left ventricle segmentation

hello...can you please help me....

I'm working on a dicom image and i wanna extract the LV (left ventricular wall) so here is the steps i used after some work on that image : 1-edge detection using canny method

2-then 'dilate' the result edge image

3-'close' to fill the holes with "disk " structuring element

4-then 'dilate' after close

5-find the 'complement' (negative) of the dilated image

6-clear borders by 'imclearborder'.....

NOW, after all of that i've an image that contains a nearly circular two concentric circles that represent the left ventricular wall and other organs such as liver, they are all in white with a black background as shown in that link http://i1131.photobucket.com/albums/m541/Lomie55/LV_wall_image_mask.jpg all i want is that the final image only contain the LV white wall with a black background in other words a mask for the LV wall...can you please help me to know how to extract these two concentric circular part as you can see in the figure (link) above...anymore questions??...thanks in advance...

[Sean de Wolski]

Perhaps you could post the original image as well and describe what you want in it. Your description isn't clear in the binary image.

[Walter Roberson]

I do not perceive any concentric circular parts in the image indicated.

[Florin Neacsu]

Hi,

Also you might want to tell us which technology was used to acquire the projections yielding this image...MRI and PET/SPECT are quite different, so the segmentation technique can be different too.

Regards,

Florin

[Lamis]

hello....this is an MRI image and i'm working on t2 star assessment on myocardium (left ventricular wall)...so i need to segment the left ventricle...as i described above the steps i did....the original image is in that link < http://i1131.photobucket.com/albums/m541/Lomie55/t2star_rsquare_redcircle.jpg > the left ventricle or as i described it by two concentric circles is highlighted by the red circle in-order to be clearer...all what i want is to know how to segment this part (left ventricle)...and the binary image i posted earlier (in the original question) was the result of the steps i described also in the original question... if there is anymore questions just ask...thanks a lot...

[Lamis]

hello again...in that link < http://i1131.photobucket.com/albums/m541/Lomie55/LV_wall_image_mask_redcircle.jpg > the binary image and i highlighted the part i wanna segment by an orange circle...this is the part which i call two concentric white circles.....thanks

Answer 1

[Walter Roberson]

Looking at your original image, I don't know if it was necessary or useful to go through all those steps. I get the impression from the image that if you were to threshold at a sufficiently high grey level, and then imlabel() and regionprops(), that the target area would be the one with the lowest eccentricity.

[Lamis]

thank you so much...i'll try it...and tell you if i got the desired result

Answer 2

[Florin Neacsu]

Hi,

I suggest registering the images (I guess you have more than one), than crop the area where the left ventricle is generally situated ('a priori' information is always useful;a good registration is required, but you could have a wider cropping window), do an edge detection on you binary image and after that apply a Hough transformation. This is one approach, maybe some else here has a better one.

Regards, Florin

[Walter Roberson]

Hough transform did pass through my mind when I read the problem description, but when I looked at the images it appeared to me that there was enough connected material that would make the circular portion hard to find. I think that part might be difficult.

[Florin Neacsu]

Indeed, but once again, the fact that he know what he's looking for is extremely helpful. He can exclude a good part of cases just by looking for a specific range of radius, which would correspond to the ventricle. Of course, it's not completely automated nor 100% certain (and it will vary from patient to patient and even slice from slice, but that can integrated in the program) but it provides a good starting point.