[00:01:24]Aleksandra: I’m gonna welcome Gregoire. Welcome, Gregoire.
Gregoire: Hi. It’s nice to meet you, Alexandra, and really thank you for this invitation.
Aleksandra: I am so excited to have you. So I’m gonna let you introduce yourself, but, let me just give a little background story and we have like a full agenda.
We have one hour scheduled for this and I have a list of questions that I need to ask you cause they’ve been like, I was really thinking of different stuff that you have to offer and how you manage this.
So let’s start with Cytomine. Cytomine is the company that we’re gonna be showcasing, but how I even learned of Cytomine.
So, Cytomine. I have this blog post that is a pretty well performing blog post on my digital pathology place where I list, I don’t know how many, softwares now, I don’t know, seven or eight, whatever. Open source software for doing tissue image analysis.
And, I was updating this and I think I updated with [00:02:24] something and then you guys reach out to me and I included Cytomine as an open source tool for doing image analysis but it’s a different open source tool
So let’s do this. Let’s start with you, Gregoire. Tell us about you, about your background and give us a little bit of an introduction to Cytomine. What does it do and what is it for?
Gregoire: Okay. Quite like you, I guess. I’m a vet and my initial formation and I work. I started working in research in toxicology and pharmacology also but I was not a pathologist.
Aleksandra: Where did you study for? Where did you?
Gregoire: It’s, here in Belgium at University of Liège where Cytomine was born. So I was firstly more, working in the pharmacology lab within the university and I always had, geek [00:03:24] side of in my way of working. So in the end of year ’90, web war was coming and quite soon I was starting to develop tools to help research and teaching at university.
And mostly in topics in bioinformatics, biostatistics and histology already at this stage. And yes, my career has drive me into several Belgium University where I had some kind of project in teaching and research with new, what we called new technologies. And in 2014 I was enrolled at again, at University of Liège to come to the Cytomine research group.
And the Cytomine research group has been built by Raphaël Marée which is the leader of this research group at University of Liège. And the team was dedicated really to the development of AI in the topics of digital [00:04:24] pathology. So it was very specific.
Aleksandra: When was that? When did you start?
Gregoire: The Cytomine team was built in 2010. And so Rafael and his colleagues start to develop Cytomine and they start some research, published the first algorithms and paper and so on, and I reached the team in 2014.
So it was in 2014 and I was there to manage a project to use Cytomine also for the teaching aspect because as a virtual microscope, it was really useful for teaching histology and pathology and cytology at university and mostly in very big classroom because in first grade there is more than 1000 students a year, so.
Aleksandra: Oh, wow.
Gregoire: Yes, it’s massive. Yeah. And then we speak. Yes, we start to work on site, line and design. And then I became somehow the coordinator of this teaching project. And in 2016, the [00:05:24] university have taken the decision to publish source code in open source. So the open source adventure started at the University of Liège for academic reason.
And when we start to publish the source code. Rafael and his colleagues has already published some papers. That we’re quite well, like, read by the community team. So we have seen a float of, I want this software, how do i install, I want it in my university, I want to teach with them and then the university.
Aleksandra: When there’s a free software showing up.
Gregoire: The university says that, oh, you are supposed to be a researcher, not a software maintainer, not a community manager. So you have to make some choices. And then with part of the team, we decide to launch a company by the university. What we call a spinoff here in Belgium and then the entrepreneurial adventure started.
Aleksandra: Okay.
Gregoire: This was in 2017.
Aleksandra: [00:06:24] So you said that Cytomine originated as a, like a collaborative teaching tool and this collaborative aspect, but it’s also an image analysis tool. So the collaborative aspect of Cytomine was something that immediately caught my attention when we were talking, because it’s such an important part of image analysis. Tissue image analysis, right?
Gregoire: Yeah.
Aleksandra: And digital pathology is like super collaborative or at least should be. You basically have people from, different, with different expertise.
And this caught my attention. Yeah. So I think that was an unmet need that you guys were addressing when you started developing the platform. So why was it important enough to branch off? Of the open source model and create a paid enterprise version of the software. So this is like something that I.
