Update MetaAnalysis_03_NeuroQuery.rst

docs/MetaAnalysis/MetaAnalysis_ShortCourse/MetaAnalysis_03_NeuroQuery.rst

@@ -1,8 +1,8 @@
 .. _MetaAnalysis_03_NeuroQuery:
 
-===============================================================
-Meta-Analysis Tutorial #3: Online Meta-Analysis with Neuroquery
-===============================================================
+==============================================================================
+Meta-Analysis Tutorial #3: Online Meta-Analysis with Neurosynth and Neuroquery
+==============================================================================
 
 ---------------
 
@@ -34,6 +34,22 @@ Neurosynth is straightforward to use, and any further questions about the interf
 Neuroquery: Refinement of Search Terms
 **************************************
 
+Although GingerALE and Neurosynth are very good at finding the overlap among studies for search terms, these terms are often restricted to single words, and moreover to words that occur frequently in the neuroimaging literature. What if we wanted to perform a meta-analysis on a more complicated phrase, such as "self-referential processing", or what if we wanted to use a relatively rare term, such as "prosopagnosia"?
 
+These questions were addressed by a new online meta-analysis tool called `Neuroquery <https://neuroquery.org/>`__, which emphasizes *prediction* for a given search term or phrase, instead of calculating the overlap from studies that have already been published. Take the search term "prosopagnosia", for example, which was also the example used in the original `Dockès et al. 2020 paper  <https://elifesciences.org/articles/53385>`__. There are only a handful of studies which have studied prosopagnosia using neuroimaging, which means there are not enough to run an effective meta-analysis.
 
-For meta-analyses that test where there is probable overlap in activation across studies, both GingerALE and Neuroynth will generate Z-statistics at each voxel, and these Z-statistics in turn can be used to calculate t-tests between maps. 
+With Neuroquery, however, this limitation can be overcome by analyzing the full text of each article (as opposed to just the abstract, as is done by Neurosynth), and then by using a technique called "semantic smoothing" to add weight to semantically related terms. This results in predictive maps of voxels that are expected to be associated with a certain cognitive process, such as prosopagnosia, which would be otherwise impossible with conventional meta-analysis.
+
+.. figure:: Neuroquery_03_Prosopagnosia.png
+
+  Figure 5 from Dockès et al., 2020, comparing the results of example search terms between Neurosynth and Neuroquery. Both perform well when examining a robust and frequent search term, such as STS or PSTS (Superior Temporal Sulcus and Posterior Superior Temporal Sulcus, respectively), but Neurosynth does not have enough activation foci to create meta-analysis maps for infrequent search terms such as "Huntington" or "Prosopagnosia". Neuroquery is able to leverage the context of each search term to calculate its correlation with other search terms, and thereby generate a predictive map for that term.
+
+Similar to Neurosynth, you type a search term into the Query box, click on ``Generate Brain Map``, and a few seconds Neuroquery will display a predictive map of that search term on an MNI brain. Below the Query box, you will see the search terms that were used to generate the map, along with the relative weight given to each one; in addition, related terms are shown below in the ``expansion`` field, and you can hover your cursor over any of these terms to see a preview of what the predictive map would look like. The orthogonal views can also be changed by clicking on ``Swap View``, and the color and transparency of the map can be edited by clicking on the heatmap box below the viewing panes. Lastly, you can download these maps by clicking the ``Download map`` button below the viewing pane. You can then load this in the software package of your choice, determine where the peak is, and create a sphere around it, if you want.
+
+.. figure:: Neuroquery_03_Website.png
+
+.. note::
+
+  For meta-analyses that test where there is probable overlap in activation across studies, both GingerALE and Neuroynth will generate Z-statistics at each voxel, and these Z-statistics in turn can be used to calculate t-tests between maps. The maps generated by Neuroquery on the other hand are likelihood estimates, and should not be used for inferential tests.
+
+We will complete our survey of meta-analysis by learning how to use NiMARE, a Python-based software package. To learn how to use it, click the ``Next`` button.