Neuroimaging Meta-Analyses

Neuroimaging meta-analyses tend to not use the same methodology as is used in meta-analyses in other disciplines.

Because neuroimaging is primarily concerned with identifying the neural coordinates of a physical, mental, or perceptual process, a different methodological approach is required; one that can estimate an effect-location in the brain as opposed to an effect-size.

Meta-analyses concerned with effect location take data from existing studies to ask two questions: where do the foci converge by synthesizing foci (singular, focus) or the x, y, z atlas coordinates of activation peaks, and where do the images converge by synthesizing statistical maps. 

Combining the effect-locations in the form of foci or statistical maps from individual studies can give researchers a more accurate estimation of the effect-location, overcoming some of the limitations associated with single-study neuroimaging results such as small sample size, low power, high rate of false positives, low test-retest reliability, poor reproducibility, and failure to adequately correct for multiple comparisons. (Samartisidis et al. 2017)

Equally important to understand is that meta-analysis methodology for neuroimaging is an active and evolving area of research. Below I will highlight some of the more common approaches and tools being used today. 

Image-based meta-analyses (IBMA) or also sometimes called 'mega-analyses' are the ideal method of analysis when the full T-statistic images are available. (Say in Neurovault)

Where do the images converge?

The use of full image data prevents the loss of information that has been observed when only foci are used. However, until the sharing of full statistical maps with effect magnitude estimates and standard errors for single studies becomes the norm, researchers conducting meta-analyses may only have access to location information in the form of a list of foci. (Salimi-Khoshidi et al. 2009)

IBMA Methods: 

• Mixed-Effects GLM 
• Fixed-Effects GML
• Fisher's IBMA
• Stouffer's IBMA
• z permutation

References: 

• Maumet, C., & Nichols, T. E. (2016). Minimal Data Needed for Valid & Accurate Image-Based fMRI Meta-Analysis. In bioRxiv (p. 048249). https://doi.org/10.1101/048249
• Lazar, N. A., Luna, B., Sweeney, J. A., & Eddy, W. F. (2002). Combining brains: a survey of methods for statistical pooling of information. Neuroimage, 16(2), 538-550.

Coordinate-based meta-analyses use foci as data and are the most common methodology used in neuroimaging meta-analysis. Foci data is readily available in neuroimaging literature and tools such as NeuroSynth, NeuroQuery, and BrainMap have made it easy to source imaging studies by topic or region of interest. 

CBMA Methods:

• Activation Likelihood Estimation (ALE)
• Kernel Density Analysis (KDA)
• Multilevel Kernel Density Analysis (MKDA)
• Specific Coactivation Likelihood Estimation (SCALE)
• Seed-Based d-Mapping
• MKDA Chi2 Extension

Reference: Samartsidis, P., Montagna, S., Nichols, T. E., & Johnson, T. D. (2017). The coordinate-based meta-analysis of neuroimaging data. Statistical Science: A Review Journal of the Institute of Mathematical Statistics, 32(4), 580–599. https://doi.org/10.1214/17-STS624

References

•  Samartsidis, P., Montagna, S., Nichols, T. E., & Johnson, T. D. (2017). The coordinate-based meta-analysis of neuroimaging data. Statistical Science: A Review Journal of the Institute of Mathematical Statistics, 32(4), 580–599. https://doi.org/10.1214/17-STS624
• Samartsidis, P., Montagna, S., Laird, A. R., Fox, P. T., Johnson, T. D., & Nichols, T. E. (2020). Estimating the prevalence of missing experiments in a neuroimaging meta-analysis. Research Synthesis Methods, 11(6), 866–883. https://doi.org/10.1002/jrsm.1448