TADPOLE nilearn Integration? (a brain atlas question)

[lbr...@mymail.mines.edu]

Hello,

First, thank you for providing the TADPOLE data-set/challenge to young researchers like myself =).

I wondered if anyone had any experience with the visualization framework provided by nilearn (http://nilearn.github.io/index.html) with regards to the TADPOLE data-set? I have some experience using this software for creating brain visualizations using standard brain atlases (AAL, Talairach, Harvard-Oxford, etc.) I'm just getting started with the TADPOLE data-set and would like to provide you with a general idea of my envisioned workflow:

1. Train a model (eg. linear multivariate regression with Lasso regularization) --> XW + b = y (X is TADPOLE data, W is coefficient matrix, y is RAVLT scores, etc.)

2. Take the sparse coefficient matrix W and collapse it into a single column. This collapsed column (derived from W) should provide me with information about which features in X are "important" for predicting RAVLT scores.

3. Then, using nilearn and an associated atlas (eg. http://nilearn.github.io/modules/generated/nilearn.datasets.fetch_atlas_harvard_oxford.html) I plot the ROIs (weighted by W) associated with the features provided X. This capability is obviously quite useful for visually communicating which parts of the brain are important to pay attention to.

Question: Is there a way for me to determine which brain atlas to use when plotting brain images when using the TADPOLE data-set? Has this mapping already been defined? Note: since the TADPOLE data-set is by definition multi-modal I would likely need multiple brain atlases to achieve this..

Furthermore, if this capability doesn't exist where would you suppose it fits inside the TADPOLE project? If I developed a mapping from the columns of X to a specific atlas, in addition to a nilearn integration, I would like to share it with the rest of the research community.

Thanks!

Lou

[Razvan Marinescu]

Hi Lou,

Sorry for the late reply, I forgot to get back on this. I don't personally have experience with nilearn, but it looks like a good set of tools for brain visualisation. 

ADNI MRI images have been pre-processed using Freesurfer, which gives parcellations for both the Desikan-Killiany and Destrieux atlases. See https://surfer.nmr.mgh.harvard.edu/fswiki/CorticalParcellation   For non-MRI images, I don't know exactly which atlas was used, but you can check the methods documents for each specific analysis (FDG, DTI, etc ..). 

Regarding sharing code or other tools, we encourage everyone to do that. You can post it on this forum. 

Regards,

Raz