, this paper investigates three hypothesis testing questions that have received widespread attention in multimodal image data analysis: (a) testing whether brain activation patterns are independent between different fMRI tasks in each region of interest; (b) Under each fMRI task, examine whether brain activation patterns are independent between different brain regions; (c) Under different fMRI tasks, examine whether brain activation patterns are independent between different brain regions. In multimodal image data analysis, the high dimensionality, strong spatial correlation, and complex structure of the data often pose challenges to statistical inference. This paper presents the three issues mentioned above as a general form of independence testing between high-dimensional random vector components, and proposes a global testing method and a multiple testing method for controlling the error detection rate (FDR) applicable to these characteristics of multimodal image data. The proposed method is not only applicable to multimodal image data analysis, but also to other problems of testing the independence between components of high-dimensional random vectors, and does not require any structural assumptions about the covariance matrix of high-dimensional random vectors. In order to improve computational efficiency, the paper also proposes a distributed algorithm. Numerical simulation and HCP multi task fMRI data analysis show that the newly proposed method has excellent performance in discovering complex correlations in multimodal image data.
Author Introduction
Jinyuan Chang is the Executive Director of the Joint Laboratory of Data Science and Business Intelligence at SWUFE. He is a Guanghua Distinguished Professor, Doctoral Supervisor, recipient of the National Science Fund for Distinguished Young Scholars of China, a Distinguished Expert of Sichuan Province, and a member of the Sichuan Province Statistical Expert Advisory Committee. His primary research interests include ultra-high dimensional data analysis and high-frequency financial data analysis.
Jing He is an Associate Professor at the Joint Laboratory of Data Science and Business Intelligence at SWUFE. Her research focuses on high-dimensional data analysis and spatiotemporal data analysis.
Jian Kang is a Professor at the University of Michigan. His research focuses on statistical methods for large-scale complex biological data.
Mingcong Wu is a Ph.D. student in statistics at the School of Statistics, SWUFE, under the supervision of Professor Jinyuan Chang. His research interests include ultra-high dimensional data analysis and Gaussian approximation.
