Dear Michael,
I am afraid Jesper is much too old and set in his ways to be convinced
of anything these days.
Having said that the non-linear fitting is already considerably less
affected by outliers by virtue of not log-transforming the data as is
done for the linear fit. Most outliers in DTI are in the negative
direction, i.e. they are cause by loss of signal that causes the data
to have a smaller value than would be predicted by the model. An
implicit consequence of the log-transformation is that small values
are considered more reliable, which means that these values caused by
signal loss is given a higher weight in the fitting than are "good"
values. So even if there is no detection (and de-weighting) of
outliers there will at least be no up-weighting of them.
Jesper
On 15 May 2010, at 08:45, Michael Scheel wrote:
> Hi Saad, thanks for the reply. That sounds like good news.
> Wouldn't it be favourable to have an outlier detection?
> Maybe Jesper could be convinced ....
>
> cheers, michael
> ________________________________________
> Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag
> von Saad Jbabdi [[log in to unmask]]
> Gesendet: Freitag, 14. Mai 2010 03:58
> An: [log in to unmask]
> Betreff: Re: [FSL] dtifit using a non-linear fit - any plans
>
> Hi Michael,
>
> Jesper has promised a nonlinear fit in dtifit soon [including
> positivity constraint and Rician noise modelling]
> (it is already implemented, just needs to make it into dtifit!).
> However this will likely _not_ include automatic outlier detection.
>
> Cheers,
> Saad.
>
>
> On 13 May 2010, at 20:09, Michael Scheel wrote:
>
>> Dear Saad and other fsl experts,
>>
>> are there any plans to implement a non-linear fit approach
>> including outlier detection as described by Chang, Jones and
>> Pierpaoli in the RESTORE algorithm.
>> I found their paper very convincing and yet would like to stay
>> within the fsl framework.
>>
>> Thanks, Michael
>>
>
> --
> Saad Jbabdi
> University of Oxford, FMRIB Centre
>
> JR Hospital, Headington, OX3 9DU, UK
> (+44)1865-222466 (fax 717)
> www.fmrib.ox.ac.uk/~saad
>
|