Common practice of the lab is to adhere to the following guidelines in scientific communication.
IN MANUSCRIPTS & PRESENTATIONS
Always state the initial sample size in the manuscript, and indicate exact criteria/reasons for why certain data sets where not included in the analyses.
Reporting fMRI results
(from Nichols, T.E et al. (2016) Best Practices in Data Analysis and Sharing in Neuroimaging using MRI. CORBIDASreport.)
For reporting single univariate outcomes, like average BOLD response in an ROI or global mean FA, there is a wealth of best practice guidelines available [Altman2008]. For mass univariate models, there are four general classes of information that need to be carefully described: Effects tested, tables of brain coordinates, thresholded maps, parcellated maps, and extracted data.
A complete itemization of the effects tested must be presented, identifying the subset that are presented. This is necessary to understand the true magnitude of the multiplicity involved and the potential danger of selection biases. For example, if a study has a multifaceted design allowing various main and interaction effects to be considered, effects tested and omitted should be enumerated, including references to previously published results on the current dataset. A full sense of how extensively the data has been explored is needed for the reader to understand the strength of the results.
Tables of coordinates historically have often been the only quantification of the results, and now should be complemented with sharing of full statistic images (see, e.g., NeuroVault ).
- If 10 coordinates are reported, each table or sub-portion of a table should be clearly labeled as to what contrast / effect it refers to (nature of the contrast, individual versus group result, group size), and should have columns for: Anatomical region, X-Y-Z coordinate, T/Z/F statistic, and the P-value on which inference is based (e.g. voxel-wise FWE corrected P; or cluster-wise FDR corrected P);
- If cluster-wise inference is used, the cluster statistic (e.g. size, mass, etc) should be included.
- Avoid having multiple columns of results, e.g. multiple XYZ columns, one for increases, one for decreases, or one for left hemisphere, one for right hemisphere.
The table caption should clearly state (even if in repetition of the body text:
- The significance criterion used to obtain these coordinates, and whether they represent a subset of all such significant results (e.g. all findings from whole-brain significance, or just those in a selected anatomical region).
- If T or F statistics are listed, supply the degrees of freedom. Whenever possible, provide effects sizes at the selected coordinates together with 95% confidence intervals. Finally, the space (i.e., Talairach, MNI, fsaverage) of the coordinate system should be noted.
The thresholded map figures perhaps garner the most attention by readers and should be carefully described. Note for Roshan's lab: we recommend and urge people to instead of the thresholded maps report dual-coded images using Slice Display, which simultaneously display contrast estimate and unthreshold t-statistic maps (Allen, E. A., Erhardt, E. B., & Calhoun, V. D. (2012). Data Visualization in the Neurosciences: Overcoming the Curse of Dimensionality. Neuron, 74(4), 603–608). If you decide to go against this recommendation, then adhere to the following bullet points:
- In the figure caption clearly state the type of inference and the correction method (e.g. “5% FWE cluster size inference with P=0.001 cluster-forming threshold”), and the form of any sub-volume corrections applied.
- For small volume or surface ROI corrections, specify whether or not the ROI was identified prior to any data analysis and how it was defined.
- Always annotate threshold maps with a color bar for the statistic values; when showing multiple maps, use a common color bar when feasible; and always indicate right and left.
- Avoid common fallacies in interpreting maps; e.g. an activation in region A but not region B doesn’t mean A is significantly more active than B [Poldrack2008], and lack of activation is not evidence of no activation. (Note: Despite being applicable in manuscripts, this point also very well apply to presentations).
- Most important, publicly share the original statistic images, unthresholded and thresholded, so readers can explore the maps themselves in 3D.
Extracted data from images aids the interpretation of the complex imaging results, and is presented as effect magnitudes, bar plots, scatter plots or activation time courses. Computed from a single voxel/vertex, or an average or principal component of a set of voxels/vertices, they however present a great risk for “circularity” [Vul2009; Kriegeskorte2009]. Specifically, when the voxels summarized are selected on the basis of a statistic map, they are biased estimates of the effect that map describes. Thus it is essential that every extracted summary clearly address the circularity problem ; e.g. “derived from independently-formed ROI”, or “values based on voxels in a significant cluster and are susceptible to selection bias”. When working with single regions and uncorrected P-values, consider the current discussions on the limitations of P-values [Wasserstein2016] and in particular how P=0.05 can amount to very weak evidence of an effect [Nuzzo2014].
Always indicate where/if the raw data and code has been shared in the manuscript. If raw data and code has not been shared, always provide a reason for why it has not been shared (e.g., privacy issues, lack of approval from participants, due to time constraints etc).