[Insight-users] automatic segmentation of bone in legs: which filter most suitable?

[Andriy Fedorov]

Mark,

Depending on your data, you might want to apply bias field correction
prior to using region growing methods on the whole MRI volume. I saw
cases where this makes a big difference in segmentation performance.
You can use this IJ contribution to suppress bias in your data:
http://www.insight-journal.org/browse/publication/640

Andriy Fedorov


> Date: Thu, 14 Jan 2010 10:59:24 +0000
> From: michiel mentink <michael.mentink at st-hughs.ox.ac.uk>
> Subject: Re: [Insight-users] automatic segmentation of bone in legs:
>        which   filter most suitable?
> To: marco giordano <marco.giord at gmail.com>
> Cc: ITK <insight-users at itk.org>
> Message-ID:
>        <5488ad6a1001140259p1f611741we4c83169f1daf460 at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
>
> I was playing around yesterday with the slice I posted to this thread.
>
> When placing one single seed, you can find the bone rather well, but there
> is a big risk that the region spills over into the tissue surrounding the
> bone.
> I haven't experimented yet with placing multiple seeds.
>
> If you look at the picture, you can see that the area surrounding the
> cartilage is delimiting the bone very well, but the side of the bone does
> not produce
> sharp edges in the edge detect filter bit of this segmentation algorithm.
>
> Also, I only tried one slice, and actually, I'd like to segment a complete
> 3D volume of the bone. That means I have to place multiple seeds along the
> bone and figure out some set of parameters that produces good results for
> every slice.
>
> I'm a bit weary about this type of segmentation but I'll continue
> experimenting with it. Please let me know if you have
> hints/tips/suggestions/experiences
> regarding this.
>
> Michael
>
>
>
> On Wed, Jan 13, 2010 at 4:30 PM, marco giordano <marco.giord at gmail.com>wrote:
>
>> Hi michiel,
>>
>> I am also involved in solving a similar problem so I am interested in any
>> possible solution.
>>
>> I did not look at the values but it seems that the bones are well delimited
>> and also quite homogeneous.
>> If you can specify a seed easily (e.g by looking at the histogram) I would
>> try with region growing first. The problem might be that the segmented
>> region overgrows if the borders are not sharp.
>>
>> Also watersheds should work but It will give you a complete segmentation,
>> then you should select the region by yourself.
>>
>> Level sets may also be an option but it might be too much for this case.
>>
>> I would be curious to see the results, please share any experience.
>>
>> Regards
>>
>>
>> 2010/1/12 michiel mentink <michael.mentink at st-hughs.ox.ac.uk>
>>
>>
>>>>> The images are made with MRI.
>>>>> You can see the two bones in the middle.
>>>>>
>>>>> Michael
>>>>>
>>>>>
>>>>>
>>>>> On Tue, Jan 12, 2010 at 3:10 PM, siqi chen <siqichensc at gmail.com>wrote:
>>>>>
>>>>>>
>>>>>> If you can post one sample image, that would be helpful to determine
>>>>>> which filter is the appropriate one.
>>>>>>
>>>>>> Siqi
>>>>>>
>>>>>> On Tue, Jan 12, 2010 at 10:08 AM, michiel mentink <
>>>>>> michael.mentink at st-hughs.ox.ac.uk> wrote:
>>>>>>
>>>>>>>
>>>>>>> Does anyone have experience with or know which segmentation algorithm
>>>>>>> can be used
>>>>>>> best for segmenting bones in a knee?
>>>>>>>
>>>>>>> There are about 100 pages of segmentation algorithms in the ITK
>>>>>>> software guide..
>>>>>>>
>>>>>>> I have a DICOM volume I want to segment into a binary (labeled) volume
>>>>>>> and ultimately
>>>>>>> convert to a surface mesh of the individual bones.
>>>>>>>
>>>>>>> thanks, Michael