[00:01:39] Aleksandra Zuraw: Today, my guest on the podcast is David Clunie, and we are going to be talking about standardizing annotation format. Hi, David. How are you today?
[00:01:48] David Clunie: Very well, thanks.
[00:01:49] Aleksandra: Thanks for joining me. And let’s start with you. Let’s tell the listeners about yourself. I’m going to just say to me your the DICOM person, and there is a lot more to your background than DICOM. So let’s talk about your background and then we’re going to dive into the annotations.
[00:02:07] David: Okay. So I started out in radiology, not pathology and was a radiologist/neuroradiologist for a little while, but was much more interested in computers and the informatics side of radiology solutions. And so I got into the question of standardization and interoperability of medical imaging formats in the radiology field. And when the dotcom standard came out in the early ’90s, I grew interested in that and started working with various different working groups that defined the standard and became the editor of the standard.
[00:02:36] And since then, I’ve been primarily consulting in that field and of late have been focusing on digital pathology and in particular annotations in digital pathology and using and extending the DICOM standard to support development of those. Mostly I work with government contracts, some private contracts related to cancer imaging. And, for example, I’m a participant in the NCI’s Imaging Data Commons project, where we’re doing a lot of radiology and pathology imaging in the standard format in an archive for cloud computation for computational pathology.
[00:03:11] Aleksandra: So basically DICOM is your main job now.
[00:03:15] David: Right.
[00:03:16] Aleksandra: Mm-hmm [affirmative]. So DICOM is an image format, but I know your working group, which I was a part of, is working on DICOM for annotations as well. So let’s start with what are annotations because annotations mean slightly different thing to everyone. I came across annotations for image analysis purposes, and it can be a contour of a tumor that’s what I experienced when I was working with annotations. This could be dots on the cells of interest.
[00:03:50] And now with the deep learning entering the image analysis scene, these are just examples of regions of interests. I know they them as vector annotations or pixel painted annotations. And you’ll tell me what the official format names for these are, but how do you define annotations for the work you’re doing for the DICOM standardization of annotation formats?
[00:04:16] David: Well, like you say, annotation, it means different things to different communities and stakeholders. And in the most general sense, an annotation is saying something about something else. And so you can have an annotation about an entire patient or case. This is a patient with cancer, or this is a normal patient with a particular cancer.
[00:04:36] Aleksandra: Sort of like a label.
[00:04:38] David: Yeah, a label or a categorical statement, or it can be down to the pixel level. So you can have an image, scanned image of a slide, and you can say that this individual pixel is part of the nucleus as opposed to part of the cytoplasm, or this is part of a mitotic figure or whatever, or it’s part of an area has tumor in it, or that has a particular kind of adenocarcinoma of a particular tissue. Or it could be an anatomical classification. This is part of a lymph node.
[00:05:10] And so the annotation spans the entire spectrum. So one needs to be capable of representing annotations at all levels, whether they be categorical or quantitative. In a sense, measurements are also annotations, the size of something or the percent of something that is a particular feature and so on.
[00:05:31] Aleksandra: Do you have a definition in your standard for annotations?
[00:05:36] David: Actually, for a long time, we didn’t. Rather, we had sort of things that could be used to represent annotations in certain contexts. And we just sort of assumed that people knew what an annotation was. It’s only relatively recently that people have actually even begun to use the word annotation in the context of DICOM.
[00:05:58] We’ve had overlays and contours and RT structure sets and presentation states and structured reports, all sorts of means of communicating annotations, but we didn’t really use the word annotation, per se, even though we had many different ways of representing an annotation. In radiology, as a case in point, traditionally we’d refer to a region of interest, ROI, is a term that pathologists probably don’t use. But for many years, that was the term of in radiology to describe what is essentially an image relative annotation.
[00:06:32] Aleksandra: Before we dive into the technical piece, let’s take a step back and talk about what is the DICOM format and how the DICOM format relates to other image formats. For the images now, not for annotations yet.
[00:06:46] David: Right. So the DICOM effort really began in the mid ’80s. The first version of DICOM, if you want to call it that, was released in 1985. And it was a sort of marriage of convenience between the vendors of radiology systems and the users of radiology systems as represented by the professional societies. And, in particular, the NEMA, National Electrical Manufacturers Association, in the United States and the American College of Radiology were kind of forced together by the FDA at the time or individuals within the FDA, because there was a burgeoning problem of lack of interoperability between new and very expensive systems.
