Segmentation Problem
Locked
 
|
Dear List
I got this Image of muscle fiber (Fiber-I white and Fiber-II red) from a user: http://www.picfront.org/d/91MZ She wants to compare white versus red cells. She needs the percentage of cells (white vs red) but only cells that have a circularity between 0.6 and 0.8 should be considered. Truncated cells should be excludes as well. I am able to seperate white and red cells but just area wise. I would be glad if anybody as a good procedure to segment the single cells (if possible at all). Thanks a lot for your help Pascal -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
|
|
I tried converting to HSV, media filtering, and segmenting each with simple threshold or find edges / variance and use logical AND to narrow to the white cells. Isolating the white cells from everything else is simple but I couldn't get the cells to separate from each other. If anybody else has a method, I'd be very interested to hear it too.
But if she has fewer than 1000 or so cells, I'd just trace them. Regards, Michael -----Original Message----- From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Pascal Lorentz Sent: Wednesday, October 02, 2013 7:54 AM To: [hidden email] Subject: Segmentation Problem Dear List I got this Image of muscle fiber (Fiber-I white and Fiber-II red) from a user: http://www.picfront.org/d/91MZ She wants to compare white versus red cells. She needs the percentage of cells (white vs red) but only cells that have a circularity between 0.6 and 0.8 should be considered. Truncated cells should be excludes as well. I am able to seperate white and red cells but just area wise. I would be glad if anybody as a good procedure to segment the single cells (if possible at all). Thanks a lot for your help Pascal -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
|
Hi Michael & Pascal,
Here is a potential approach that might work with some fine tuning: 1) Use Trainable Weka Segmentation to separate white cells from red cells from background. 2) From the white cells mask, clean up the mask as best as possible using morphological operations (e.g., Process > Binary > Open). 3) Since cells are blob-shaped (i.e., not too long and skinny), use Ultimate Points to compute cell centers from the Euclidean distance map. Here is a macro for steps 2 & 3: call("trainableSegmentation.Weka_Segmentation.getResult"); selectWindow("Classified image"); setAutoThreshold("Default dark"); setThreshold(0, 60); // threshold to white cells only run("Create Mask"); // create a mask of the white cells run("Open"); // clean up the mask a little run("Ultimate Points"); // compute segmentation centers from EDM setAutoThreshold("Default dark"); setThreshold(30, 255); // threshold to only centers of radius 30+ run("Create Selection"); selectWindow("Classified image"); close(); selectWindow("mask"); close(); selectWindow("2151833.png"); // switch back to the original image run("Restore Selection"); // show cell centers overlaid on original data It assumes you have opened the data (2151833.png), then run Trainable Weka Segmentation to perform the initial segmentation (see attached trainable-weka.jpg). My initial results are not perfect (see attached results.jpg) but not bad. Further fiddling of the TWS classifier might improve matters. And once you have a really good classifier, you can fully automate the process for additional images; for details see: http://fiji.sc/Trainable_Weka_Segmentation#Macro_language_compatibility Regards, Curtis On Thu, Oct 3, 2013 at 11:08 AM, Cammer, Michael <[hidden email] > wrote: > I tried converting to HSV, media filtering, and segmenting each with > simple threshold or find edges / variance and use logical AND to narrow to > the white cells. Isolating the white cells from everything else is simple > but I couldn't get the cells to separate from each other. If anybody else > has a method, I'd be very interested to hear it too. > > But if she has fewer than 1000 or so cells, I'd just trace them. > > Regards, > > Michael > > > > -----Original Message----- > From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of > Pascal Lorentz > Sent: Wednesday, October 02, 2013 7:54 AM > To: [hidden email] > Subject: Segmentation Problem > > Dear List > > I got this Image of muscle fiber (Fiber-I white and Fiber-II red) from a > user: > > http://www.picfront.org/d/91MZ > > She wants to compare white versus red cells. She needs the percentage of > cells (white vs red) but only cells that have a circularity between 0.6 and > 0.8 should be considered. Truncated cells should be excludes as well. I am > able to seperate white and red cells but just area wise. I would be glad if > anybody as a good procedure to segment the single cells (if possible at > all). > > Thanks a lot for your help > > Pascal > > -- > 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 |
