New program TRR for test-retest reliability analysis

AFNI Message Board
1

TRR performs test-rest reliability analysis for behavior data as well as region-based neuroimaging data. If no multiple trials are involved in a dataset, use the conventional intraclass correlation (ICC) with, for example, 3dICC for neuroimaging data. However, when there are multiple trials for each condition, the traditional intraclass correlation may underestimate the reliability to various extent. For more details, check out the help of the program and the following manuscript:

https://www.biorxiv.org/content/10.1101/2021.01.04.425305v1

P.S.- 3dLMEr could be utilized with the option -TRR to estimate test-retest reliability with trial-level data for whole-brain analysis; however, it may only work for data with situations such as a single effect (e.g., one condition or average across conditions). In other words, 3dLMEr -TRR could fail for contrasts.

2

Hello,expert!
I have this situation: four subjects took part in a taste fmri experiment,Each one have five same session,and Each session has five run.Each run have the beat effects of five different task (A,B,C,D,E) and Each run has executed Each type task five times.Thus,each session has 665=180 trails. In addition,
all of the five session are the same.So I want to calculate the ICC value of the 4*5 session concluding five different tasks.How do I use the ‘cond’ to distinguish between different tasks,because there are five tasks.Dose it only work in this situation:‘the situation with two conditions for a group of subjects that went
two sessions’.
Thanks for your reply!

3

Unfortunately it is difficult for me to understand your goal and your data structure.

  1. Are you trying to calculate the ICC at the whole brain or region level?
  2. ICC for each task separately or for some contrast among the 5 tasks?
  3. “each session has 665=180 trails” - what are the two 6s?
4

Dear experts:
Thanks for your reply.In my situation,1)I want to calculate the ICC at whole brain.By the way ,Is there difference of calculating ICC between at whole brain and at the region level? Dose the region level just mean a smaller mask?
2) In my task,subjects tasted bitter,salty,sour ,sweet,and umami solution.ICC for each task separately.
3) each session has 6 run.and a run has six times of five tastes stimulation.So each session has 665=180 trails.In addition,I collected four subjects fMRI data and each subject has the same five session.By the way,Is there a good way perform group analysis for this data structure,because number of participants is too small,Can I just think the 20 sessions as 20 subjects to perform group analysis?It is really a difficult situation for me.
Looking forward to your early reply!

5

With such as small subject sample size as 4 subjects, it could be challenging to perform group analysis as well as reliability estimation.

For conventional ICC, you can obtain the effect estimates for each task, and feed them into 3dICC.

6

Hello, As I described before, each session has 6 run.and a run has six times of five tastes stimulation.So each session has 665=180 trails.In addition,I collected four subjects fMRI data and each subject has the same five session.In subhect level, can I use 3dmema to make one sample t-test of five session results of each subjects,Is it a correct way or can you give me some right suggestion to get the individual level results using the five session results? Thanks you so much.I feel so bad to collect such data structure.

7

So, you have 36 trials for each taste? It is very difficult for me to make suggestions when the information your provided is not clear to me. What is your ultimate goal: compare the 5 tastes at the group level, ICC estimation, or something else?

Also, this thread is for the program test-rest reliability. So, could you start a new thread and provide more complete information?