 |
 |
Dear Mark, This is very helpful -- thank you! I'll try to implement your suggestions and see if I have any further questions. Best, David On Tue, Sep 17, 2013 at 1:42 AM, Mark Jenkinson < [log in to unmask]> wrote: > Dear David, > > The -kernel file option is the right one for you, but this only selects > the kernel - you still need to apply the kernel in a convolution operation. > From your description I think the -fmeanu option would be right for you > (add this after the "-kernel file weight_array" options). > > The weight_array file needs to be a NIFTI image file, which in your case > should be a 3D image that has 5x5x5 voxels, containing the intensities as > you've specified them. The voxel dimensions (in mm) will be ignored when > loading a kernel, as it will consider them to have the same voxel > dimensions as the input image. > > You can fairly easily make the weight_array image with fslcreatehd and > fslview (to manually set the voxel intensities) or by using matlab (with > save_avw from $FSLDIR/etc/matlab, or any other equivalent nifti image > saving routine). > > All the best, > Mark > > > > On 16 Sep 2013, at 23:14, David Romano <[log in to unmask]> wrote: > > > Hi everyone, > > > > Suppose I have a 3D image in a file called brain.nii.gz, and that I'd > like to produce a new .nii.gz file by replacing -- at each voxel -- the > weighted average of all the voxels in a 5x5x5 cube centered at that voxel, > where the central voxel is given a weight of 3, the remaining voxels of > the inner 3x3x3 cube are given weights of 2, and the outer voxels of the > 5x5x5 cube are given a weight of 1. > > > > I'm used to thinking of this as the result of 1) convolving the original > image with the 5x5x5 array of weights, and then 2) selecting those entries > of the resulting array that correspond to the original image array, but I'm > not sure how to implement this in fslmaths (if it's even possible). I can > see that it's possible to specify "kernels", which I think corresponds to > my array of weights, so maybe I could specify the convolution with the > command: > > > > fslmaths brain -kernel file weight_array convolved_image > > > > but I'm not sure what kind of file format should be used for > 'weight_array' or the dimensions of the output file convolved_image.nii.gz > will be, or its voxels' relationships to the those of the original > brain.nii.gz image. > > > > I'd be grateful for any advice or pointers to relevant documentation. > > > > Thanks in advance for your help! > > > > David Romano >