|
|
Hi Pascal,
> Unfortunately I could not find mentioned attachments. I would be > curious to see your results. Would you mind sending them directly > again? Looks like the ImageJ list-serve strips many attachments. So I posted them at: http://curtis.imagej.net/2013-10-07/trainable-weka.jpg http://curtis.imagej.net/2013-10-07/results.jpg > I tried to follow the process and it seems to be really a good > approach. If I understood it correctly I will end up with the centre > of the cells. Right. > This is for sure enough to count them but I am not able to exclude > cells that don't fit the circularity criteria. Do you see a > possibility for that as well? Off the top of my head, I do not know the best way to reconcile the cell centers with the imperfect binary mask segmentation. To exclude based on circularity you would need to use Analyze Particles with a less messy binary mask. There may be a way to use the center points as the starting point for an operation that cleans up the mask, but I do not know what it is. Maybe someone else on this list has an idea. > Or is this already taken into account in the "ulitmate points" > function? I did not really understand what this function exactly does. The function is described here: http://imagej.net/docs/guide/146-29.html#sub:Ultimate-Points Sorry I cannot be of more assistance but I don't currently have the time to play with this analysis further right now. Regards, Curtis On Mon, Oct 7, 2013 at 5:00 AM, Pascal Lorentz <[hidden email]>wrote: > Hi Curtis > > Thanks you very much for your advice and even the macro code. > Unfortunately I could not find mentioned attachments. I would be curious > to see your results. Would you mind sending them directly again? > I tried to follow the process and it seems to be really a good approach. > If I understood it correctly I will end up with the centre of the cells. > This is for sure enough to count them but I am not able to exclude cells > that don't fit the circularity criteria. Do you see a possibility for that > as well? Or is this already taken into account in the "ulitmate points" > function? I did not really understand what this function exactly does. > > Best regards > > Pascal > > > Am 03.10.2013 19:07, schrieb Curtis Rueden: > > Hi Michael & Pascal, >> >> Here is a potential approach that might work with some fine tuning: >> >> 1) Use Trainable Weka Segmentation to separate white cells from red cells >> from background. >> >> 2) From the white cells mask, clean up the mask as best as possible using >> morphological operations (e.g., Process > Binary > Open). >> >> 3) Since cells are blob-shaped (i.e., not too long and skinny), use >> Ultimate Points to compute cell centers from the Euclidean distance map. >> >> Here is a macro for steps 2 & 3: >> >> call("trainableSegmentation.**Weka_Segmentation.getResult"); >> selectWindow("Classified image"); >> setAutoThreshold("Default dark"); >> setThreshold(0, 60); // threshold to white cells only >> run("Create Mask"); // create a mask of the white cells >> run("Open"); // clean up the mask a little >> run("Ultimate Points"); // compute segmentation centers from EDM >> setAutoThreshold("Default dark"); >> setThreshold(30, 255); // threshold to only centers of radius 30+ >> run("Create Selection"); >> selectWindow("Classified image"); >> close(); >> selectWindow("mask"); >> close(); >> selectWindow("2151833.png"); // switch back to the original image >> run("Restore Selection"); // show cell centers overlaid on original data >> >> It assumes you have opened the data (2151833.png), then run Trainable Weka >> Segmentation to perform the initial segmentation (see attached >> trainable-weka.jpg). >> >> My initial results are not perfect (see attached results.jpg) but