vasculature (looking for current state of the art)
Locked
vasculature (looking for current state of the art)
 
|
I’m just starting a literature search. The topic is finding the branchng artery structure as seen in images of human retina - specifically Fundus AutoFlouresence (FAF) images as produced by instruments such as the Spectralis. I’m also interested in anything dealing with similar images (roughly centered on the fovea, with the optic disc just barely inside the field of view.
In these images, the blood vessels are very dark (except where they become very thin) and form a single branching tree emanating from the optic nerve head (which is similarly dark, but a large, roughly circular blob. There are issues involved in defining the edge of the ONH - but I’d be happy to ignore these and settle for many trees, each rooted at the ONH blob. I’d even be happy to manuallyh select the root of each tree and have it grown from there. I *might* be happy with a segmentation of “long skinny blood vessel-like structures” without any explicit branching structure. The ultimate goal is a binary mask (*this* is vasculature..and *that* is not). I anticipate a final, manual editing phase. We have a crude system working (using Phansalkar adaptive thresholding - with no information on the fact we’re looking for elongated features forming a branching structure). I’m trying to improve it (to reduce the amount of time it takes to manually edit). I get the feeling that *someone* has decent methods for this, but I’m unfamiliar with that literature. I don’t want to re-invent the wheel. Citations to papers would be good. Pointers to implementations (FIJI would be a huge bonus) are most appreciated. Common problems: imaging artifacts, especially at either the top or bottom of the image, and in the four corners. Less common (so far) problems: pathological cases with large dark patches that are *not* vasculature. On one end, simple adaptive thresholding does a pretty good job. On the other hand, I thinking of experimenting to find “long, thin, dark segments” and then trying to join them together. And finally…the manual editing phase now is primarily concerned with very small blood vessels which are “obviously” extensions of already found larger vessels. These consist of “mixed pixels”, and thresholding simply won’t work. On the other hand, we might be able to get away with *not* masking these. I’ve started to consider methods of extending the thicker vessels by growing the ends (constrained by direction, and the presence of a “U” shaped intensity cross-section through the putative vessel. All clues gratefully rented. — Kenneth Sloan [hidden email] -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
Re: vasculature (looking for current state of the art)
|
|
Dear Kenneth,
On 10/08/2016 20:37, Kenneth Sloan wrote: > I’m just starting a literature search. The topic is finding the branchng artery structure as seen in images of human retina - specifically Fundus AutoFlouresence (FAF) images as produced by instruments such as the Spectralis. I’m also interested in anything dealing with similar images (roughly centered on the fovea, with the optic disc just barely inside the field of view. > > In these images, the blood vessels are very dark (except where they become very thin) and form a single branching tree emanating from the optic nerve head (which is similarly dark, but a large, roughly circular blob. > > There are issues involved in defining the edge of the ONH - but I’d be happy to ignore these and settle for many trees, each rooted at the ONH blob. I’d even be happy to manuallyh select the root of each tree and have it grown from there. I *might* be happy with a segmentation of “long skinny blood vessel-like structures” without any explicit branching structure. > > The ultimate goal is a binary mask (*this* is vasculature..and *that* is not). > > I anticipate a final, manual editing phase. We have a crude system working (using Phansalkar adaptive thresholding - with no information on the fact we’re looking for elongated features forming a branching structure). I’m trying to improve it (to reduce the amount of time it takes to manually edit). I get the feeling that *someone* has decent methods for this, but I’m unfamiliar with that literature. I don’t want to re-invent the wheel. > > Citations to papers would be good. Pointers to implementations (FIJI would be a huge bonus) are most appreciated. > > Common problems: imaging artifacts, especially at either the top or bottom of the image, and in the four corners. > > Less common (so far) problems: pathological cases with large dark patches that are *not* vasculature. > > > On one end, simple adaptive thresholding does a pretty good job. On the other hand, I thinking of experimenting to find “long, thin, dark segments” and then trying to join them together. before thresholding? It helped enormously when I tried to analyse growth of neurons in cell culture. For that I used the following strategy, which I think should be applicable to your problem (for soma read optic nerve head and for neurites blood vessels): - apply tubeness filter to image - threshold image to create mask of all putative neurites - manually mark soma and create mask of soma - use a binary reconstruct to only select putative neurites that are connected to soma - any neurites that were missed due to breaks/gaps in image can easily be added in an editing step - the resulting mask can then be skeletonised and analysed further If you want, send me a sample image and I will run it through my macro tools. If it looks promising, I am very happy to share my macros. Best, Volko > > And finally…the manual editing phase now is primarily concerned with very small blood vessels which are “obviously” extensions of already found larger vessels. These consist of “mixed pixels”, and thresholding simply won’t work. On the other hand, we might be able to get away with *not* masking these. I’ve started to consider methods of extending the thicker vessels by growing the ends (constrained by direction, and the presence of a “U” shaped intensity cross-section through the putative vessel. > > All clues gratefully rented. > > — > Kenneth Sloan > [hidden email] > > > > > -- > ImageJ mailing list: http://imagej.nih.gov/ij/list.html -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
Re: vasculature (looking for current state of the art)
|
|
Not yet - I’ll look into it.
