I am interested in combining three thresholded language activation maps within a single subject. The three language tasks are similar (block-related designs) and tend to produce some degree of overlapping clusters in the major language regions. My goal is to summarize the language-related activation clusters in a clean and statistically relevant manner.
I am somewhat new to ANFI and recently came across the 3dmerge program with options to perform Fisher transformation on intensity data (-gfisher) or thresholded data (-tgfisher). This seems to accomplish my main goal, but I am concerned about potentially violating certain assumptions associated with the Fisher method, namely the assumption of independence (i.e., each language task is expected to produce relatively similar activation patterns, thus not independent?). I am curious to hear from the experts on this forum about three main things:
Does it make sense to be concerned about combining non-independent thresholded correlation maps within a single subject using 3dmerge –tgfisher or -gfisher?
Are there other issues with this approach that stand out?
Are there other methods of statistically combining thresholded activation maps within a single subject that may be better than the fisher method?
Could you provide more context? Is your goal to make inference on one subject or at the group level? What is your research hypothesis about those three regions?
My goal is to summarize the language-related activation clusters in a clean and statistically relevant manner.
It's difficult to define a clean manner because statistics is meant to measure and reveal uncertainty, not certainty.
Sorry for my delayed response, I forgot to select the option to be emailed about replies (it is selected now though).
Context: I’m assessing different ways to present language activation maps from a single subject to neurosurgeons as part of a patient’s pre-surgical workup and planning. The current clinical workflow includes running ~3 simple block design language paradigms meant to elicit activation in a few key language areas, then displaying results via 3 separate thresholded activation maps overlaid on an anatomical. The specific tasks selected sometimes vary based on a subjects ability level but all are geared toward eliciting activation in classic Broca’s (e.g., verbal fluency) or Wernicke’s (e.g., listening comprehension) or both (e.g., reading an incomplete sentence and thinking of a word to fill in a blank). Activation is also commonly observed around visual word form area, DLPFC, along the middle temporal gyrus, and other language-related regions. The goal is to capture the maximum amount of activation associated with the broader construct of language which is then combined with other clinical data to help determine potential risk of language impairment after some surgical intervention. Given this clinical application, a little more emphasis is placed on type II error compared to fMRI in the research arena but type I error is still of significant concern. Additionally, the construct of interest (i.e., language) is more loosely defined than what is common in research.
I want to eventually alter this workflow by displaying only one merged thresholded language map, but I don’t know what the best method is for doing this. The gfisher option of 3dmerge seems like a simple place to start.
a little more emphasis is placed on type II error compared to fMRI in the research arena but type I error is still of significant concern
We probably all agree that it would make more sense to keep some kind of balance between the two sides of the coin. I don't have any definite suggestion to your question, but I would like to say that the statistical information extracted from the data is dependent on several factors: the quality of the data, the amount of data (e.g., samples), experimental design, etc. There is always some extent of uncertainty involved. Regardless of the methodology you decide to adopt to combine the three maps, it might be better to present the three separate maps, together with the combined final result, in a gradated fashion, not a dichotomized/thresholded one. In doing so, the surgeon can combine all sources of information, not only the statistics but also anatomical information plus expertise, before making the final judgment.
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.