Elsevier

NeuroImage

Volume 4, Issue 1, August 1996, Pages 16-33
NeuroImage

Regular Article
Functional Magnetic Resonance Image Analysis of a Large-Scale Neurocognitive Network

https://doi.org/10.1006/nimg.1996.0026Get rights and content

Abstract

Many “higher-order” mental functions are subserved by large-scale neurocognitive networks comprising several spatially distributed and functionally specialized brain regions. We here report statistical and graphical methods of functional magnetic resonance imaging data analysis which can be used to elucidate the functional relationships (i.e., connectivity and distance) between elements of a neurocognitive network in a single subject. Data were acquired from a normal right-handed volunteer during periodic performance of a task which demanded visual and semantic processing of words and subvocalization of a decision about the meaning of each word. Major regional foci of activation were identified (by sinusoidal regression modeling and spatiotemporal randomization tests) in left extrastriate cortex, angular gyrus, supramarginal gyrus, superior and middle temporal gyri, lateral premotor cortex, and Broca's area. Principal component (PC) analysis was initially undertaken by singular value decomposition (SVD) of the “raw” time series observed at 170 activated voxels. This revealed a large functional distance (negative connectivity) between visual processing systems and all other brain regions in the space of the first PC. SVD of a matrix of fitted time series, and a matrix of six sinusoidal regression parameters estimated at each activated voxel, were developed as less noisy (more informative) alternatives to SVD of the “raw” data. Canonical variate analysis of denoised data was then used to clarify functional relationships between the major regional foci. Visual input analysis systems (extrastriate cortex and angular gyrus) were colocalized in the space of the first canonical variate (CV) and significantly separated from all other brain regions. Semantic analysis systems (supramarginal and temporal gyri) were colocalized and significantly separated in the space of the second CV from the subvocal output system (Broca's area). These results are provisionally interpreted in terms of underlying hemodynamic events and cognitive psychological theory.

References (0)

Cited by (134)

  • BHCR: RSVP target retrieval BCI framework coupling with CNN by a Bayesian method

    2017, Neurocomputing
    Citation Excerpt :

    Actually, we had conducted a comparison with respect to the EEG decoding methodology in Sections 4.3 and5.3. Firstly, the comparison setting of feature extraction methods in Section 4.3 had taken the classical unsupervised method PCA [31,32] and the commonly used supervised method LDA into consideration [1,9,33]. Secondly, the candidate classification algorithms we had inspected on include SVM [15] and LR [1], which were frequently used in previous works.

  • Fundamentals of Brain Network Analysis

    2016, Fundamentals of Brain Network Analysis
View all citing articles on Scopus
1

To whom correspondence should be addressed. Fax: +44 (0)171 277 1390; E-mail: [email protected].

View full text