Distance Concentration in High-Dimensional fMRI Datasets: Possible Analysis Implications

Abstract

Event Abstract Back to Event Distance Concentration in High-Dimensional fMRI Datasets: Possible Analysis Implications Jo Etzel1* and Todd Braver2 1 Washington University in St Louis, Psychology, United States 2 Washington University in St Louis, United States An intrinsic part of fMRI analysis is working with very high dimensionality datasets, because of the large number of brain voxels and timepoints (images) collected in most studies. Analyses often attempt to avoid dimensionality problems by working with small numbers of voxels at a time (e.g. searchlight analysis, mass-univariate GLM), but many hypotheses require describing large groups of voxels (e.g. regions of interest). An unintuitive property of high dimensional datasets is that distances can become concentrated: all points become essentially equidistant (François, Wertz, et al. 2007, "The Concentration of Fractional Distances." IEEE Transactions on Knowledge and Data Engineering). If distances are concentrated, methods relying on nearest-neighbor calculations will fail; meaningful distances (similarity) between points cannot be computed as usual (e.g., Euclidean distance, correlation). Approaches for calculating and compensating for distance concentration are studied within machine learning fields, but generally unknown within the neuroimaging community, despite their possible importance: techniques using distance metrics (e.g. representational similarity analysis, RSA) are unlikely to provide meaningful results when distances are concentrated. We estimated distance concentration in several neuroimaging datasets using the techniques described in Kabán (2012, "Non-parametric detection of meaningless distances in high dimensional data." Statistics and Computing), then examine RSA and classification algorithm performance in these datasets. We found that neuroimaging datasets vary in their degree of distance concentration, and that this variability is related to RSA and MVPA results. Quantifying intrinsic dimensionality, through techniques such as estimating distance concentration, may provide important insights into the characteristics of neuroimaging datasets, as well as providing guidance for the development and interpretation of appropriate analysis techniques. Keywords: fMRI, MVPA, representational similarity analysis, dimensionality, Distance concentration Conference: XII International Conference on Cognitive Neuroscience (ICON-XII), Brisbane, Queensland, Australia, 27 Jul - 31 Jul, 2014. Presentation Type: Poster Topic: Methods Development Citation: Etzel J and Braver T (2015). Distance Concentration in High-Dimensional fMRI Datasets: Possible Analysis Implications. Conference Abstract: XII International Conference on Cognitive Neuroscience (ICON-XII). doi: 10.3389/conf.fnhum.2015.217.00131 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 19 Feb 2015; Published Online: 24 Apr 2015. * Correspondence: Dr. Jo Etzel, Washington University in St Louis, Psychology, St Louis, United States, jetzel@wustl.edu Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Jo Etzel Todd Braver Google Jo Etzel Todd Braver Google Scholar Jo Etzel Todd Braver PubMed Jo Etzel Todd Braver Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.