Q:To whom it may concern, Good morning. My name is Fidel, and I am a Spanish computer scientist doing a PhD in Artificial Vision. At the moment, I am trying to reproduce some algorithms regarding the segmentation of brain tumors and I want to use some of the brain models you give in your web http://mouldy.bic.mni.mcgill.ca/brainweb/anatomic_normal_20.html.

I am trying to create a brain atlas using your brain models, but I have a problem: I want to download the T1 simulation of a brain model and I want to use it as a template, using the ‘CRISP’ model to know the tissue classification of each pixel. Nevertheless, the T1 simulation and the crisp model have different dimensions ( [181,256,256] & [362,434,362] ),being this way impossible to “adjust” one to the other, as the axis have not the same proportion.

Why are you creating the simulations this way? Do you have T1 simulation with the same dimensions as the rest of the brain tissue models? Do you know an easy way to adjust the T1 simulation to the rest of the files.

Thanks in advance for your hard and useful work!

A:The reason why the simulation model has higher resolution than the resulting simulated MRI, is to allow the simulations to contain partial volume effects - as you would encounter in an actual MRI. In other words, this was done to increase the realism of the simulations.

However, the images live in the same space, meaning that the world coordinate spaces of the model and the simulation line up. This in turn means that you can simply resample the model to the sampling space of the simulaton (or vice versa) using something like:

  mincresample -nearest -like <T1.mnc> <crisp.mnc> <crisp_like_T1.mnc>

Note that when resampling a label (discrete) volume you will need to use nearest neighbor “interpolation” (-nearest), otherwise you will be left with non-integer label values. A better way to do this is to use resample_labels which goes about it somewhat more intelligently and will result in a less voxellated resampled label volume.