Dear experts,

I want to clarify the setting of longitudinal studies since I found my results (correlation matrices) are different. 

I have 16 patient data at baseline and follow-up. 

I import my data by "New (import)" -> "from fMRIprep dataset", after preprocessing and ICA-AROMA denoising the data. Then, I replace all functional data with ICA-AROMA denoised data in "SETUP" -> "Functional".

I tried 2 different ways to import my data and found them result in different correlation matrices.

(1) import baseline and follow-up data as different subjects, resulting in 32 subjects in "Basic information" of "SETUP". CONN automatically generates one rs condition

(2) import 16 subjects, each with 2 sessions (baseline, and follow-up). CONN automatically generates 2 rs conditions (baseline_rs, followup_rs).

All the other settings are the same.

I know the first setting is wrong since it leads to a wrong GLM model which doesn't take the subject random effect into account. I thought that correlation matrices would be the same for both ways, but, I found the correlation matrices are different. It looks like the subject random effect is also considered during the calculation. However, I want to extract the correlation matrices (I use "tanh()" function to convert them back to the Pearson correlation), and use them to calculate topological properties with the Brain Connectivity Toolbox. 

I'm confused about which way to set my analyses since I plan to calculate longitudinal changes in patients (baseline vs. follow-up) and compare healthy controls (at the baseline) with patients at follow-up (HC-baseline vs. Patients-follow-up).

I also checked the result of another analysis (3) which imports healthy controls and patients at baseline, each subject has one session.

(2) and (3) generated the same correlation matrix for the same patient, however, (1) generated a different matrix.

Best regards,

Chih-Hao

Dear experts.

I'd like to provide more information about my question.

I have 16 patient data at baseline and follow-up.

All data were preprocessed and denoised via fMRIprep and ICA-AROMA denoising.

I use "New (import)" -> "from fMRIprep dataset" to import preprocessed data into CONN, and then replace functional data with denoised data.

I tested the ROI-to-ROI matrices generated by 3 different settings.

(1) only import follow-up data 
(2) import baseline and follow-up data as 2 sessions for each patient
(3) import baseline and follow-up data separately, resulting in 32 subjects in the setting, each subject has one session (baseline or follow-up)

I extracted ROI-to-ROI matrices from  “resultsROI_Subject*_Condition001.mat" in "...\results\firstlevel\SBC_01" (For (2), also from “resultsROI_Subject*_Condition002.mat"). I noticed that (1) and (2) generated the same correlation matrices, but (3) generated different results.

I know (2) is the correct way to set up a longitudinal study. But, given that (1) and (3) result in different correlation matrices, I'm wondering why they generate different results. Would the same situation happen when I compare healthy controls and patients? (e.g. separately import healthy controls and patients as 2 conn project vs. import all data as one conn project).

An update: I did the same analysis as (1) again. However, the matrices are different with (2), but as the same as (3) this time.

I noticed that the second time automatically extracted more confounds (e.g. 'scrubbing_Dim20', 'QC_aroma_motion_Dim40', 'QC_w_comp_cor_Dim27'), but I only used CSF, white matter, and effect of rs for denoising, and these confounds/regressors were not included in SBC analysis, too. Thus, I don't really know why did I get different results from the same data and the same setting.