Gregoire: Oh, it’s a large question.
Aleksandra: Yeah. But I’m like, okay, is this open source or is this not open source? And you guys have like two logo. The [00:07:24] blue logo is for open source. And the non blue is for the pink that you have behind is the commercial. Usually, you know, you have the hardcore open source people that never wanna have this cost any money.
And, you know, there’s plenty of effort going into it. But you have both. So you have like free and paid. What’s the difference? Why do you have it like that?
Gregoire: So first of all, about collaboration. So the status we have made when we start to work in 2010 and then in 2014 is that if you want to collaborate within digital pathology, you must have a central place where all the data, annotations, logging, history, et cetera, is stored.
And we were struggling because we were developing AI or giving teaching courses in digital and pathology, and we must ask the student to install desktop application on their own computer. So everything, what they were doing remains on the computer. We do not have any feedback. And even if [00:08:24] we want to develop an AI for example, we will have to ask as a pathologist with our hospital partners to annotate, exports annotation within the text files, send the text files, imports, annotation in our own servers and try to work with it.
And when the model is ready, send it back to them. No, no, it was very too easy. So we decide that we will put everything on servers and we have developed since the beginning a web application. So Cytomine, since the first day, was always web based for this meanings. That everything is centralized. So you get connected to the platform, you retrieve your data, you can see what your colleague have done.
You see the annotations, you can see that it’s your annotations. You can decide if you share them with the, with other people in the room, et cetera, et cetera. Exactly the same in teaching. So each student can get connected, have only access to the collection of slides the teachers want him to have access to.
The teachers decide if the students are allowed or not to annotate or to follow annotation of the teachers and so. There is a lot of scenarios [00:09:24] that can, that you can bring. So this is for what we call a collaboration. Collaboration means that people like of different level of expertise, student teachers, developer, pathologists, laboratory managers and pathologists and so on have to share the same platform to be sure to be efficient and that you have traceability, reputability, and so on.
Everything, which is definitely needed in research and teaching and in diagnostic. So this being done, we have published the code in open source. And the second part of is how do we manage this duality between the open source and the closed source. So it’s a quite common model in the open source community.
Aleksandra: Okay.
Gregoire: It’s a quite common.
Aleksandra: Do we have somebody in the digital pathology community doing this as well, or like more other software?
Gregoire: There is a lot of project which are open source, but only academics. And there is some projects which are also open source and based on economic model, which is Amira [00:10:24] with Glencoe.
Aleksandra: Ah, okay.
Gregoire: And others. So it’s not uncommon in the open source community. So a lot of software you have a version open source which is allowed for you to get installed to make it run on your computer, on servers, and start to work with. And the idea at the start of the company was we met, so as it’s a server based application. We met a lot of teams that had struggle because they don’t have servers to owe the solutions or they don’t have enough storage in this, on these servers, or they have the servers, but they don’t have the IT guys to make it run and to maintain and to correct when there is a problem and so on.
So we sell mostly services. So as a start, we sell mostly services to install the solution on premise, on the IT infrastructure of the customer. So installation, configuration, we install the monitoring, we decide the rule of the, and the frequency of the maintenance of the troubleshooting. [00:11:24] We start also a lot of consultancy on how to prepares the rise of digital pathology in the different place where they acquires a solution because as you may know, digital pathology is now very well known, but it’s still in transition. So a lot of people still work with microscope and generally when we come into a place, it’s the first.
Yeah. But it’s quite normal. Generally, people as just the first try in digital pathology. So we have to be a companion in this transition also. And then we. The number of customers were rising, and then we had some. Also some struggle with open source because for example, if you want to add a new image format, which is not open source, and you contact the company which has developed this format, and the company said, no, you are not allowed to make it open, you must build a plugin within your open source software, but the plugin is not open [00:12:24] source.
So, quite soon we were quite forced in some manner on another to develop some plugins, not open source. And we have an agreement with different people that we have gonna sell this in this condition and so on. So we start to develop modules like that but also a Cytomine out of the box and in an open source version. If you download it and install it, you will see that it’s quite normal.
