info
Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
Digital Pathology
166 views
Skip to first unread message
Bohsen
Nov 1, 2022, 9:56:52 AM11/1/22
to
Hi
We are considering acquiring a digital pathology system and would like to use DICOM in the solution. But there is concern about whether the suppliers can ensure the full functionality of their solutions if we demand DICOM.
I've looked into the Orthanc project and tested their Dicomizer on various proprietary file formats with mixed success. Also I haven't been able to find other projects/open source tools that specifically touch on Digital Pathology and I am therefore worried if we could be out too early.
I am also particularly concerned about if DICOM WSIs are ready for use in relation to GSPS objects. I consider this to be particularly important as we aim to support AI in our solution.
Does anyone here in the forum have experience or perspectives on it?
Kind regards,
Thomas
We are considering acquiring a digital pathology system and would like to use DICOM in the solution. But there is concern about whether the suppliers can ensure the full functionality of their solutions if we demand DICOM.
I've looked into the Orthanc project and tested their Dicomizer on various proprietary file formats with mixed success. Also I haven't been able to find other projects/open source tools that specifically touch on Digital Pathology and I am therefore worried if we could be out too early.
I am also particularly concerned about if DICOM WSIs are ready for use in relation to GSPS objects. I consider this to be particularly important as we aim to support AI in our solution.
Does anyone here in the forum have experience or perspectives on it?
Kind regards,
Thomas
sojanmat...@gmail.com
Nov 1, 2022, 10:42:54 PM11/1/22
to
If you are looking for a developer to build this DICOM solution, please let me know, I am interested.
Here is my email - sojanmathewshaan at gmail . com
BR,
Mathew
herman o.
Nov 2, 2022, 3:47:59 AM11/2/22
to
On Tuesday, November 1, 2022 at 9:42:54 PM UTC-5, sojanmat...@gmail.com wrote:
> >
> > Does anyone here in the forum have experience or perspectives on it?
> >
Here is my perspective:
> >
> > Does anyone here in the forum have experience or perspectives on it?
> >
1. The US is many years behind with deploying digital path compared with other countries, notably European. For example, in the Netherlands, exchanging digital path images is common and some hospitals have been completely digital in path for 5+ years if not longer. I believe that there are 2 reasons for the slow adoption in the US, first the FDA lagging approvals and second the lack of short-term financial return on investment as the slide scanners are not inexpensive. But the technology is mature as witnessed by those implementations and the potential for AI is going to make a big impact as many of the AI detections are relatively straightforward (e.g. simple cell counting).
2. Vendors are scrambling to implement DICOM in their systems. As a DICOM consultant, I have been approached by several of them who tell me that if they are not offering DICOM within the next 1-2 years that their customers will revolt and throw them out for those who do.
3. I believe that digital path is still somewhat at the bleeding edge in the US and there is going to be a fall-out from vendors who try to stick to their proprietary interfaces (I won't mention any names but if you look at them carefully you'll see who does and who does not support open standards). I think that it will only be 1-2 years before reliable, open systems (as well as cloud based solutions) will be available for digital path.
In conclusion, I believe that everyone should be looking today for digital path, be careful with picking your vendor to make sure they support standard interfaces, do due diligence with checking if they have sites up and running in the US that are well supported and operating, make sure your infrastructure is ready (networks, archive) as everything needs to be upscaled about a factor of 10, and consider enterprise solutions to share resources (VNA...).
Herman Oosterwijk
Andreas Vendel
Nov 2, 2022, 9:31:24 AM11/2/22
to
On Tuesday, November 1, 2022 at 14:56:52 UTC+1, Bohsen wrote:
>
> ...
There are challenges still. One is getting LIMS connections to scanners in order to populate the header. Another challenge is that there are some degrees of freedom in standards interpretation. You also can’t assume that software dealing generally with DICOM will handle whole slide imaging well in all aspects, especially not on the viewing side.
I think there is a first stage of DICOM adoption and maturity where it facilitates not having to deal with a plethora of proprietary WSI formats. My opinion is that we are already at this stage. I work as a software architect at Sectra, and we have customers using DICOM for digital pathology successfully in production.