[00:07:22] You’d have CT scanners coming on the market that would store and share their images in a proprietary format, and yet need to be used on third-party workstations or stored in third-party archives. So if you had, for example, a General Electric CT scanner and a Siemens CT scanner, and a Philips CT scanner, they would put out different formats and they would not be compatible with each other’s workstations and archives. And this was recognized to be a problem, especially when you were spending many millions of dollars on it.
[00:07:52] And so as a consequence, the vendors and the users and the regulators realized that interoperability was necessary. So they looked at what was available off the shelf, and there were really no standards for the interchange of images and certainly not images with the kind of metadata that is needed to manage clinical images including patient identification, scanner identification, and scanner characteristics. And so they all got together and defined the DICOM standard. It was about the same time that the TIFF standard and the JPEG standard were being defined for consumer applications, and so it follows similar principles.
[00:08:25] Aleksandra: So we don’t have to talk about it more because there is another podcast episode for a different podcast, Beyond the Scope by DPA, that you gave an interview on that. And I’m going to link to this podcast episode in the show notes. So everybody who wants to learn about DICOM format for the images itself, I can refer you, but what is DICOM for annotations and how does DICOM relate to other annotation formats?
[00:08:50] And my limited knowledge on annotation formats is limited to the Imperial and XML format. And I like that you can send this annotation via email, and it can be detached from your image. And if the other person has the same image, it’s going to superimpose on this image. So from that stand, I like this a lot, but tell me about DICOM for annotations and how it relates to, for example, XML or other formats.
[00:09:20] David: So DICOM has had several different mechanisms for doing annotations on images and on sub-parts of images for a long time. And, as you can imagine, for example, in the radiotherapy world, which is heavily standardized using DICOM because they have means of drawing contours around tumors that need to be radiated during therapy or tissue that needs to be protected. We’ve needed to be able to identify lesions on CT and MR scans and communicate that and display them on the screen. And so we’ve had annotation formats for that, and they have been designed around two patterns, essentially.
[00:09:54] One is to draw contours around something and code those coordinates in a way that can be understood by the recipient relative to the images to which they were intended to apply. Or you can have a sort of a bit mask or a bit plane or paint over the pixels, depending on how you want to think of it. And so each pixel in the image has a corresponding pixel in the annotation, which is either a zero or a one to put it simply, which indicates whether you are or are not within the region of interest.
[00:10:22] And so this is well for radiology, where the images are generally relatively small in terms of individual frames or sections or slices. But when you get to digital pathology where you have the entire slide scanned, and you have hundreds of thousands of tiles and millions of pixels, and you may be annotating not just an area of tumor in a lymph node, for example, but rather every nucleus on the entire slide, then you get into a problem of scale. And just as you can do it with radiology, you can use the existing radiology mechanisms for pathology, but they become very large objects.
[00:10:59] So in DICOM we decided to create a very efficient representation that’s specific to the digital pathology task and intended to scale to millions of objects. And it is a contour-based format that outlines the locations either on the slide, relative to the image, or in a 3D volume on a series of Z-stacks. And it’s specifically intended for the computational pathology and whole slide imaging use case. So the existing mechanisms that DICOM has are still applicable for digital pathology and usable for it.
[00:11:30] But we do now have as of earlier in the year, when we finalize this in the standard, a new mechanism for what we call bulk microscopy annotations, which is intended to be very compact. And it is a binary format rather than an XML or JSON format, because it just becomes easier to transfer and easier to import if you have essentially a flat, numerical array of coordinates that can be easily ingested by an AI application.
[00:11:56] Aleksandra: Okay. You say it’s the contour-based format. So that will be the vector annotation model rather than the bitmap annotation model.
[00:12:04] David: Right. We already have a bitmap annotation model. So the DICOM segmentation object, which is a rasterized bitmap, has already been extended to support the pyramidal tiled approach of encoding whole site images that’s used in DICOM just as in, for example, TIFF or FCS.
[00:12:21] Aleksandra: So DICOM for images, it’s not just the format in which the pixels are encoded, it’s also a bunch of metadata associated with the image, right?
