general-discussion > new template for region growing process
Showing 1-2 of 2 posts
Aug 10, 2016 09:08 PM | Kangjoo Lee
new template for region growing process
Dear Pierre,
I'm wondering if NIAK is planning to include the new brain parcellation template in Glasser et al., (2016, Nature) for the region growing process, on behalf of or as an another option for the AAL template. Thanks!
Best,
Kangjoo
I'm wondering if NIAK is planning to include the new brain parcellation template in Glasser et al., (2016, Nature) for the region growing process, on behalf of or as an another option for the AAL template. Thanks!
Best,
Kangjoo
Aug 11, 2016 01:08 AM | Pierre Bellec
RE: new template for region growing process
Hi Kangjoo,
Would be happy to investigate the possibility, however I have a couple of reservations:
1. (and foremost) this parcellation has little impact on end results. BASC is applied on the full brain so there is no reason why two regions from distinct AAL regions wouldn't merge latter. That is assuming you are applying BASC or a similar algorithm on top of the region growing, of course.
2. The parcels need to be adapted to the registration procedure of niak. By default niak uses a symmetric space, which is not standard. The original AAL parcellation has been modified before inclusion in niak (see detailed explanations here http://niak.simexp-lab.org/pipe_region_growing.html). The same type of modifications would need to be applied to an alternative parcellation.
3. We are currently working on a replacement of the region growing pipeline which implement bootstrap aggregation of many parcellations into regions, generated from subsets of the data "a la BASC", while working at the voxel level. And using little memory, regardless of the number of time points and subjects. And also running fast. So soon, the question of the prior parcellation will become irrelevant.
Anyway, if you want to try out this new parcellation in the region growing, you can always manually specify it in `files_in.areas`, see the doc http://niak.simexp-lab.org/pipe_region_growing.html#inputs
Best regards,
Pierre Bellec
Would be happy to investigate the possibility, however I have a couple of reservations:
1. (and foremost) this parcellation has little impact on end results. BASC is applied on the full brain so there is no reason why two regions from distinct AAL regions wouldn't merge latter. That is assuming you are applying BASC or a similar algorithm on top of the region growing, of course.
2. The parcels need to be adapted to the registration procedure of niak. By default niak uses a symmetric space, which is not standard. The original AAL parcellation has been modified before inclusion in niak (see detailed explanations here http://niak.simexp-lab.org/pipe_region_growing.html). The same type of modifications would need to be applied to an alternative parcellation.
3. We are currently working on a replacement of the region growing pipeline which implement bootstrap aggregation of many parcellations into regions, generated from subsets of the data "a la BASC", while working at the voxel level. And using little memory, regardless of the number of time points and subjects. And also running fast. So soon, the question of the prior parcellation will become irrelevant.
Anyway, if you want to try out this new parcellation in the region growing, you can always manually specify it in `files_in.areas`, see the doc http://niak.simexp-lab.org/pipe_region_growing.html#inputs
Best regards,
Pierre Bellec