TY - JOUR
T1 - Designing weighted correlation kernels in convolutional neural networks for functional connectivity based brain disease diagnosis
AU - Jie, Biao
AU - Liu, Mingxia
AU - Lian, Chunfeng
AU - Shi, Feng
AU - Shen, Dinggang
N1 - Funding Information:
This study was supported by National Natural Science Foundation of China (nos. 61976006 , 61573023 , 61703301 , 61902003 ), Foundation for Outstanding Young in Higher Education of Anhui, China ( gxyqZD2017010 ), NGII Fund, China (no. NGII20190612), and AHNU Fundamental Research Funds (nos. 1708085MF145, 1808085MF171).
Funding Information:
This study was supported by National Natural Science Foundation of China (nos. 61976006, 61573023, 61703301, 61902003), Foundation for Outstanding Young in Higher Education of Anhui, China (gxyqZD2017010), NGII Fund, China (no. NGII20190612), and AHNU Fundamental Research Funds (nos. 1708085MF145, 1808085MF171).
Publisher Copyright:
© 2020 Elsevier B.V.
Copyright:
Copyright 2020 Elsevier B.V., All rights reserved.
PY - 2020/7
Y1 - 2020/7
N2 - Functional connectivity networks (FCNs) based on functional magnetic resonance imaging (fMRI) have been widely applied to analyzing and diagnosing brain diseases, such as Alzheimer's disease (AD) and its prodrome stage, i.e., mild cognitive impairment (MCI). Existing studies usually use Pearson correlation coefficient (PCC) method to construct FCNs, and then extract network measures (e.g., clustering coefficients) as features to learn a diagnostic model. However, the valuable observation information in network construction (e.g., specific contributions of different time points), as well as high-level and high-order network features are neglected in these studies. In this paper, we first define a novel weighted correlation kernel (called wc-kernel) to measure the correlation of brain regions, by which weighting factors are learned in a data-driven manner to characterize the contributions of different time points, thus conveying the richer interaction information among brain regions compared with the PCC method. Furthermore, we build a wc-kernel based convolutional neural network (CNN) (called wck-CNN) framework for learning the hierarchical (i.e., from local to global and also from low-level to high-level) features for disease diagnosis, by using fMRI data. Specifically, we first define a layer to build dynamic FCNs using our proposed wc-kernels. Then, we define another three layers to sequentially extract local (brain region specific), global (brain network specific) and temporal features from the constructed dynamic FCNs for classification. Experimental results on 174 subjects (a total of 563 scans) with rest-state fMRI (rs-fMRI) data from ADNI database demonstrate the efficacy of our proposed method.
AB - Functional connectivity networks (FCNs) based on functional magnetic resonance imaging (fMRI) have been widely applied to analyzing and diagnosing brain diseases, such as Alzheimer's disease (AD) and its prodrome stage, i.e., mild cognitive impairment (MCI). Existing studies usually use Pearson correlation coefficient (PCC) method to construct FCNs, and then extract network measures (e.g., clustering coefficients) as features to learn a diagnostic model. However, the valuable observation information in network construction (e.g., specific contributions of different time points), as well as high-level and high-order network features are neglected in these studies. In this paper, we first define a novel weighted correlation kernel (called wc-kernel) to measure the correlation of brain regions, by which weighting factors are learned in a data-driven manner to characterize the contributions of different time points, thus conveying the richer interaction information among brain regions compared with the PCC method. Furthermore, we build a wc-kernel based convolutional neural network (CNN) (called wck-CNN) framework for learning the hierarchical (i.e., from local to global and also from low-level to high-level) features for disease diagnosis, by using fMRI data. Specifically, we first define a layer to build dynamic FCNs using our proposed wc-kernels. Then, we define another three layers to sequentially extract local (brain region specific), global (brain network specific) and temporal features from the constructed dynamic FCNs for classification. Experimental results on 174 subjects (a total of 563 scans) with rest-state fMRI (rs-fMRI) data from ADNI database demonstrate the efficacy of our proposed method.
KW - Alzheimer's disease
KW - Classification
KW - Convolutional neural network
KW - Correlation kernel
KW - Functional connectivity network
UR - http://www.scopus.com/inward/record.url?scp=85084556251&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85084556251&partnerID=8YFLogxK
U2 - 10.1016/j.media.2020.101709
DO - 10.1016/j.media.2020.101709
M3 - Article
C2 - 32417715
AN - SCOPUS:85084556251
VL - 63
JO - Medical Image Analysis
JF - Medical Image Analysis
SN - 1361-8415
M1 - 101709
ER -