NIfTI: — Neuroimaging Informatics Technology Initiative

phorum - NIfTI Message Board - Re: Size limitations in NIfTI

NIfTI Message Board

New Topic | Go to Top | Go to Topic | Search

Re: Size limitations in NIfTI

Author: Davide (134.107.216.---)
Date: 02-17-09 09:37

It is a practical question, I�m working with such kind of data, and I cannot find a straightforward solution to this problem that is practical, computation and storage efficient and compatible with nifti.
Till now we are working with vista data, but for compatibility issues we are porting the most part of our SW to support nifti.
I�m working with tractography based clustering of the brain.
we compute thousands of tractograms of one brain and for a number of reasons, mostly due to memory occupation and computational speed, we store them in a "compact format". In a "compact format" we have a subset of all image voxels stored in a single mono-dimensional vector. A 3D volume is made by some million of "empty" voxels and about one million of informative voxels, thus we mask out the "background" using an irregular mask (tipically white matter mask) and save the voxels of interest as a vector.
The following step is for instance the computation of the correlation or mutual information matrix, which requires to confront each tractogram with all others.
Using a compact representation this task is solved in 2 hours to a couple of days depending on the size of the dataset. For big datasets which does not fit into the memory of the workstation we have to repeatedly load/unload some tractogram. Hence a representation standard that allows to reduce the amount of data to fetch from teh disk and avoid on-the-fly masking is able to reduce the computational cost by a good factor.

Cheers.
Davide.

Reply To This Message | Flat View

Newer Topic | Older Topic

Topics