Problem with fMRI - deleting potentials in raw data



Dear experts,

I am very new to fMRI and I would like some help in a problem I am currently facing. I run a fMRI event- related model consisting of 4 runs. The appropriate number of images I obtain from the scanner (adapted to the duration of my experiment) are 10.188 images or potentials per run divided in 283 volumes in order to have 36 slices per volume.In some subjects examined and in some runs I obtained less images or potentials e,g 9551 DICOM images,266 volumes.(by fault during the experiment) .When I made the conversion from DICOM to NIFTII via dcmniigui.exe the runs which had the less DICOM images could not be converted.This makes sense because the number of images in series (9551) are not divisible by the number of volumes(266).In order to solve this i divised the number of images acquired with the number of slices(36) to find the proper number of volumes (9551/36=265,3) and then multiplied 265 vol x 36=9540 images.Then I substracted 9540 from th 9551 and the 11 images found I substracted them from the end.After I made the conversion to NIFTII without a problem.But when I ran afterwards the Art Repair program in these runs which I had modified I found an abnormal number of bad slices.Note, that all the other runs in the same subject which had not been modified were without bad slices at all.I suppose that its due to the modification.I am wondering if the problem is in Art Repair program which does not read the volumes having less images.Does anybody have any idea??Thank you in advance!

Jenny F.
MD neurology resident


I would check which slices/timepoints are flagged by Art Repair and look at them in the input data (using a nifti viewer such as Mango). Perhaps the last volume in the timeseries was not reconstructed correctly (half of the slices are missing and filled with zeros) and this is what is throwing outlier detection off. If that’s the case I would remove that last volume.


The distribution of bad slices is scattered.There are too many bad slices 2000 from total 9000 and in different volumes…


So why some runs have less images - you interrupted them? If so, wouldn’t you want just to discard them (entire runs)? Otherwise I guess you might be getting disbalance in the design of you balanced your design for full run etc.
In general I either have used dcm2niix not dcmniigui.exe which is I guess the GUI version of older and no longer developed afaik dcmnii.


I did not interrupt them I had a problem with the scanner.But if my design will be unbalanced I will have to exclude so many data…There is no solution to this?


The model is event related just to know.Could I include subject with less runs?


I guess it all depends on details of the study and analyses you are going to do.