Multivoxel analysis for functional magnetic resonance imaging (fMRI) based on time-series and contextual information: relationship between maternal love and brain regions as a case study

• Published in: Multimedia tools and applications Vol. 75; no. 9; pp. 4851 - 4865
• Main Authors: Chen, Bo-Wei, Ou, Yang-Yen, Kung, Chun-Chia, Yeh, Ding-Ruey, Rho, Seungmin, Wang, Jhing-Fa
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2016; Springer Nature B.V
• Subjects: Activation; Adrenal glands; Analysis; Brain; Brain research; Case studies; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Discriminant analysis; Emotions; Feature extraction; Love; Magnetic resonance imaging; Mathematical models; Medical imaging; Mothers; Multimedia computer applications; Multimedia Information Systems; NMR; Noise; Nuclear magnetic resonance; Pattern recognition; Principal components analysis; Scientists; Special Purpose and Application-Based Systems; Statistical methods; Studies; Support vector machines; Variance; Variance analysis; Maternal love recognition; Blood-oxygen-level-dependent (BOLD) contrast edge; Functional magnetic resonance imaging (fMRI); BOLD contrast centroid; Multivoxel pattern analysis
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-014-2020-4

This study explores the relationship between maternal love and brain regions by using functional magnetic resonance imaging (fMRI). Also, a novel pattern analysis for fMRI based on the discovered brain regions is proposed in this work. Firstly, to identify which region responds to stimuli, a statistical t -test is used after the scan. Based on these preliminary regions of interest, this study develops discriminant features extracted from multivoxels for cognitive modeling. In total, five parameters are used in the time-series and contextual analysis, including the proposed blood-oxygen-level-dependent (BOLD) contrast edge, BOLD contrast centroid, activated voxels, mean, and variance. Furthermore, this study also proposes a test function for examining voxel activation based on variance, so that insignificant voxels and irrelevant outliers can be removed from the features. After the feature extraction from brain regions of interest, the analysis subsequently uses Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) for reducing the feature size. Lastly, this study adopts a computer-aided pattern recognizer, the Support Vector Machine (SVM), to facilitate automation of the proposed analysis. A dataset consisting of brain-scanning images from 22 subjects was used for evaluation. The statistical result shows that the neural circuitry associated with maternal bonds indeed appears in the relevant brain regions as indicated by the other research. Such regions are subsequently used for assessment of the proposed analysis. Classification result shows that the proposed approach can effectively identify activated samples. Besides, our system achieves an accuracy rate of as high as 83.33 %. A comparison among different systems reveals that the proposed system is superior to the others and establishes its feasibility.