vim 2: two questions

Hello all,

I have two questions regarding vim 2 dataset:

What causes NaNs to be found in BOLD response voxels? I have to elaborate this question little bit more, so to follow my contemplation please take a look at this picture http://picpaste.com/NaN_values_ROI_voxels-tRU6masC.png (subject 1, first BOLD response), there you can see highlighted/masked ROI pixels in each 64x64 slice. White pixels correspond to mentioned NaNs, red ones are all ROI's indexes together. What bothers me most is that some ROI indexes point right to these NaNs. So, at first, I thought that there was some mismatch in pyTables library, therefore I tried to load the same data with h5py, Octave, even Matlab itself, but results were always the same. Am I doing something wrong or is this actually due to how fMRI data acquisition/processing works and everything is in fact ok? Well, there is offcourse no problem with getting rid of these NaNs using numpy.nan_to_num, but anyway I need to understand clearly to this discrepancy before any experiments. Maybe I am only uncertain and this is not a problem at all. Last thing, only a small remark that is not related to this question - indexing of "data" variable in example load code is swapped (vim 2 data description pdf, "how to get started" section: "v1lh_resp = data[:,v1lh_idx]" should be "v1lh_resp = data[v1lh_idx,:]"). 

My second question is simpler, is there any possibility of getting minimally preprocessed fMRI data for vim 2, like the gzipped NIFTI files attached to vim 1 data set? This could be really helpful for my experiments.

Please let me know what you think about it. Thank you in advance.

Jan

Hi Jan,

Thanks for your questions.

1) In the movie dataset, the NaNs appear around the edges of the scanning volume due to motion correction. Since the ROIs were defined using separate localizer scans that used a different scanning prescription, there are some voxels that fall within the ROI definitions that have NaN values in the movie dataset. Using numpy.nan_to_num to get rid of the NaNs for analysis should be fine. Thank you for pointing out the error in the indexing order - we will fix the README document.

2) Unfortunately we don’t have the raw data available in NIFTI format, since they were acquired on a 4T Varian magnet that used FID format. We do not currently have any code for converting the FID data to NIFTI format. However, we will annotate and release the raw data in FID format and as an addition to this dataset in a few weeks.

Cheers,

Natalia Bilenko, Gallant lab

Previously Jan Sileny wrote:

Hello all,

I have two questions regarding vim 2 dataset:

What causes NaNs to be found in BOLD response voxels? I have to elaborate this question little bit more, so to follow my contemplation please take a look at this picture http://picpaste.com/NaN_values_ROI_voxels-tRU6masC.png (subject 1, first BOLD response), there you can see highlighted/masked ROI pixels in each 64x64 slice. White pixels correspond to mentioned NaNs, red ones are all ROI's indexes together. What bothers me most is that some ROI indexes point right to these NaNs. So, at first, I thought that there was some mismatch in pyTables library, therefore I tried to load the same data with h5py, Octave, even Matlab itself, but results were always the same. Am I doing something wrong or is this actually due to how fMRI data acquisition/processing works and everything is in fact ok? Well, there is offcourse no problem with getting rid of these NaNs using numpy.nan_to_num, but anyway I need to understand clearly to this discrepancy before any experiments. Maybe I am only uncertain and this is not a problem at all. Last thing, only a small remark that is not related to this question - indexing of "data" variable in example load code is swapped (vim 2 data description pdf, "how to get started" section: "v1lh_resp = data[:,v1lh_idx]" should be "v1lh_resp = data[v1lh_idx,:]"). 

 

My second question is simpler, is there any possibility of getting minimally preprocessed fMRI data for vim 2, like the gzipped NIFTI files attached to vim 1 data set? This could be really helpful for my experiments.

 

Please let me know what you think about it. Thank you in advance.

Jan

Hello, I just wanted to mention that we added raw data for this dataset (available for download via NERSC) in case it might be useful. We apologize for the delay.

Best regards,

Natalia Bilenko, Gallant Lab