| Next revision | Previous revision |
| plp.mkda [2011/08/24 16:11] – created jochen | plp.mkda [2011/08/24 17:47] (current) – Added more info jochen |
|---|
| studies that have either no points or the maximum number of points</file> | studies that have either no points or the maximum number of points</file> |
| |
| ===== Arguments ===== | ===== Options ===== |
| | |
| | ==== .applymask ==== |
| | This is a boolean flag (''true'' or ''false'') with default ''false''. If set to true, the resulting summary statistical map will be masked with the dataset used to determine the randomization space. Please be aware that this sets all voxels outside the mask to ''0'' which makes interpolated images have a "smaller" fringe given the fact that interpolated values will be lower. |
| | |
| | ==== .asimiter ==== |
| | The number of Monte-Carlo simulation iterations, default value ''5000''. |
| | |
| | ==== .asimkeep ==== |
| | This is a boolean flag (''true'' or ''false'') with default ''false''. If set to true, the resulting VMP object not only contains the (thresholded) map of the actual (true) coordinates but also the average map of the Monte-Carlo simulated maps (which can be useful to understand why certain locations might be missing in the output). |
| | |
| | ==== .asimmask ==== |
| | This specifies the mask from which the randomization space is estimated. By default, the ''_files/colin/colin_brain_ICBMnorm.vmr'' dataset is used (which also contains cerebellum and brain stem; in case you wish for a more stringent mask, you can use the drawing capabilities of NeuroElf to further restrict the search space). |
| | |
| | ==== .asimsmpl ==== |
| | This flag can be set to either '''full''' or '''near''', with '''full''' being the default. When '''full''' is selected, the Monte-Carlo simulation draws the null-distribution coordinates from the entire search space, regardless of where the actual (true, reported) coordinates are located. This is clearly the default, as without further information the assumption can only be that, in absence of additional restrictions, false-positive peak coordinates would be distributed throughout the entire search space (usually the brain/gray matter voxels). However, given that subsequent publications are often required to already reconcile their findings with prior reported data, a considerable (but hard to estimate!) bias might already be in the data (such that coordinates which don't fit into the already existing view of a phenomenon might be under represented, leading to a non-uniform null-distribution for false-positives). This can be addressed using the '''near''' setting, for which random coordinates are drawn from the mask but, for each reported coordinate, come from a sub-set of this mask which is defined as the sphere of twice the kernel radius, excluding all coordinates that are up to half the kernel radius away. **In short, setting this flag to '''near''' uses a more stringent null distribution.** |
| | |
| | ==== .asimthr ==== |
| | This flag selects the thresholding method and is one of |
| | * '''fweXX''' - which sets a family-wise error criterion, such that the highest summary statistic value is found in at most 5% of the simulated maps |
| | * '''rescale''' - this approach stores all simulated maps and, at the end, for each voxel converts the summary statistic into a Z-score which represents the proportion (stochastic probability) of maps for which this voxel had a higher value under the null distribution |
| | |
| | The '''rescale''' approach subsequently allows to use arbitrary thresholding techniques. |
| | |
| | ==== .bbox ==== |
| | This optional setting is passed into the [[newnatresvmp]] function and defines the maximal space for which the computation is performed. The default value is the empty array. |
| | |
| | ==== .cond ==== |
| | A conditional statement can be given, which restricts the selection of coordinates (e.g. to specific types of contrasts, populations, fMRI methods, timing types, etc.). The statement must be made up of logical expressions that, for each row of data, evaluate to either ''true'' (row is included in the selection) or ''false'' (row is not included). The syntax for each such expression is by using the Dollar sign ($) followed by the name of the header field (e.g. ''Year'' or ''ContrastType'') and a numerical expression (at the moment string constants cannot be used). Parentheses can be used to group expressions accordingly. |
| | |
| | ==== .contcomp ==== |
| | This flag sets the computation type for cases where a differential contrast is investigated (e.g. "negative > neutral images"). The possible values are |
| | * '''diff''' - a straight difference between the maps is computed |