[00:12:33] David: Absolutely. And that’s the key distinguishing characteristic of DICOM, as opposed to, for example, TIFF is intended to encode the pixels and describe the pixels. And it is extensible in the sense you could add tags to the TIFF header that that said other things like who the patient was, but that’s not what people have done with TIFF. And that’s not what the whole site imaging vendors do when they use TIFF in their proprietary formats like SVS.
[00:12:56] Whereas DICOM, having come from a clinical background, is intended to be a format and a protocol that allow for management of the images such that the metadata is included within the image header. So the metadata in a DICOM image header says who the patient is, who took the image. [crosstalk] –
[00:13:14] Aleksandra: What’s an image header?
[00:13:15] David: Header is just the text or binary information that accompanies the pixel data. So it’s just as in a-
[00:13:22] Aleksandra: Like a description.
[00:13:23] David: Like a description. So for example, in a JPEG file or a TIFF file, there’ll be tags that say the number of rows and columns and how bits many there are in the image pixel data. There are additional tags in DICOM that are defined to encode who the patient is. So that’s put into the header of the image because it’s usually in front of the bulk pixel data.
[00:13:44] Aleksandra: Okay. So is this header information in DICOM for annotations similar to what it is in DICOM for images?
[00:13:55] David: Yes, it’s exactly the same, in fact. So basically a DICOM annotation is like a DICOM image, except you sort of cross out the pixel data and put in the annotation data as the bulk data instead, but the header is largely the same in that the same tags are used to identify the patient. The same tags are used to identify the date that it was obtained. The same tags are used to provide the unique identifiers that relate the images and the annotations together.
[00:14:23] Aleksandra: Mm-hmm [affirmative]. So I have an opinion, or I have… I feel a certain way why we need to standardize annotations. What’s your reason, or what’s the DICOM rationale for standardizing annotation? Why do we need a standard for annotations?
[00:14:41] David: Because we have a pressing need for an interoperability boundary between the applications that create the annotations and the applications that consume the annotations. So particularly given the explosion of interest in computational pathology, where we need annotations to train deep learning models. And we need annotations that are the output of deep learning models. If we imagine a world where pathologists use one platform to create the annotations that are used for training, and they need to be ingested by another application from a different vendor or a different software supplier that are going to use it to train the model.
[00:15:18] And in that model, when it’s running, either in the laboratory or in the clinic is going to be producing annotations that another system needs to be able to view or display to the pathologist who’s doing the report. Each of those solutions may come from a different vendor. So just as the images need to be interoperable, the annotations need to be interoperable. So we have defined a clear boundary where a standard is needed, and there is certain logic in using the same standard for the annotations and the images. And it’s up to the users to make sure that the solutions they buy, particularly from commercial vendors, are interoperable across those boundaries.
[00:15:55] Aleksandra: Mm-hmm [affirmative], so what I think about it is a different side of what you said, you are focusing on the interoperability. To me, it’s like there is, especially for machine learning and artificial intelligence model development, there is a lot of work going into making those annotations to train the models. And I feel those notations are mine. I want to be able to take them out. So first to export them regardless of the format, but basically take them out and use somewhere else if I choose to. Basically kind of more personal or more selfish motivation for interoperability.
[00:16:38] David: It’s your data and you own it. So you should be able to pick it up and take it somewhere else as you need it.
[00:16:43] Aleksandra: Yeah, exactly. I want that. And I might have collaborators and I want to have the data from my collaborators with the information attached, who made it, for what was it made and everything that would be included in the DICOM standard. And in the Beyond the Scope podcast that we already mentioned, and I will link to, by the DPA, you said vendors only provide what the customers ask for.
[00:17:08] So I think the pathology community is not really used to this approach. I don’t know if radiology community was, or it was developed over time. What can we do as end users to get what we want to get DICOM and annotation interoperability down the road, even if it’s not yet incorporated? And by the way, do you know vendors who already incorporate this both for images and for annotations?
[00:17:33] David: Well, yes, there are some who are starting to, and I mean, as you alluded to in my previous podcast, I said, what the customers ask for. So you, as the buyer, may have a certain amount of power in the marketplace and they, as the seller, may have a certain amount of power. And it’s not a free market because it’s skewed by all sorts of factors, including limited competition, the regulatory actions on what you can and can’t do for medical devices and so on and so forth.
