Hi Mark,
I just tried athe bias field correction included in Fast and used the
resored image as input in Flirt, which gave me a much better result.
Thanks a lot for the help,
wolf
Mark Jenkinson wrote:
> Hi wolf,
>
> Without seeing the data it is difficult to be precise, but in general,
> surface coil data is difficult to register because of the large bias
> field effects in it. The best remedy for this (if you can't remove
> the bias field by some form of calibration) is to weight the surface
> coil image such that regions of high signal have a near unity
> weight and regions where the signal is low receive near zero weight.
> You'll probably need to define this by hand, and to start with a simple
> binary mask, drawn in fslview might suffice.
>
> As for scaling and mcflirt - if you have a whole brain volume then
> mcflirt should work well regardless of the input resolution. However,
> if you are dealing with a small FOV, then it may have problems still
> (as it works at 8mm and 4mm resolutions) and "rescaling" the voxel
> dimensions by an appropriate factor may work better. Alternatively,
> you can script flirt up to do the same job if that's easier.
>
> All the best,
> Mark
>
>
> wolf zinke wrote:
>
>> Hi,
>>
>> I was happy to read that rescaling is not an issue anymore for flirt.
>> Is it the same for mcflirt, or does this function still require data
>> with the right scales.
>>
>> I was trying now to apply flirt to my data and wanted to coregister a
>> coronal set of slices acquired with a surface coil to a full brain
>> scan, both T1 weighted sequences. The results are not satisfying me.
>> I was playing around with the cost functions and the DOF, but got no
>> acceptable results. Are there any suggestions, how to achieve a good
>> coregistration.
>>
>> thanks,
>>
>> wolf
>
|