| | * '''excl''' - after computing the difference an exclusion is performed (all voxels that are implicated, even if only by a fraction, in both contrasts are set to ''0'') |
| | * '''wexcl''' - a weighted exclusion, such that voxels for which both contrasts yield values greater than a threshold are set to ''0''); this is the default |
| | |
| | This computation is performed **within study**. |
| | |
| | ==== .contcompw ==== |
| | Threshold for the '''wexcl''' setting of the ''.contcomp'' flag. The default value is 0.5. |
| | |
| | ==== .contnames ==== |
| | A cell array describing the names used in the output VMP object. Must match the number of defined contrasts. |
| | |
| | ==== .contnull ==== |
| | This flag sets the type of null distribution for contrasts (where two sets of coordinates are compared). Possible values are |
| | * '''full''' - this shuffles the assignments of contrasts to studies and labels of coordinates to contrasts |
| | * '''labels''' - this shuffles only the labels of coordinates to contrasts within studies |
| | * '''spatial''' - this constructs a null distribution by randomized spatial sampling |
| | |
| | The difference between spatial and label randomization is that the null distribution is of a different kind. Whereas the spatial null distribution tests whether the given coordinates form a non-randomly observable pattern, the label null distribution tests whether the assignment of coordinates to a specific kind of stimulus or process is random. Given that some areas in the brain are repeatedly implicated in various tasks (within a sub-class of tasks, naturally), the spatial null distribution might not be the most appropriate type of test, particularly if the number of contrasts/studies reporting the results from certain tasks is imbalanced. |
| | |
| | ==== .contrasts ==== |
| | A cell array defining the contrasts for which coordinates will be selected (for positive and negative terms). Each cell must be a ''1xC'' double array containing at least one contrast identifier. Positive numbers refer to contrasts which will be used for as positive terms, negative numbers refer to contrasts which will be used as negative terms. E.g. the array ''[1, -2]'' would specify a contrast where the first and positive term is the contrast identified by ''1'' and the second and negative term is the contrast identified by ''2''. |
| | |
| | ==== .grpmeth ==== |
| | This option sets the method used to compute the summary statistic. Possibly values are |
| | * '''ost''' - performs a one-sample t-test (highly experimental and not thoroughly tested!) |
| | * '''sum''' - computes the plain sum of the study-based maps |
| | * '''wsum''' - computes a weighted sum of the study-based maps; this is the default |
| | |
| | ==== .indivmaps ==== |
| | This is a boolean flag (''true'' or ''false'') which defaults to ''true''. If set to true, the output VMP will contain one map per contrast and study representing the actual coordinates (which allows to inspect those maps for errors and further to visualize several maps at once for additional visualization options). |
| | |
| | ==== .jbmeth ==== |
| | This option defines how the "blobs" (kernel-filled spheres) are combined across coordinates, within study. If set to '''max''', the maximum value will be used; if set to '''rsum''' the sum is computed (but restricted to a maximum value of ''1''). This setting is experimental and the '''max''' value should only be used in special cases and for testing. |
| | |
| | ==== .pbar ==== |
| | An optional progress bar object, which then visualizes the progress across the Monte-Carlo simulations. |
| | |
| | ==== .res ==== |
| | Resolution of the output VMP, default value is ''3''. Please note that **coordinates are sub-voxel exact computed** and that using higher resolutions (smaller values for this option) are only required if the data needs to be available in higher resolutions, e.g. for ROI generation, etc. |
| | |
| | ==== .scale ===== |
| | This flag determines whether the kernel is interpreted as an indicator sphere ('''indic''' setting) or as a gaussian kernel (which will be scaled to a maximum of ''1'', '''toone''' setting, which is the default). |
| | |
| | ==== .smkern ==== |
| | Smoothing kernel(s). If multiple values are given, the MKDA analysis is performed for several kernels. The default is to use a single value of ''8''. |
| | |
| | ==== .smkinterp ==== |
| | Smoothing kernel interpolation setting. Any value available for [[flexinterpn_method]] can be used. This determines what interpolation kernel will be used to determine the sub-voxel data. Given that the spatial frequency of the smoothing kernel is usually larger than twice the resolution, '''linear''' is sufficient to prevent additional smoothing. |
| | |
| | ==== .smkmdist ==== |
| |