[00:18:00] So it’s a little different from a free market, but the bottom line is if you won’t buy a solution because it doesn’t have what you want in it, and you will say, “I will buy from your competitor instead,” then you have power. So if you prioritize interoperability and the adoption of the DICOM standard, be it for the pathology images or for the pathology annotations or both, then the vendors will respond to that.
[00:18:23] And product managers and vendors are always looking for opportunities to gain market advantage. And if they can deliver what the customers are asking for, then they will. But of course they have limited resources. So if you don’t ask for it, they won’t build it. They may have some foresight and say, “We anticipate that the users will need this. And so we will prioritize interoperability,” over some other feature that they might add, but it is essentially a competition, and you as a buyer have some power.
[00:18:53] And, furthermore, you don’t necessarily have to have it today, but when you buy something, you can put it in contractually and say, “You don’t have to deliver it today. But if you deliver it within 12 months, then I’ll buy your solution.” And that gives them a little breathing room, but they still have to deliver in the end for fear of a contractual penalty. And so there are many tools that one you can use to exercise your buying power.
[00:19:16] One has to use that in moderation, obviously, but right now, for example, in the United States, there are two general purpose whole slide scanning solutions available, one from Philips and one from Leica. And one of them uses a completely proprietary solution internally and will grudgingly convert things to DICOM if pressed. And the other-
[00:19:36] Aleksandra: That’s Philips.
[00:19:36] David: … is going in the direction… That’s Philips. And the other is going in the direction of having DICOM come out natively from their scanners. So assuming all other factors were equal, the qualities of the scanners, the throughput of the scanners, usability of the scanners, number of slides they break, all that kind of stuff… Assuming everything else was equal, then one might have an opportunity to use one’s buying power to pick a vendor that was going down a standards pathway, as opposed to a vendor that is not going down the standards pathway.
[00:20:05] And hopefully in the longterm, all the vendors will realize that it’s better to use a standard solution. And that allows the customer to mix and match best of breed all of those kind of buzzwords that people use. And that applies equally to annotations. Obviously, annotations are not yet as important as getting the images standardized, but it’s getting there. And so we’re trying in the standard to provide the necessary mechanisms for the vendors to implement as the users ask for these features.
[00:20:31] Aleksandra: To my understanding, DICOM technologically is a format is like any format. So why are the vendors not choosing to use this format? Okay. We mentioned that there’s a little bit more to this format than just encoding the pixels, but what’s the greatest challenge in implementing DICOM, for both images and down the road for annotations?
[00:20:55] David: I think one of the issues is lack of familiarity of the developers. DICOM is a medically specific format. So if you take a developer off the street, they’re probably familiar with JPEG and they might be familiar with TIFF because it’s used for faxes and stuff like that.
[00:21:10] Aleksandra: Yeah, but if they are now working on a, let’s say, medical device or a product for medical market, I would assume you need to have enough background knowledge of what’s there. I’ve seen people reinventing the wheel so many times for slide viewers, for things that are out there for free. You can look it up. Let’s take opensource [inaudible] . You can look up all the features that are there and just build stuff that works.
[00:21:41] David: Well, in radiology, it’s pretty easy for DICOM because there are so many solutions out there. There are toolkits and libraries and viewers and archives and so on and so forth. So it makes no sense to reinvent something, but in digital pathology, it’s not really the case. I mean, all of the libraries for handling images for analysis while viewing are forced to read all of the different proprietary formats.
[00:22:05] And as these sort of nascent vendors in the whole slide imaging field have started to release solutions to clinical, they have used their research, software research tools, research hardware, which has been proprietary. Libraries, like OpenSlide and Bio-Formats, try to keep up with reading every different format out there. And every time a scanner develops a new feature or something, then they have to learn that new feature of their proprietary format and add it to the library.
[00:22:36] So the presence of these kind of read-anything libraries has to some extent reduced the pressure on the vendors to standardize in that they just assume that you’re going to have OpenSlide and hope that will work, but that doesn’t scale very well. And it certainly doesn’t scale operationally to volumes in terms of millions of slides per year, as opposed to a few hundred or a few thousand slides per year in a research setting. So the scalability ultimately depends on the adoption of standards, but the interim solutions, if you want to call it that have sort of taken the pressure off the vendors a little bit.
