Updates to HCP data loader (including visualization)

michaelwaskom · July 23, 2020, 3:03am

We’ve just shipped an update to the HCP project dataset loader notebook.

The update includes additional information useful for visualizing the parcel-wise data in Nilearn, including showing ROI statistics on the fsaverage5 surface:

and connectome plots using approximate volumetric coordinates for the regions:

Additional changes:

Fixed a bug where selective_average did not use the skip= parameter
Made condition_frames more robust when events extend past the duration of the scan

Abhinav_Singh · July 24, 2020, 2:31pm

Hi!
I was exploring the visualization update according to my different correlation matrix and threshold values but after a few plots, I’m getting this warning that I have created more than 10 plots.
I want to know how to clear my colab RAM? I’m just plotting 1 plot and then after downloading it as png image I delete the cell and create a new plot with different data but still, I’m getting the warning.

/usr/local/lib/python3.6/dist-packages/nilearn/reporting/html_document.py:60: UserWarning: It seems you have created more than 10 nilearn views. As each view uses dozens of megabytes of RAM, you might want to delete some of them.
  MAX_IMG_VIEWS_BEFORE_WARNING))

mishkanemes · July 25, 2020, 1:49pm

In my group, we’re looking at the HCP data and we would like to do some modelling that bridges between the fMRI and the behavioural data. We are particularly interested in the Language task (story/math) and whether we can get hold of the trial by trial accuracy or difficulty for the questions that are asked after the stories or maths problems. We here define a trial as an individual run in each task (e.g. each story that the subjects listen to in language task -> story condition).

Could anyone point us where to find this, or whether this data exists at all? Many thanks

emeyer121 · July 27, 2020, 4:33pm

Hi, thanks for the updates! I have a question about the group_contrast[atlas[“labels_L”]] part of the surface plots. group_contrast has a length of 360 for each of the parcels, but atlas[“labels_L”] has a length of over 10,000? Can you provide some insight to how this indexing is working?

michaelwaskom · July 27, 2020, 5:40pm

The 3D visualization uses a mesh that has ~10K vertices. The atlas["labels_L"] maps from the indices in a parcel-wise array to a vertex-wise array.

When indexing into a numpy array with integer values, the size of the index array determines the size of the output. The size of the input determing the range of possible index values. Does that help?

raul.rodriguezcruces · July 29, 2020, 4:37pm

In my pod we prepared two ipynb with visualization of the Glasser parcellations over the surface, I couldn’t make them work on colab, but they work fine locally (if all the dependencies are install).
Could be useful for visualization.

github.com

rcruces/2020_NMA_surface-plot/blob/master/README.md

# NMA Glasser surface plot with BrainSpace.  

## Requiriments: BrainSpace  
### Python installation
[BrainSpace](https://brainspace.readthedocs.io/en/latest/index.html)  
> `pip install brainspace`. 
  
BrainSpace works on `Python 3.5+`, and probably with older versions of Python 3,
although it is not tested.

**Dependencies**   
To use BrainSpace, the following Python packages are required:

* [`numpy`](https://numpy.org/)
* [`scipy`](https://scipy.org/scipylib/index.html)
* [`scikit-learn`](https://scikit-learn.org/stable/)
* [`vtk`](https://vtk.org/)
* [`matplotlib`](https://matplotlib.org/)
* [`nibabel`](https://nipy.org/nibabel/index.html)
* [`nilearn`](https://nilearn.github.io/)

This file has been truncated. show original

yamin_bagheri · July 29, 2020, 9:25pm

hi
yes this data exists if i well understood your question, you want EVs of language task.
you can find that in AWS or compact subject data in a path like this
104416_3T_tfMRI_LANGUAGE_preproc//104416/MNINonLinear/Results/tfMRI_LANGUAGE_LR/EVs/