The next stage is where we can leverage all DICOM tools, frameworks, products, and solutions that we are used to in other -ologies also for whole slide imaging. DICOM working group 26 has arranged several connectathons demonstrating parts of this. Interoperability has been demonstrated between scanners, archives, image management systems, and viewers, from different vendors, all using DIMSE and DICOMweb protocols.
You mention AI support. There is nothing about DICOM that would prevent AI or image analysis in general. The pixel data should be the same as in the proprietary formats, only wrapped in DICOM. I don’t think all AI vendors support DICOM yet, though, but that is probably due to lack of customer requirements and a matter of time. For AI/IA results, there is a quite recent module added to the standard for this purpose, the Microscopy Bulk Simple Annotations Module (https://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_C.37.html#sect_C.37.1.2).
// Andreas Vendel
>
> ...
> Does anyone here in the forum have experience or perspectives on it?
>
Currently, you will probably find more software vendors supporting the most common proprietary formats than vendors supporting DICOM for whole slide imaging. This is changing, though, and the major scanner vendors are either already producing DICOM or working on it. In Europe, having a requirement for DICOM in tenders is quite common.
>
There are challenges still. One is getting LIMS connections to scanners in order to populate the header. Another challenge is that there are some degrees of freedom in standards interpretation. You also can’t assume that software dealing generally with DICOM will handle whole slide imaging well in all aspects, especially not on the viewing side.
I think there is a first stage of DICOM adoption and maturity where it facilitates not having to deal with a plethora of proprietary WSI formats. My opinion is that we are already at this stage. I work as a software architect at Sectra, and we have customers using DICOM for digital pathology successfully in production.
The next stage is where we can leverage all DICOM tools, frameworks, products, and solutions that we are used to in other -ologies also for whole slide imaging. DICOM working group 26 has arranged several connectathons demonstrating parts of this. Interoperability has been demonstrated between scanners, archives, image management systems, and viewers, from different vendors, all using DIMSE and DICOMweb protocols.
You mention AI support. There is nothing about DICOM that would prevent AI or image analysis in general. The pixel data should be the same as in the proprietary formats, only wrapped in DICOM. I don’t think all AI vendors support DICOM yet, though, but that is probably due to lack of customer requirements and a matter of time. For AI/IA results, there is a quite recent module added to the standard for this purpose, the Microscopy Bulk Simple Annotations Module (https://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_C.37.html#sect_C.37.1.2).
// Andreas Vendel
David Clunie
Nov 2, 2022, 10:01:14 AM11/2/22
to
Orthanc was certainly an early adopter, but I am not sure that it has kept up with current WSI features and trends in the DICOM standard, so check with dciodvfy, etc., as well as testing with other open source DICOM WSI viewers.
There are several other open source projects related to creating (converting), storing, displaying and analyzing DICOM WSI, which can be found with a simple Internet search. At the very least, a Google search should find Google's own WSI tools, including a converter (which, by the way, has lots of undocumented options that may be useful to you).
Ideally, vendors would provide the DICOM format natively, which some vendors can do with some models of their scanner, in some countries (depending on what has regulatory approval, which as I am sure you know is a roadblock as it was early on with mammography). If the vendor can't or won't do it, then they (or you) can partner with a third party converter vendor who will (just like some folks did with their old CT and MR scanners back in the days of Merge boxes, etc.).
Conversion from some proprietary formats is possible, but it is important to ensure that the conversion is LOSSLESS (at least for the high resolution layer, if not all the layers of the pyramid). "Lossless" in this case means that if the original tiles are baseline JPEG or J2K compressed, then the compressed bitstream (at least at the DCT coefficient level) is reused, rather then decompressed and recompressed (which causes unacceptable further loss). Not all converters have this "lossless" conversion capability, but some do.
As Andreas mention in his reply, the other key question is where does the metadata come from and is it sufficiently well populated? Strange as it may seem, most slide scanners (and thence their proprietary format output) know nothing about who the patient is, what kind of fixation, embedding or stain was used, what date the specimen was acquired, and do little more in this respect than scan and OCR the slide label barcode with its single identifier. At some point, a lookup in the AP-LIS needs to be performed and the metadata populated, in order for the images to be detachable from a device connected to the AP-LIS, for them to be useful and not be orphaned.