[00:23:10] Aleksandra: Okay, so what failures did you experience, and what did you learn from them when trying to implement DICOM in pathology in comparison to radiology? And, I don’t know, you mentioned that the CT scanners and the radiology equipment is so much more expensive, like millions of dollars so I guess this is a bigger pressure than pathology hardware and equipment. Even though everybody says that cost is prohibitive still, but compared to radiology I don’t think it is. So, okay, we might have less pressure because the things cost less, but what failed compared to radiology, or what is still in the workings?
[00:23:55] David: I think it’s premature to make any conclusions about successes or failures. The adoption of digital pathology in the world is very minimal at the moment. Only a few major centers, aside from research, of course, aside from preclinical research. But in clinical use, a number of hospitals, a number of clinics around the world that are doing everything digitally, that have set aside their microscopes, can be almost counted on one hand. And so it is a very immature market, and there’s a lot remains to be shaken out.
[00:24:23] There are very few products that are approved by the regulators yet. And so the standard that’s expected in terms of validation and regulatory approval is still fairly high and, to some extent, mitigates against interoperability. So the FDA, for example, has been focused on the entire pixel pathway. Scanners have to be married with archives and display software and display hardware to develop the full pixel pathway end to end as the FDA [crosstalk] .
[00:24:51] Aleksandra: That’s basically pushed the Philips system to be such a closed system.
[00:24:56] David: Yeah, and the Leica system. The Leica system has to be coupled with its own viewer, and the FDA has approved third-party viewers, both one for the Philips system and one for the Leica system.
[00:25:07] Aleksandra: Which one are-
[00:25:10] David: Sectra and Paige. And so there is some interoperability there, but it remains for them to relax that stance and separate the components and approve them separately. So, in radiology, for example, you approve CT scanners separately from display software, separately from display hardware, separately from archives, if the archives even need approval at all, because they’re just basically arrays of discs with a little bit of management software. So the componentization, the deconstruction of packs as it was called for a while in radiology, is not really enabled by the current regulatory market for digital pathology in the same way it is for radiology. So when that change-
[00:25:47] Aleksandra: And why is that? Is that because it’s easier to control a closed system? It’s already done in radiology. Why are we doing this from a different angle in pathology?
[00:26:00] David: Well, it’s a question of proof. So the regulators want to have proof that it’s safe to decompose your solution. So, in radiology, they are now and have long been familiar with the notion that the display software and hardware is completely independent of the scanner software and that you don’t put the patients at risk by using a third-party display system. Whereas in pathology, because image quality is pushing the limits of scanning technology and display technology, and there is lack of familiarity with what’s safe, they are waiting for evidence.
[00:26:33] So when a vendor goes to them with a scanning solution, they have to be able to display it. And they have to show clinical evidence as well as technical data that the images as seen on the screen and read by the pathologist are equivalent to those conclusions drawn from looking at the same tissue on a microscope slide down an optical microscope. And so that proof has to be provided.
[00:26:57] So when the vendor starts to provide proof of equivalence, then they will chill out. And that’s happening with the Leica and Paige viewers. They provided data, not necessarily clinical data, in some cases technical data, that was sufficient to convince the FDA that it was a technically equivalent and therefore clinical equivalent solution. And the more proof they get, the more they’ll relax. And then eventually it will just become like radiology, I would imagine.
[00:27:22] Same goes for AI. The regulatory challenges for AI safety and efficacy are incredibly complex and many different approaches have been described. And, as you’re aware I’m sure, there’s a lot of controversy over whether AI products that are on the market are really efficacious and whether they’re being held to the necessary standards and what those standards should be. I think that annotations are not really a regulatory problem in that the way in which you encode annotations is essentially moot if it meets certain technical criteria. So I’m hoping that the use of interoperable annotations in the DICOM format will not face any kind of regulatory barriers.
[00:27:59] Aleksandra: Okay. So from what I understand, it kind is of a natural cycle. Somebody starts in whichever capacity it’s being allowed at the moment. So let’s take the Philips, it had to be a closed one, and then people start expanding upon it by showing proof of equivalency and enough data so that the regulators can relax a little bit.
[00:28:24] David: Well, I mean, even the Philips solution didn’t have to be the way it is. They could have used the DICOM standard in their product. They elected not to because the DICOM standard predated their product and their [crosstalk] approval.
