TY - GEN
T1 - The effect of SOM size and similarity measure on identification of functional and anatomical regions in fMRI data
AU - O’Driscoll, Patrick
AU - Merényi, Erzsébet
AU - Karmonik, Christof
AU - Grossman, Robert
N1 - Funding Information:
This work was partially supported by the Program for Mind and Brain, Department of Neurosurgery, Houston Methodist Hospital. Figures are in color, request a color copy by email: [email protected], [email protected]
Publisher Copyright:
© Springer International Publishing Switzerland 2016.
PY - 2016
Y1 - 2016
N2 - We demonstrate the advantage of larger SOMs than those typically used in the literature for clustering functional magnetic resonance images (fMRI). We also show the advantage of a connectivity similarity measure over distance measures for cluster discovery and extraction. We illustrate these points through maps generated from a multiple-subject investigation of the genesis of willed movement, where clusters of the fMRI time-courses signify functional (or anatomical) regions, and where accurate delineation of many clusters is critical for tracking the relationships of neural activities across space and time. While we do not provide an automated optimization of the SOM size it is clear that for this study increasing it up to 40 × 40 facilitates clearer discovery of more relevant clusters than from a 10 × 10 SOM (a size frequently used in the literature), and further increase has no benefits in our case despite using large data sets (all data from whole-brain scans). We offer insight through data characteristics and some objective justification.
AB - We demonstrate the advantage of larger SOMs than those typically used in the literature for clustering functional magnetic resonance images (fMRI). We also show the advantage of a connectivity similarity measure over distance measures for cluster discovery and extraction. We illustrate these points through maps generated from a multiple-subject investigation of the genesis of willed movement, where clusters of the fMRI time-courses signify functional (or anatomical) regions, and where accurate delineation of many clusters is critical for tracking the relationships of neural activities across space and time. While we do not provide an automated optimization of the SOM size it is clear that for this study increasing it up to 40 × 40 facilitates clearer discovery of more relevant clusters than from a 10 × 10 SOM (a size frequently used in the literature), and further increase has no benefits in our case despite using large data sets (all data from whole-brain scans). We offer insight through data characteristics and some objective justification.
KW - CONNvis
KW - Cluster extraction
KW - Conscience self-organizing map
KW - Data-driven model
KW - Functional magnetic resonance imaging
KW - Willed movement
UR - http://www.scopus.com/inward/record.url?scp=84955294573&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84955294573&partnerID=8YFLogxK
U2 - 10.1007/978-3-319-28518-4_22
DO - 10.1007/978-3-319-28518-4_22
M3 - Conference contribution
AN - SCOPUS:84955294573
SN - 9783319285177
T3 - Advances in Intelligent Systems and Computing
SP - 251
EP - 263
BT - Advances in Self-Organizing Maps and Learning Vector Quantization - Proceedings of the 11th International Workshop WSOM 2016
A2 - O’Driscoll, Patrick
A2 - Mendenhall, Michael J.
A2 - Merényi, Erzsébet
PB - Springer-Verlag
T2 - 11th International on Advances in Self-Organizing Maps and Learning Vector Quantization Workshop, WSOM 2016
Y2 - 6 January 2016 through 8 January 2016
ER -