 |
 |
The segmentation, bias correction and spatial normalisation are all combined within the same model of the data. I would therefore suggest just feeding the original data into the model. If you find that the bias correction is not doing enough, then change the regularisation settings. If there is no intensity nonuniformity artifact, then a model with heavy regularisation is better (as you are providing prior knowledge about there being little nonuniformity in the data). If there is lots of nonuniformity, then one with less regularisation may be preferable. Best regards, -John On 30 September 2010 11:45, Joćo Duarte <[log in to unmask]> wrote: > Thank you for you quick reply John. > > I understand. So you are saying that the segmentation procedure > includes itself a bias correction routine. In this case, I can use the > native space image in input, and the segmentation procedure will > automatically perform bias correction before writing the GM, WM and > CSF normalized images. In practice, the bias corrected image that is > written in the output of "Estimate and Writing" (despite being a > separated option from the GM, WM and CSF writing options in the > module) is the one used during the segmentation procedure to write GM, > WM and CSF, is it true? > > So what would you do, keep the native space image as input? > > Thank you very much. > > Regards, > > JD > > PS: The volumes are in ml, right? > > On Thu, Sep 30, 2010 at 11:36 AM, John Ashburner <[log in to unmask]> wrote: >> Different data will give different results. The fact that your >> results differ slightly is entirely expected. >> >> The difference comes from the fact that one image is nonuniformity >> corrected, which will have a slight influence on the initial affine >> registration, as well as on the bias correction part of the >> segmentation model. Bias correction is regularised, so the result of >> doing it twice will be different from doing it just the once. >> >> Best regards, >> -John >> >> 2010/9/30 Joćo Duarte <[log in to unmask]>: >>> Hello to all. >>> >>> Following my e-mail yesterday, I did the segmentation of one T1 image >>> using both the native space image and the bias corrected image. Then I >>> checked the GM and WM segmented images and the results are different >>> when using the native space or the bias corrected image. Also the >>> pxxx_seg8.txt have different values for GM, WM and CSF volumes (by the >>> way, are these values in ml?) when using the 2 different images in >>> input... Can someone help me to understand the difference, is this >>> normal? >>> >>> Thank you in advance >>> >>> JD >>> >> >