After applying my algorithm,I got a segmented image which contains small unwanted regions.I just need the vetebrae and disc (portion with blue in the middle).Can you suggest any method to eliminate it.Connected Component analysis returns a single object.Image is also attached
Median filtering is one way of removing such small objects. The second way: identify area of each individual region using regionprops fucntion in MATLAB and eliminate such small objects.
Opening removes small objects, while closing removes small holes.
Can you tell me more about your data?
If you have a CT scan available, then treating it as a 3D problem makes the segmentation much simpler. I mean, the dorsal spikes of the vertebrae are connected in 3D, but not (or not all) in 2D. If you indeed have 3D CT data, thresholding half-way between bone and water is usually good enough.
I disagree with all of the above that suggest morphological operations.
Opening followed by closing will not get you anywhere, because the contour of what you want to segment is as thick as the junk you want to remove. Opening(s) followed by closing(s) only work when the object of interest is thicker than the rest!!
Depends on how far you want to go and what additional information you might have. I agree with Lambert Zijp that morphology is of no use here. For a fully automatic algorithm that would work reliably on high volumes of data you'd probably want to use some kind of template that you match to the vertebrae and then once you have prior probabilities at each voxel, connected component analysis or segmenting based on geometry features (or rather a combination thereof) can get you further down the road.
1- use bucket fill algorithm to group all the segmented areas.
2- the vertebrae and disc are connected and they will form a large group
3- remove the smallest groups from the image leaving the largest group(s).
as shown in the image the green and pink are the largest groups this include the spine area.
you may design a special erosion filter that remove all pixels only if they are connected and surrounded by let us say 10x10 connected pixels
Ahmad, I don't see how that could work. OK, eliminating small structures, but that was not really the problem. If you use bucket fill, and than apply your erosion, the net result of those two operations is: nothing happened! The vertebrae column is still attached to unwanted junk...
Lambert, the question was "How do I get rid of small regions in an image"
You are right, that is what was asked. But it is not what was meant. :-)
Lambert, you made me challenge this question, despite the heavy work am doing I will solve it, perhaps one of my students would help
here spinal cord is having high intensity.is it necessary to remove spinal cord before proceeding intensity based segmentation
Try to dilate the image until the blue part (disc) is connected to the required part (vertebrae). And then apply connected component algorithm(Blob coloring). I am not sure whether this works or not but from the image you attached it looks like this may work.
I would defintely recommend you visit www.MIPAR.us. It is perfectly suited for the kind discovering the multi-step process necessary to solve your problem.
It is very powerful and intuitive 2D/3D image analysis software written by scientists/end-users and is about to be released as a free trial on December 14th. You can sign up on the site as well as submit images/datasets to test.
There are many image analysis options available, but I think you may just find MIPAR to be something special if you give it a try!
- Badges
- Science topic