I’m planning to train and test a CNN using ADHD200 preprocessed data- specifically rsfMRI (also aiming to apply data augmentation techniques to enhance the dataset). However, I’m not sure which files i should download from the ADHD200 collection on NITRIC. (The neuro bureau preprocessing initiative has different pipelines, each having different output files in here)
Understanding the data and how exactly I should use it is not exactly straightforward to me, what do you recommend?
Hi @neurobiffie, and welcome to neurostars!
Have you read the corresponding article (Bellec et al., 2016)? It might contain information to help you understand the data types. Without knowing more about your analysis it is hard to give you precise recommendations.
Best,
Steven
Hi @neurobiffie,
I’ll also note that more up-to-date minimal preprocessing with fMRIPrep has been done, and can be accessed via AWS S3 at fcp-indi/data/Projects/ADHD200/Outputs/fmriprep/fmriprep/. Using these outputs you can have more flexibility with how denoising and postprocessing is done, if that is of interest to you.
So you can copy the data with something like aws s3 cp -r fcp-indi/data/Projects/ADHD200/Outputs/fmriprep/fmriprep/ /path/to/local/. This requires you to install the AWS command line interface.
Best,
Steven
Hello Steven,
Thank you, I appreciate your reply! Well noted i will look more into it!
I was trying to go for the data that is most prevalent in literature (for deep learning approaches), even though I don’t really understand it exactly as I’m new to neuroimaging data (some papers mentioned fALFF output data were most useful for classification purposes on this dataset). Problem is even when I try to access the fALFF data files (which are supposed to be 3D NifTI files, according to the paper), I cannot seem to find its testing/training data on the nitric prepocessed website.
Best,
Depending on your storage and processing capability, you can run the fMRIPrep outputs into XCP_D (making sure to use parameters that will enable ALFF outputs). I don’t know much about fALFF, but other outputs of XCP_D such as connectivity matrices could also be useful for you, as they are also commonly used in fMRI classification studies.
Best,
Steven
Dear @Steven
I have a question related to the preproc ADHD200 data. I am using mountain duck to access the repository and trying to understand what the volumetric 4D bold files mean. Specifically, what does “Volumes.noindex/Amazon S3.localized/ADHD200/Outputs/fmriprep/fmriprep/sub-0010002/ses-1/func/sub-0010002_ses-1_task-rest_run-1_space-MNI152NLin2009cAsym_desc-smoothAROMAnonaggr_bold.nii.gz” as an example represent? does it mean the full pipeline of xcp-d applied to it? and if yes where can I find the exact parameters passed to the command line of xcp-d? and whether “Volumes.noindex/Amazon S3.localized/ADHD200/Outputs/fmriprep/fmriprep/sub-0010002/ses-1/func/sub-0010002_ses-1_task-rest_run-1_space-MNI152NLin2009cAsym_desc-preproc_bold.nii.gz” would be preferred to be used (without AROMA) to build a group connectome? I am trying to understand which one of them have global signal, csf and wm signal regressed out and which one has smoothing besides bandpass filtering to be precise!
Thanks in advance
Bassam
Hi @bassam.alfatly,
This is the preprocessed data with ICA-AROMA non aggressively applied to it. As part of the process, the data are smoothed. You can read more about aggressive vs non-aggressive on other threads (e.g., ICA AROMA agg vs non-agg).
No, that is an fmriprep output.
That is up to you, some people prefer AROMA some do not.
Neither of them have global signal, csf, wm regressed out.
Best,
Steven