not bad. >> Further fiddling of the TWS classifier might improve matters. And once you >> have a really good classifier, you can fully automate the process for >> additional images; for details see: >> http://fiji.sc/Trainable_Weka_**Segmentation#Macro_language_** >> compatibility<http://fiji.sc/Trainable_Weka_Segmentation#Macro_language_compatibility> >> >> Regards, >> Curtis >> >> >> On Thu, Oct 3, 2013 at 11:08 AM, Cammer, Michael < >> [hidden email] >> >>> wrote: >>> I tried converting to HSV, media filtering, and segmenting each with >>> simple threshold or find edges / variance and use logical AND to narrow >>> to >>> the white cells. Isolating the white cells from everything else is simple >>> but I couldn't get the cells to separate from each other. If anybody >>> else >>> has a method, I'd be very interested to hear it too. >>> >>> But if she has fewer than 1000 or so cells, I'd just trace them. >>> >>> Regards, >>> >>> Michael >>> >>> >>> >>> -----Original Message----- >>> From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of >>> Pascal Lorentz >>> Sent: Wednesday, October 02, 2013 7:54 AM >>> To: [hidden email] >>> Subject: Segmentation Problem >>> >>> Dear List >>> >>> I got this Image of muscle fiber (Fiber-I white and Fiber-II red) from a >>> user: >>> >>> http://www.picfront.org/d/91MZ >>> >>> She wants to compare white versus red cells. She needs the percentage of >>> cells (white vs red) but only cells that have a circularity between 0.6 >>> and >>> 0.8 should be considered. Truncated cells should be excludes as well. I >>> am >>> able to seperate white and red cells but just area wise. I would be glad >>> if >>> anybody as a good procedure to segment the single cells (if possible at >>> all). >>> >>> Thanks a lot for your help >>> >>> Pascal >>> >>> -- >>> ImageJ mailing list: http://imagej.nih.gov/ij/list.**html<http://imagej.nih.gov/ij/list.html> >>> >>> -- >>> ImageJ mailing list: http://imagej.nih.gov/ij/list.**html<http://imagej.nih.gov/ij/list.html> >>> >>> -- >> ImageJ mailing list: http://imagej.nih.gov/ij/list.**html<http://imagej.nih.gov/ij/list.html> >> > > -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
|
|
Dear Curtis, dear Michael
Thank you very much for your suggestions and your time that you spend on my problem. We are already pretty close and I will try to optimize the settings to improve the results. Thanks again Pascal Am 07.10.2013 17:18, schrieb Curtis Rueden: > Hi Pascal, > >> Unfortunately I could not find mentioned attachments. I would be >> curious to see your results. Would you mind sending them directly >> again? > Looks like the ImageJ list-serve strips many attachments. So I posted them > at: > > http://curtis.imagej.net/2013-10-07/trainable-weka.jpg > http://curtis.imagej.net/2013-10-07/results.jpg > >> I tried to follow the process and it seems to be really a good >> approach. If I understood it correctly I will end up with the centre >> of the cells. > Right. > >> This is for sure enough to count them but I am not able to exclude >> cells that don't fit the circularity criteria. Do you see a >> possibility for that as well? > Off the top of my head, I do not know the best way to reconcile the cell > centers with the imperfect binary mask segmentation. To exclude based on > circularity you would need to use Analyze Particles with a less messy > binary mask. There may be a way to use the center points as the starting > point for an operation that cleans up the mask, but I do not know what it > is. Maybe someone else on this list has an idea. > >> Or is this already taken into account in the "ulitmate points" >> function? I did not really understand what this function exactly does. > The function is described here: > http://imagej.net/docs/guide/146-29.html#sub:Ultimate-Points > > Sorry I cannot be of more assistance but I don't currently have the time to > play with this analysis further right now. > > Regards, > Curtis > > > On Mon, Oct 7, 2013 at 5:00 AM, Pascal Lorentz <[hidden email]>wrote: > >> Hi Curtis >> >> Thanks you very much for your advice and even the macro code. >> Unfortunately I could not find mentioned attachments. I would be curious >> to see your results. Would you mind sending them directly again? >> I tried to follow the process and it seems to be