If you were thinking about using Presentation States for annotations (e.g., locations on a slide), as you probably are since you mentioned AI, then you are barking up the wrong tree, since Presentation States are semantically meaningless. For coordinate based annotations, DICOM SR or the new Microscopy Bulk Simple Annotation objects (to which Andreas provided a link) should be used, and for bit masks, DICOM Segmentations have been extended to support pyramids and tiling.
For visualization rather than annotation tasks though various Color Presentation States may have their uses for WSI. Markus Herrmann from MGH did a nice job of creating some Advanced Blending Presentation States to select, pseudo-color and fuse multiplex immunofluorescence channels. You can find examples of these in the HTAN collections in the Imaging Data Commons (IDC). That is an exotic use case, and not applicable to brightfield true color H&E images (many DICOM-encoded examples of which you can also find at IDC).
Also, don't forget that there is more to this than the DICOM files. You also need the DICOM protocols to get them from the scanner to the archive (usually C-STORE, but could be STOW-RS), and to fetch metadata and tiles from the archive for viewing or analysis (for which WADO-RS (DICOMweb) has proven to be extremely useful for virtual microscopy zero footprint web browser based viewers). There are no WSI-specific requirements that prevent a generic VNA DICOMweb server being used for WSI. We use Google's Cloud Healthcare API DICOM store off-the-shelf implementation in IDC for radiology and WSI, for example.
For AP-LIS integration, see also the IHE PaLM DPIA Integration Profile.
David
PS. For a recent discussion of DICOM for WSI at a high level, you may find a recent discussion with Aleks Żuraw helpful "https://www.youtube.com/watch?v=9aIL61yL-WQ".
There are several other open source projects related to creating (converting), storing, displaying and analyzing DICOM WSI, which can be found with a simple Internet search. At the very least, a Google search should find Google's own WSI tools, including a converter (which, by the way, has lots of undocumented options that may be useful to you).
Ideally, vendors would provide the DICOM format natively, which some vendors can do with some models of their scanner, in some countries (depending on what has regulatory approval, which as I am sure you know is a roadblock as it was early on with mammography). If the vendor can't or won't do it, then they (or you) can partner with a third party converter vendor who will (just like some folks did with their old CT and MR scanners back in the days of Merge boxes, etc.).
Conversion from some proprietary formats is possible, but it is important to ensure that the conversion is LOSSLESS (at least for the high resolution layer, if not all the layers of the pyramid). "Lossless" in this case means that if the original tiles are baseline JPEG or J2K compressed, then the compressed bitstream (at least at the DCT coefficient level) is reused, rather then decompressed and recompressed (which causes unacceptable further loss). Not all converters have this "lossless" conversion capability, but some do.
As Andreas mention in his reply, the other key question is where does the metadata come from and is it sufficiently well populated? Strange as it may seem, most slide scanners (and thence their proprietary format output) know nothing about who the patient is, what kind of fixation, embedding or stain was used, what date the specimen was acquired, and do little more in this respect than scan and OCR the slide label barcode with its single identifier. At some point, a lookup in the AP-LIS needs to be performed and the metadata populated, in order for the images to be detachable from a device connected to the AP-LIS, for them to be useful and not be orphaned.
If you were thinking about using Presentation States for annotations (e.g., locations on a slide), as you probably are since you mentioned AI, then you are barking up the wrong tree, since Presentation States are semantically meaningless. For coordinate based annotations, DICOM SR or the new Microscopy Bulk Simple Annotation objects (to which Andreas provided a link) should be used, and for bit masks, DICOM Segmentations have been extended to support pyramids and tiling.
For visualization rather than annotation tasks though various Color Presentation States may have their uses for WSI. Markus Herrmann from MGH did a nice job of creating some Advanced Blending Presentation States to select, pseudo-color and fuse multiplex immunofluorescence channels. You can find examples of these in the HTAN collections in the Imaging Data Commons (IDC). That is an exotic use case, and not applicable to brightfield true color H&E images (many DICOM-encoded examples of which you can also find at IDC).