[00:28:34] Aleksandra: Why did they elect not to? Why do people elect not to if it’s already there? And let’s say they’re starting development from scratch, it’s there. You don’t have to think. You can adopt it. Why people choose not to?
[00:28:48] David: It’s a very interesting question. People convince themselves that their homegrown invented technology is in some way better. And in the case of Philips, they had a wavelet-based compression scheme, because lossy compression is needed for practical and digital pathology. And they were of the opinion that their homegrown compression scheme was better than using JPEG or JPEG 2000 in DICOM because DICOM uses standard compression schemes in order to minimize the burden on the implementers.
[00:29:18] In retrospect, that probably proves not to be the case, but at some point in time inside Philips one must assume that they thought that there was an advantage to doing it this way, in terms of image quality or speed of retrieval and review or whatever parameters they were trying to optimize for. And the price they have paid for it is a closed monolithic solution.
[00:29:40] They could have built their solution as such that their compression technology used standards instead with the same microscope hardware and hence the same inherent image quality and prioritized interoperability over whatever other parameters they were authorizing for, but they elected not to. And they will pay the price for that in the longterm because they will ultimately have to retool to use DICOM when everybody else starts using DICOM.
[00:30:05] Aleksandra: So when do you think everybody else is going to start using DICOM? How long do you think it’s going to take pathology to adhere to a unified standard?
[00:30:14] David: Well, the other vendor that’s approved in the United States, Leica, already uses DICOM in its next generation product. And DICOM is the format that they will be putting out by preference. They can, of course, encode it in their legacy format, their SVS format, but it’s got DICOM built in from the start. So the more vendors come to market, and it’s surprising how long it’s taking, honestly, for other slide scanner vendors to get approval in the United States. I don’t know what the reason for the slowdown is, but many vendors have demonstrated, in our Connectathons and in research platforms, they can put out DICOM. And so I think it’s only a matter of time.
[00:30:52] The other thing that people fail to consider is how you deploy this in a large institution. So, for example, you go to a place like Mayo Clinic or Cleveland Clinic, or Memorial Sloan Kettering or some very large institution, Texas MD Anderson. Their enterprise IT and security people are not going to let you build a data center in the pathology department. So your pathology images will have to fit in with the rest of the enterprise. And if every other image in the hospital from every other department is in DICOM, they’re not going to be excited with you wanting to deploy a proprietary solution.
[00:31:24] You might be able to, but the enterprise imaging approach to using a standardized solution and a centralized platform for all images in the hospital provides enormous benefits in terms of economy to scale and simplicity of management, as well as allowing pathology images mixed with radiology images and images from other specialties, dermatology, dermoscopy images and so on, OCT images.
[00:31:46] So the dermatopathologists will be able to see their dermoscopy images side by side with their whole slide images and their gross images and their immunohistochemistry images, all using the same technologies essentially. And so leveraging central IT solutions, central enterprise imaging solutions, is a tremendous motivator for picking a standard that’s already in widespread use.
[00:32:09] And that applies to annotations too. So, for example, if you’re annotating in radiology and you’re creating structured reporting annotations for radiology images, it makes sense to do the same thing for gross microscopy and human annotations on slides. So, of course, for deep learning applications where you’re creating millions of annotations for nuclei and so on, you will need specialized solutions and specialized viewers and specialized analytic tools, but they can plug into the same central imaging enterprise imaging platform.
[00:32:37] Aleksandra: The DICOM standard for images, this is done. It’s already set, and people can use it. How far are you with the standard for annotations? Where are you along the way in development and deployment of the standard?
[00:32:51] David: Very early. So just as it’s been relatively early [crosstalk] –
[00:32:55] Aleksandra: So I cannot pressure anybody to adopt it yet?
[00:32:57] David: Well, you can. Well, you should start pressuring them immediately. It is part of the standard. So we formalized it earlier this year. I don’t have the date offhand, although now that you mention it, I can probably find it easily enough, but-
[00:33:12] Aleksandra: This is something available that I can link to, right?
[00:33:16] David: Oh, yeah, absolutely.
[00:33:17] Aleksandra: Okay, I will put this in the show notes as well.
[00:33:20] David: Yeah, not that anybody who’s a normal human being would want to read a supplement to the DICOM standard, but-
[00:33:26] Aleksandra: If I want to pressure somebody, I want to point to this episode and point to that link.