really a good approach. >> If I understood it correctly I will end up with the centre of the cells. >> This is for sure enough to count them but I am not able to exclude cells >> that don't fit the circularity criteria. Do you see a possibility for that >> as well? Or is this already taken into account in the "ulitmate points" >> function? I did not really understand what this function exactly does. >> >> Best regards >> >> Pascal >> >> >> Am 03.10.2013 19:07, schrieb Curtis Rueden: >> >> Hi Michael & Pascal, >>> Here is a potential approach that might work with some fine tuning: >>> >>> 1) Use Trainable Weka Segmentation to separate white cells from red cells >>> from background. >>> >>> 2) From the white cells mask, clean up the mask as best as possible using >>> morphological operations (e.g., Process > Binary > Open). >>> >>> 3) Since cells are blob-shaped (i.e., not too long and skinny), use >>> Ultimate Points to compute cell centers from the Euclidean distance map. >>> >>> Here is a macro for steps 2 & 3: >>> >>> call("trainableSegmentation.**Weka_Segmentation.getResult"); >>> selectWindow("Classified image"); >>> setAutoThreshold("Default dark"); >>> setThreshold(0, 60); // threshold to white cells only >>> run("Create Mask"); // create a mask of the white cells >>> run("Open"); // clean up the mask a little >>> run("Ultimate Points"); // compute segmentation centers from EDM >>> setAutoThreshold("Default dark"); >>> setThreshold(30, 255); // threshold to only centers of radius 30+ >>> run("Create Selection"); >>> selectWindow("Classified image"); >>> close(); >>> selectWindow("mask"); >>> close(); >>> selectWindow("2151833.png"); // switch back to the original image >>> run("Restore Selection"); // show cell centers overlaid on original data >>> >>> It assumes you have opened the data (2151833.png), then run Trainable Weka >>> Segmentation to perform the initial segmentation (see attached >>> trainable-weka.jpg). >>> >>> My initial results are not perfect (see attached results.jpg) but not bad. >>> Further fiddling of the TWS classifier might improve matters. And once you >>> have a really good classifier, you can fully automate the process for >>> additional images; for details see: >>> http://fiji.sc/Trainable_Weka_**Segmentation#Macro_language_** >>> compatibility<http://fiji.sc/Trainable_Weka_Segmentation#Macro_language_compatibility> >>> >>> Regards, >>> Curtis >>> >>> >>> On Thu, Oct 3, 2013 at 11:08 AM, Cammer, Michael < >>> [hidden email] >>> >>>> wrote: >>>> I tried converting to HSV, media filtering, and segmenting each with >>>> simple threshold or find edges / variance and use logical AND to narrow >>>> to >>>> the white cells. Isolating the white cells from everything else is simple >>>> but I couldn't get the cells to separate from each other. If anybody >>>> else >>>> has a method, I'd be very interested to hear it too. >>>> >>>> But if she has fewer than 1000 or so cells, I'd just trace them. >>>> >>>> Regards, >>>> >>>> Michael >>>> >>>> >>>> >>>> -----Original Message----- >>>> From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of >>>> Pascal Lorentz >>>> Sent: Wednesday, October 02, 2013 7:54 AM >>>> To: [hidden email] >>>> Subject: Segmentation Problem >>>> >>>> Dear List >>>> >>>> I got this Image of muscle fiber (Fiber-I white and Fiber-II red) from a >>>> user: >>>> >>>> http://www.picfront.org/d/91MZ >>>> >>>> She wants to compare white versus red cells. She needs the percentage of >>>> cells (white vs red) but only cells that have a circularity between 0.6 >>>> and >>>> 0.8 should be considered. Truncated cells should be excludes as well. I >>>> am >>>> able to seperate white and red cells but just area wise. I would be glad >>>> if >>>> anybody as a good procedure to segment the single cells (if possible at >>>> all). >>>> >>>> Thanks a lot for your help >>>> >>>> Pascal >>>> >>>> -- >>>> ImageJ mailing list: http://imagej.nih.gov/ij/list.**html<http://imagej.nih.gov/ij/list.html> >>>> >>>> -- >>>> ImageJ mailing list: http://imagej.nih.gov/ij/list.**html<http://imagej.nih.gov/ij/list.html> >>>> >>>> -- >>> ImageJ mailing list: http://imagej.nih.gov/ij/list.**html<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 |
«
Return to ImageJ
|
1 view|%1 views
| Free forum by Nabble | Edit this page |