Also, don't forget that there is more to this than the DICOM files. You also need the DICOM protocols to get them from the scanner to the archive (usually C-STORE, but could be STOW-RS), and to fetch metadata and tiles from the archive for viewing or analysis (for which WADO-RS (DICOMweb) has proven to be extremely useful for virtual microscopy zero footprint web browser based viewers). There are no WSI-specific requirements that prevent a generic VNA DICOMweb server being used for WSI. We use Google's Cloud Healthcare API DICOM store off-the-shelf implementation in IDC for radiology and WSI, for example.
For AP-LIS integration, see also the IHE PaLM DPIA Integration Profile.
David
PS. For a recent discussion of DICOM for WSI at a high level, you may find a recent discussion with Aleks Żuraw helpful "https://www.youtube.com/watch?v=9aIL61yL-WQ".
Bohsen
Nov 2, 2022, 2:39:14 PM11/2/22
to
Hi everyone.
Thank you all for all your feedback. It's really appreciated.
> ... if they are not offering DICOM within the next 1-2 years that their customers will revolt and throw them out for those who do.
This is for sure here in Denmark. The Danish strategy on this area is full DICOM in the near future as Pathologists are scarce and DICOM makes it possible to share the workload more easily across the country/world and also take advantage of our scalable storage solutions that we use in Radiology.
> There are challenges still. One is getting LIMS connections to scanners in order to populate the header.
The problem with metadata is definitely something I'm worried about also. The glass-scanner-vendors that we've approached so far promise a solution for this in near future, but none of them specifically mention DICOM worklists (or anything alike). I've been told that our LIS has an API (proprietary solution) that expose data that we could use somehow. But I would really love to not have to use this.
> If you were thinking about using Presentation States for annotations (e.g., locations on a slide), as you probably are since you mentioned AI, then you are barking up the wrong tree, since Presentation States are semantically meaningless. For coordinate based annotations, DICOM SR or the new Microscopy Bulk Simple Annotation objects (to which Andreas provided a link) should be used, and for bit masks, DICOM Segmentations have been extended to support pyramids and tiling.
I was thinking annotations both in regards to AI, but also for the pathologists documentation and use at MDT conferences and I was wondering how to solve the coordinate-based annotation problem. I didn't know about Microscopy Bulk Simple Annotation objects. I will take a closer look at this. The problem here would then be the vendors adoption of this profile. Our VNA/PACS already supports GSPS and this came closest to what we could think of using for documenting findings. Structured Reports is of course also a possible solution. The adoption of using SRs here in Denmark is sadly very limited to my knowledge, but I definitely see your point. Our VNA/PACS has build-in support for SRs. Probably a good time to take a closer look at what functionality it has to offer.
I'm aware of the IHE profiles and I think it's really great work.
Again thanks for all the great feedback.
Kind regards,
Thomas
Thank you all for all your feedback. It's really appreciated.
> ... if they are not offering DICOM within the next 1-2 years that their customers will revolt and throw them out for those who do.
This is for sure here in Denmark. The Danish strategy on this area is full DICOM in the near future as Pathologists are scarce and DICOM makes it possible to share the workload more easily across the country/world and also take advantage of our scalable storage solutions that we use in Radiology.
> There are challenges still. One is getting LIMS connections to scanners in order to populate the header.
> If you were thinking about using Presentation States for annotations (e.g., locations on a slide), as you probably are since you mentioned AI, then you are barking up the wrong tree, since Presentation States are semantically meaningless. For coordinate based annotations, DICOM SR or the new Microscopy Bulk Simple Annotation objects (to which Andreas provided a link) should be used, and for bit masks, DICOM Segmentations have been extended to support pyramids and tiling.
I'm aware of the IHE profiles and I think it's really great work.
Again thanks for all the great feedback.
Kind regards,
Thomas
herman o.
Nov 9, 2022, 6:09:38 AM11/9/22
to
I found another interesting article which summarizes some of the business roadblocks and drivers:
https://www.itnonline.com/article/will-2023-be-year-digital-pathology-us-labs
Herman O.
https://www.itnonline.com/article/will-2023-be-year-digital-pathology-us-labs
Herman O.
Bohsen
Nov 10, 2022, 9:32:01 AM11/10/22
to
After reading it, I think the US will have their own set of difficulties adopting Digital Pathology. I'm glad I only (mostly) have to worry about the technical/integration details.
/Thomas
0 new messages