AUTHOR=Liu Liangliang , Chang Jing , Wang Ying , Liang Gongbo , Wang Yu-Ping , Zhang Hui 

TITLE=Decomposition-Based Correlation Learning for Multi-Modal MRI-Based Classification of Neuropsychiatric Disorders

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 16 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2022.832276

DOI=10.3389/fnins.2022.832276

ISSN=1662-453X

ABSTRACT=Multi-modal magnetic resonance imaging (MRI: Structural MRI (sMRI) and functional MRI (fMRI)), which has been widely used for brain disease diagnosis in clinical practice. However, using multiple types of MRIs is challenging due to their high-dimensionality and limited availability. In addition, utilizing multiple MRI modalities jointly is even more challenging. We develop a decomposition-based correlation learning (DCL) method. To overcome the above challenges by capturing the complex relationship between sMRI and fMRI. Under the guidance of matrix decomposition, DCL takes into account the spike magnitude of leading eigenvalues, sample number, and dimensionality in the matrix. Canonical correlation analysis (CCA) is used to analyze the correlation and construct matrixes. We evaluate DCL with a classification of multiple neuropsychiatric disorders on the Consortium for Neuropsychiatric Physics (CNP) dataset. Our experimental result shows higher accuracy in comparison with several existing methods. Moreover, we find interesting feature connections from brain matrixes based on DCL that can differentiate disease and normal cases, as well as different subtypes of the disease. We believe DCL is a powerful tool that provides a better way to analyze multi-modal MRIs with a limited number of samples.