NeuroQuery wrapper or algorithm

[tsalo]

Add a wrapper or even the algorithm for NeuroQuery.

References

Dockès, Jérôme, et al. "Text to brain: predicting the spatial distribution of neuroimaging observations from text reports." International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, Cham, 2018. https://doi.org/10.1007/978-3-030-00931-1_67

[choucavalier]

Note that a richer and more recent reference is: https://elifesciences.org/articles/53385 (NeuroQuery, comprehensive meta-analysis of human brain mapping). Also, anyone working on this might be interested in reading more details about NeuroQuery in @jeromedockes's thesis freely available here https://pastel.archives-ouvertes.fr/tel-02495783.

[jeromedockes]

thanks a lot @tgy ! @tsalo let me know if you are still planning to add neuroquery and if I can help

[tsalo]

Thank you both.

@jeromedockes I definitely want to add NeuroQuery, but it's still up in the air whether I should write a light wrapper around the NeuroQuery package or incorporate the code directly. Do you have a preference? If you plan to do long-term maintenance on your NeuroQuery Python package, the former is probably better, but if you ultimately want to move onto other stuff, then adding a version of the source code to a collaborative library (e.g., NiMARE) might be preferable. I remember at OHBM last year, Gael recommended using the source code, but I think that was before there was an actual library.

I also want to make a NiMARE Dataset using the NeuroQuery data at some point, but I was a little confused about which data actually went into the model on the website. I can follow up about that in neuroquery_data though.

[jeromedockes]

Taylor Salo [email protected] writes:

Thank you both.

@jeromedockes I definitely want to add NeuroQuery, but it's still up in the air whether I should write a light wrapper around the NeuroQuery package or incorporate the code directly. Do you have a preference?

I don't have a preference, I'll ask Gaël if he does

If you plan to do long-term maintenance on your NeuroQuery Python package, the former is probably better, but if you ultimately want to move onto other stuff, then adding a version of the source code to a collaborative library (e.g., NiMARE) might be preferable. I remember at OHBM last year, Gael recommended using the source code, but I think that was before there was an actual library.

I plan to maintain it but it is hard to guarantee for how long. If you add it as a dependency I can improve the continuous integration and add scheduled jobs with the versions of nibabel, numpy etc. that are supported by NiMARE. In any case I would also add links to NiMARE from the neuroquery repo and the website

I also want to make a NiMARE Dataset using the NeuroQuery data at some point, but I was a little confused about which data actually went into the model on the website. I can follow up about that in neuroquery_data though.

I need to improve the documentation :) it is the data in this folder: https://github.com/neuroquery/neuroquery_data/tree/master/neuroquery_model (the data downloaded by fetch_neuroquery_model) This example shows how it could be used to train a new model: https://github.com/neuroquery/neuroquery/blob/master/examples/training_neuroquery.ipynb

don't hesitate to ask more questions here or in a separate issue

[jeromedockes]

I just had a quick word with Gaël and I confirm that we plan to keep maintaining the neuroquery package.