Hi David and Mark,
for fully automated lesion segmentation what has proven very reliable to
our opinion (in MS patients, and also for normal aging) is the
EMS algorithm. It's core algorithm has been published. Also see:
http://www.fil.ion.ucl.ac.uk/spm/ext/#EMS
In brief, its input is multimodal, e. g. FLAIR, T2 and T1 (which proved
well here, but you may used PD instead of T2, for example).
The nice thing, it works "outlier based", so only, if voxels carry certain
outlier features, it will allow them to enter the "lesion" class - this is
important as otherwise you get lots of false positive lesion voxels.
The output is 6 physiological classes and one lesion class (all probability
maps), it is then up to you how conservatively you threshold those.
Attached is one example with VERY conservative thresholding - note that the
lesion class was subdivided into pure T2- and T1-lesions (post-hoc, based
on T1-values) here, usually you'd get this class as one lesion class.
Overall, very good experiences with the algorithm.
I should that you need OPTIMAL coregistration and fair spatial
normalisation to MNI in order to make best of the EMS technique,
as some steps are atlas based. So, best you optimize these steps outside
the tool (my experience) and start off with the "lesions"
step.
Hope this helps,
Philipp
At 23:58 21.09.2011 +0100, Mark Jenkinson wrote:
>Dear David,
>
>I'm afraid that there is no automatic, robust and reliable way of doing
this with
>publicly available tools at present (as far as I know). You could try
looking at
>the ratio of the PD and T1 image, as this can often highlight changes, but I
>think you'll have to go with a manual or semi-automatic method of delineating
>the hyperintensities (for example, a combination of thresholding, tissue
>segmentation, and manual editing might work).
>
>All the best,
> Mark
>
>
>On 21 Sep 2011, at 20:04, David Lahna wrote:
>
>> Can anyone suggest a tool or workflow pipeline I could use to
automatically segment white matter hyperintensities?
>>
>> I have T1, T2 and proton density scans to work with but FLAIR images are
unfortunately not available.
>>
>> Any ideas are appreciated. Thanks in advance.
>>
>
>
Max Planck Institute of Psychiatry
NMR Research Group
Kraepelinstr. 2-10
80804 Munich
Mail: [log in to unmask]
Phone: 0049-89-30622-413
|