Meeting Banner
Abstract #0279

Bias and instability in graph theoretical analyses of neuroimaging data

Mark Drakesmith 1 , Karen Caeyenberghs 2 , Anirban Dutt 3 , Glyn Lewis 4 , Anthony S David 3 , and Derek K Jones 1

1 CUBRIC, Cardiff University, Cardiff, Wales, United Kingdom, 2 Department of Physical therapy and motor rehabilitation, Ghent University, Gent, Belgium, 3 Institute of Psychiatry, Kings College London, London, United Kingdom, 4 Academic Unit of Psychiatry, University of Bristol, Bristol, United Kingdom

Graph theory (GT), a powerful tool for quantifying network properties from tractography, is subject to bias and instability due to false positives (FPs). This study illustrates this bias in GT metrics and examines the effects of thresholding to reduce this bias. Thresholding does reduce the effects of FPs but also introduce their own biases. Statistical comparisons of GT metrics are also shown to be highly unstable across thresholds compared to non-GT metrics, although genuine group differences tend to be more stable. A multi-threshold permutation correction strategy is suggested to improve sensitivity of statistical comparisons of GT metrics to genuine group differences.

This abstract and the presentation materials are available to members only; a login is required.

Join Here