Using the Adaptive Smoothing Plugin
Using the Adaptive Smoothing Plugin
Overview
This plugin performs adaptive smoothing and inference for single subject task related fMRI experiments. It has to be executed after GLM analysis on a specific contrast, see
Jörg Polzehl, Henning U. Voss, Karsten Tabelow (2010) 'Structural adaptive segmentation for statistical parametric mapping', NeuroImage 52(2), 515-523
for details on the method.
Before running the plugin
Please open a VMR or FMR document and run or load a general linear model single study.
You may transform the data with percent transformation, but not with z-transformation, because the algorithm is not suited for z-transformed data.
Please make also sure that the data is not smoothed or interpolated in advance, because the algorithm assumes approximately uncorrelated data in space and therefore keeping correlation to a minimum is essential.
Parameter settings
The current GLM document and the format (VMR/FMR) is shown for information only.
- contrast: Choose one of the defined contrasts of the GLM. If multiple are choosen only the first will be considered. If you want to adaptively smooth a different contrast re-run the plugin.
- alpha: Significance level for the multiple test corrected test (values between 0.01 and 0.2)
- delta: Minimal signal value (beta) for significance test (adjust to minimal percentage change if data is percent transformed)
- bandwidth: Given in units of voxels, NOT millimeters! Larger bandwidths require considerably more computation time, but result in more smoothness in homogeneous regions. Edges between activated and non-activated regions are automatically preserved. Large spatial correlation in the data hampers the algorithm, choose a small bandwidth in such cases.
Adjusting visualization
The plugin will automatically replace the result of the BrainVoyager GLM analysis. It shows the segments of voxels where the test detected a value significantly larger (red) or lower (blue) than +/- delta at significance level alpha. The color codes beta values (percent values) NOT p-values. As there might occur single significant voxels with very large values, you can adjust the color contrast in the "Analysis -> Overlay Volume Maps..."-dialog (see Map Options tab) of BrainVoyager.
Authors
Algorithm: Karsten Tabelow, Jörg Polzehl, WIAS
Implementation: Felix Anker, WIAS