Thanks very much for the reply. -- Kenneth Sloan [hidden email] Vision is the art of seeing what is invisible to others. > On Aug 11, 2016, at 00:19 , Volko Straub <[hidden email]> wrote: > > Dear Kenneth, > > On 10/08/2016 20:37, Kenneth Sloan wrote: >> I’m just starting a literature search. The topic is finding the branchng artery structure as seen in images of human retina - specifically Fundus AutoFlouresence (FAF) images as produced by instruments such as the Spectralis. I’m also interested in anything dealing with similar images (roughly centered on the fovea, with the optic disc just barely inside the field of view. >> >> In these images, the blood vessels are very dark (except where they become very thin) and form a single branching tree emanating from the optic nerve head (which is similarly dark, but a large, roughly circular blob. >> >> There are issues involved in defining the edge of the ONH - but I’d be happy to ignore these and settle for many trees, each rooted at the ONH blob. I’d even be happy to manuallyh select the root of each tree and have it grown from there. I *might* be happy with a segmentation of “long skinny blood vessel-like structures” without any explicit branching structure. >> >> The ultimate goal is a binary mask (*this* is vasculature..and *that* is not). >> >> I anticipate a final, manual editing phase. We have a crude system working (using Phansalkar adaptive thresholding - with no information on the fact we’re looking for elongated features forming a branching structure). I’m trying to improve it (to reduce the amount of time it takes to manually edit). I get the feeling that *someone* has decent methods for this, but I’m unfamiliar with that literature. I don’t want to re-invent the wheel. >> >> Citations to papers would be good. Pointers to implementations (FIJI would be a huge bonus) are most appreciated. >> >> Common problems: imaging artifacts, especially at either the top or bottom of the image, and in the four corners. >> >> Less common (so far) problems: pathological cases with large dark patches that are *not* vasculature. >> >> On one end, simple adaptive thresholding does a pretty good job. On the other hand, I thinking of experimenting to find “long, thin, dark segments” and then trying to join them together. > Have you tried using tubeness filter (Analyse -> Tubeness) on the images before thresholding? It helped enormously when I tried to analyse growth of neurons in cell culture. For that I used the following strategy, which I think should be applicable to your problem (for soma read optic nerve head and for neurites blood vessels): > - apply tubeness filter to image > - threshold image to create mask of all putative neurites > - manually mark soma and create mask of soma > - use a binary reconstruct to only select putative neurites that are connected to soma > - any neurites that were missed due to breaks/gaps in image can easily be added in an editing step > - the resulting mask can then be skeletonised and analysed further > > If you want, send me a sample image and I will run it through my macro tools. If it looks promising, I am very happy to share my macros. > > Best, > Volko >> >> And finally…the manual editing phase now is primarily concerned with very small blood vessels which are “obviously” extensions of already found larger vessels. These consist of “mixed pixels”, and thresholding simply won’t work. On the other hand, we might be able to get away with *not* masking these. I’ve started to consider methods of extending the thicker vessels by growing the ends (constrained by direction, and the presence of a “U” shaped intensity cross-section through the putative vessel. >> >> All clues gratefully rented. >> >> — >> Kenneth Sloan >> [hidden email] >> >> >> >> >> -- >> ImageJ mailing list: http://imagej.nih.gov/ij/list.html > > -- > ImageJ mailing list: http://imagej.nih.gov/ij/list.html -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
Re: vasculature (looking for current state of the art)
|
|
In reply to this post by Kenneth Sloan-2
Hello Kenneth.
Based on tha same principle that Volko suggests you can look at these papers: M. E. Martinez-Perez, A. D. Hughes, S. A. Thom, A. A. Bharath and K. H. Parker. Segmentation of Blood Vessels from Red-free and Fluorescein Retinal Images. Medical Image Analysis. 11 (1): 47-61, 2007. Alun D. Hughes, M. Elena Martinez-Perez, Abu-Sufian Jabbar, Assif Hassan, Nick W. Witt, Paresh D. Mistry, Neil Chapman, Alice V. Stanton, Gareth Beevers, Toberto Pedrinelli, Kim H. Parker and Simon A. McG. Thom. Quantification of topological changes in retinal vascular architecture in essential and malignant hypertension. Journal of Hypertension, 24 (5): 889-894, 2006. cheers |
«
Return to ImageJ
|
1 view|%1 views
| Free forum by Nabble | Edit this page |