Dataset Selection Criteria
How you choose your dataset defines the question that your meta-analysis is asking.
If you choose a particular task to focus on, then your meta-analysis is looking for where in the brain there is agreement across studies using this particular task. If don't pick a specific task or paradigm to focus on, then you are asking about where there is concordance across all tasks. Either way can be valid, but pay attention to how you are building your dataset to make sure you know what your inclusion criteria are and that you follow them.
General Coordinate-Based Criteria
Practically, any coordinates can be grouped together to perform a meta-analysis that finds regional agreement across the dataset. However there are some basic criteria should be used:
Coordinates should use the same reference space. The error introduced by not converting between Talairach and MNI can be 5-10mm throughout the brain due to differences in size, origin and rotation. This can be enough to disrupt agreement you might have found if the coordinates were transformed into the same space.
Studies should use whole-brain analysis, not a priori regions. Limiting the analysis to a particular region is fine for a single analysis. However, it is very difficult to find several studies that use the same region, and combining several regional studies in a whole-brain meta-analysis is not valid.
The BrainMap database inclusion criteria cover these two points, so if you are using Sleuth then these criteria have already been met. However if you are compiling your own coordinates, pay close attention to reference spaces and regional analyses. They can get pretty tricky!
Turkeltaub, 2012: Within-Group Effects
As we discussed how to limit the effects of differing numbers of foci within each experiment, the effect of a single subject group that is re-compared across multiple experiments can also be limited. In order to avoid over-sampling a particular subject group, the ... or sub-groups of larger subject groups... combine the coordinates of multiple experiment into one large coordinate group. It's fine to have very large groups of coordinates, since we have a way to limit the Within-Experiment Effects.
Sample Size vs. Data Coherency
Meta-analyses require several data-sets to be reliable, with 20 or 30 studies as a rough minimum recommendation. This can lead toward loose criteria that include a broad sampling of related studies. However, each new type of task or new analysis metric might have a difference in effect that could dilute the results you would get with stricter criteria and more coherent data. Eickhoff et al, Science 2016
Each decision should be justified one way or the other. For example: "we included both MRI and PET data because the location of the activation shouldn't be affected by the modality" or "we included only high-resolution MRI data because we are doing a MACM investigating smaller regions and need the precision" or "we used only fMRI studies because we only found one relevant PET study and increasing the dataset size so slightly didn't seem like enough of a benefit to reducing the homogeneity of the imaging modality in our dataset"