[00:33:32] David: Yes. It’s supplement 222, and it’s called Microscopy Bulk Simple Annotations Storage SOP Class. And it was released as final text on the 12th of July this year. So it is fairly fresh, brand new, but it’s not rocket science. It’s very straightforward. So anybody who already has a DICOM toolkit could easily store and regurgitate these annotations or convert their internal proprietary format to this format in a very straightforward way.
[00:33:59] So if one has contours, whether they be image relative or 3D relative to a 3D volume of the slide, this supplement will work for them. If they have contours that are polygons or circles or rectangles or dots in the middle of something, this supplement supports all of that on a very large scale. And, furthermore, if they have smaller numbers of annotations, or if they have annotations stored that are bit planes, as opposed to contours, then there are other existing parts of the standard that have long been standardized which can be used to do that as well.
[00:34:31] So, for example, if you have bit plane representation of annotations, those can be stored as what we call segmentation objects, and those segmentation objects support the pyramidal tiled approach to encoding whole slide images as individual frames, as opposed to one monster image. And the same applies to the bit planes that contain the segmented annotations.
[00:34:53] Structured reports are objects that allow you to encode measurements and categorical statements about smaller numbers of annotations. And, again, it supports both bit planes and segmentation objects that are referenced and also contours, 2D or 3D, that may be relative to any kind of image, whether they be radiology or pathology or dermatology or whatever.
[00:35:16] Aleksandra: Do you have images in this standard, like figures, to explain different concepts?
[00:35:20] David: Probably not. We probably could do with some more explanatory material. There is a presentation accompanying supplement 222, which I should send you the link to [crosstalk] –
[00:35:31] Aleksandra: Please do.
[00:35:32] David: … explains it in words of one syllable on a handful of slides, but it doesn’t go to any real detail and it is kind of hand wavy.
[00:35:40] Aleksandra: If we can work on it offline, if you need some help with that, I would be happy to help translate it into normal language.
[00:35:48] David: Yeah, and to be honest, people often refer to major features of DICOM in terms of their supplement. So when they talk about the DICOM whole slide image format, they refer to supplement 145, which was added in 2010 to extend DICOM to support whole slide images. And so if in your contracts, you refer to supplement 222, that’s not the end of the world in terms of demanding a particular choice of format for annotation interoperability. And the vendors will either understand or come to learn what that means, and that can be your point of compliance in the contract with respect to having support for this feature.
[00:36:24] Aleksandra: I’m going to ask you a surprise question that I didn’t prepare in advance, has nothing to do with DICOM. So on your profile picture for your email, you’re in a plane. Are you a pilot?
[00:36:36] David: I am. Actually on my profile picture, I’m in a helicopter, I think [crosstalk] .
[00:36:39] Aleksandra: Oh, sorry, helicopter. I knew it was some flying vehicle.
[00:36:44] David: Yep. Yep. I am.
[00:36:48] Aleksandra: Okay.
[00:36:49] David: That’s my hobby. That’s what I go out and do every afternoon when I get sick of working on DICOM.
[00:36:54] Aleksandra: Really?
[00:36:55] David: [crosstalk] I walk the dog.
[00:36:55] Aleksandra: Do you have a helicopter?
[00:36:57] David: Yes.
[00:36:58] Aleksandra: Oh my goodness. Sorry. It’s just I did not expect that. I thought it was one time thing and not your passion that you do every afternoon. Cool. Okay. So thank you so much for this time that you spent with us, with me and the listeners. We’re going to link to everything we mentioned in the show notes, but is there a place, a go-to place, where the listener can go online to look up all those things we talked about? Where can they find this?
[00:37:30] David: That’s actually a good question to which I don’t have a good answer. There should be is the bottom line, but there probably isn’t. We have not yet written the papers that should be in the academic literature to describe the use of the new DICOM annotations or, for that matter, to go into detail about using existing DICOM features for annotation, and there probably should be. So, I guess, watch this space is the best I can say.
[00:37:53] Aleksandra: Okay. And if anything appears on online, I’m going to be just adding it to the links of this episode. Thank you so much.
[00:38:00] David: That’d be great. Thanks.
[00:38:02] Aleksandra: Have a great day.
[00:38:04] David: Okay.
