Re: [Gdcm2] Automatically categorizing and polygonizing scanned tissue
Cross-platform DICOM implementation
Brought to you by:
malat
From: Mark R. <mm...@gm...> - 2011-10-03 16:52:41
|
Hi John, Yes, a seed point is a starting point for a segmentation routine. Seed points can be chosen any number of ways (for instance, by a user clicking on a point internal to the organ), and then the algorithm takes the seed point and other parameters to define an organ. Mathieu is also right-- knowing how your data are acquired is essential. Segmentation routines for CT will be very different than for MR, projectional X-ray, ultrasound, etc. Knowing where to start, what body part to concentrate on, is a definite necessity. Mark On Mon, Oct 3, 2011 at 9:11 AM, John Dexter <jdx...@go...> wrote: > I'm aware of ITK but haven't used it or looked in detail. Query - "seed > points" sounds like the kind of thing I described as telling it specific > points belong to specific tissue/organs? > > On 3 October 2011 17:05, Mark Roden <mm...@gm...> wrote: >> >> I think the best place to get started is ITK: >> >> http://www.itk.org/ >> >> There are a number of different segmentation routines in the toolkit >> (and gdcm is used to read in DICOM files there). Some tasks are >> considered relatively easy (lung field segmentation, bone segmentation >> if you don't want the marrow), some start with seed points, and so >> forth. >> >> Good luck! >> Mark >> >> On Mon, Oct 3, 2011 at 7:41 AM, John Dexter <jdx...@go...> >> wrote: >> > Presumably there are particular problem areas, and other areas which are >> > much easier to understand? e.g detecting the skeleton compared to a scan >> > of >> > the abdomen, etc? My focus is not a full-body categorization, but >> > picking >> > out things like the heart, esophagus and large veins/arteries. >> > If you're able to give the system some clues, does that help? e.g load >> > the >> > scan, click on one voxel and say "this is the heart" so it can then >> > trace >> > outwards from that point? >> > >> > On 3 October 2011 15:22, Mark Roden <mm...@gm...> wrote: >> >> >> >> Polygonization is a more or less solved problem; check up on Marching >> >> Cubes and Marching Squares. >> >> >> >> The problem is determining where organs begin and end. Human experts >> >> can (and do) disagree on where organs are in these scans; Warfield >> >> developed an algorithm to try to resolve the differences in human >> >> raters: http://crl.med.harvard.edu/software/STAPLE/index.php >> >> >> >> Since humans disagree, it's very hard to get computers to come up with >> >> reasonable estimates that everyone can agree upon. >> >> >> >> On Mon, Oct 3, 2011 at 7:16 AM, John Dexter >> >> <jdx...@go...> >> >> wrote: >> >> > Is it the polygonization, or the tissue/organ categorization, which >> >> > is >> >> > the >> >> > big problem? My guess is the latter? >> >> > >> >> > On 3 October 2011 14:34, Mark Roden <mm...@gm...> wrote: >> >> >> >> >> >> This is _extremely_ hard to do, and worthy of a couple of doctoral >> >> >> dissertations. >> >> >> >> >> >> Take a look at the MICCAI segmentation challenges (Just google for >> >> >> 'MICCAI grand segmentation challenge') to see the state of the art >> >> >> in >> >> >> segmenting various organs from various modalities. >> >> >> >> >> >> >> >> >> On Mon, Oct 3, 2011 at 6:30 AM, John Dexter >> >> >> <jdx...@go...> >> >> >> wrote: >> >> >> > If I receive say a CT/MRI patient scan, how easy is it to >> >> >> > automatically >> >> >> > identify which is the heart, skeleton, etc? And how easy is it to >> >> >> > automatically build polygonal representations from volume data? >> >> >> > The >> >> >> > latter >> >> >> > part I think one of the other toolkits might solve, but could you >> >> >> > ever >> >> >> > trust >> >> >> > a completely automated solution or would it always involve manual >> >> >> > review/optimization? >> >> >> > >> >> >> > >> >> >> > >> >> >> > ------------------------------------------------------------------------------ >> >> >> > All the data continuously generated in your IT infrastructure >> >> >> > contains a >> >> >> > definitive record of customers, application performance, security >> >> >> > threats, fraudulent activity and more. Splunk takes this data and >> >> >> > makes >> >> >> > sense of it. Business sense. IT sense. Common sense. >> >> >> > http://p.sf.net/sfu/splunk-d2dcopy1 >> >> >> > _______________________________________________ >> >> >> > Gdcm-developers mailing list >> >> >> > Gdc...@li... >> >> >> > https://lists.sourceforge.net/lists/listinfo/gdcm-developers >> >> >> > >> >> >> > >> >> > >> >> > >> >> > >> >> > >> >> > ------------------------------------------------------------------------------ >> >> > All the data continuously generated in your IT infrastructure >> >> > contains a >> >> > definitive record of customers, application performance, security >> >> > threats, fraudulent activity and more. Splunk takes this data and >> >> > makes >> >> > sense of it. Business sense. IT sense. Common sense. >> >> > http://p.sf.net/sfu/splunk-d2dcopy1 >> >> > _______________________________________________ >> >> > Gdcm-developers mailing list >> >> > Gdc...@li... >> >> > https://lists.sourceforge.net/lists/listinfo/gdcm-developers >> >> > >> >> > >> > >> > >> > >> > ------------------------------------------------------------------------------ >> > All the data continuously generated in your IT infrastructure contains a >> > definitive record of customers, application performance, security >> > threats, fraudulent activity and more. Splunk takes this data and makes >> > sense of it. Business sense. IT sense. Common sense. >> > http://p.sf.net/sfu/splunk-d2dcopy1 >> > _______________________________________________ >> > Gdcm-developers mailing list >> > Gdc...@li... >> > https://lists.sourceforge.net/lists/listinfo/gdcm-developers >> > >> > > > > ------------------------------------------------------------------------------ > All the data continuously generated in your IT infrastructure contains a > definitive record of customers, application performance, security > threats, fraudulent activity and more. Splunk takes this data and makes > sense of it. Business sense. IT sense. Common sense. > http://p.sf.net/sfu/splunk-d2dcopy1 > _______________________________________________ > Gdcm-developers mailing list > Gdc...@li... > https://lists.sourceforge.net/lists/listinfo/gdcm-developers > > |