It’s every steps is manual. You can create a user by yourself and so on, but you will not spend times to create 1000 user a year if you are. If you are a university, so we developed some close source models to add some features to the platform for example, to automate it all the authentication of the user within the company or the university, et cetera.
Or some tools to directly and automatically upload slides from a repository where the standards are putting the slide and based of the name of the slide, which is also depending of the [00:13:24] barcode, which is on the glass slide directly open, uploaded to the platform in the correct folder for the correct teacher or for the correct lab manager and so on.
So we make lot of automation into the process. And these are also the services that we sell within the company. And now we are ready to release in few months our first enterprise edition. Will be a closed source, full distribution with a collection of new features that we are really targeting, first of all, the sector of education where we are quite active.
Aleksandra: Established.
Gregoire: Yeah, and then six months later for research and within the company, we have taken the decision to target in two years to have a version dedicated to the hospital. So we have started the IVDR certification process and et cetera.
Aleksandra: Oh, congratulations on that.
Gregoire: Or I dunno if it’s because it’s long.
Aleksandra: Well, congratulations on making this decision and good luck implementation.[00:14:24]
Gregoire: And so opens the question directly and you will understood if you can, if you’re gonna get certified. It’s not the same game that building open source. So you have to fix your code. You have to describe every part of the code to be sure that it can be audited and so on. So also the closed source is also linked to the different certifications that we must attach to our software if we wanted to be, run into hospitals for management of image of patients because our main motivation is the first day is to be, an actor and the way to, enhance the quality of the healthcare. On our cases based on the digital pathology place and the rise of AI in this case. So within this strategy, it’s not only based on the platform, it’s also based about building a certified, AI store, in which we will able to accept AI made by us for sure, but also by other companies that want to push their [00:15:24] own AI into the Cytomine of our customers. So it’s a global, global strategy, which is based on this enterprise services. But once again, it’s quite common if you look at some, very well known open source solutions outside of digital pathology.
We have in Belgium here, Odoo, which is a CRM, which is work on the same if you have some Linux distribution, it’s the same. You can have the open source version, but you can have services and extra products if you need, et cetera.
Aleksandra: Okay, so two questions that I have. One is the open source is on premise and the paid one is cloud based. Is that correct?
Gregoire: No, most of our customers are asking us to install On-premise.
Aleksandra: Okay.
Gregoire: The big part.
Aleksandra: So even the one that is cloud?
Gregoire: So most of our customers want to have, and it’s quite normal because digital pathology slides are very sensitive data. So it’s definitely not enough to be hosted on the cloud. So we and [00:16:24] they prefer generally to have it on their own server. But this is cultural, so it’s in Europe, it’s mostly on-premise, on the IT facility of the hospital, of the university, of the research center and so on. And in the North America, in America in general.
In Asia, it’s more asking to host it on Azure, or AWS. But once again, as soon we’re the server side and web based applications, the kind of hosting solution doesn’t really matters.
Aleksandra: So, because many, many companies have On-premise software and then they’re building like a cloud access.
Gregoire: Yeah.
Aleksandra: Is that easy? How does that work? Or do you have to have, so can you like copy the On-premise software and just move it to the cloud? Or do you have to develop the technology is the same?
Gregoire: The technology is the same, when the IT facility of the customers. So the servers are just located in their own IT facility. But, and [00:17:24] when you install it on the cloud, it does it, it just means that you install it on servers that are owned by cloud companies like AWS or Azure. But at the end it still servers whenever they’re looking at it.
Aleksandra: You access it from a browser.
Gregoire: Yeah, every time.
Aleksandra: Ah, every time. Okay. Now I understand.
Gregoire: No. So, it’s a only web-based application for the viewer part. And there you have the key of what we call the collaboration. Let me explain.
Aleksandra: Yes.
Gregoire: A biologist, a pathologist, cytologist, will quite exclusively used the web based viewer and annotating tools and managing data information in the web viewer. But the data scientists, bioinformatics an AI developer will definitely, will prefer to work in the terminal using Python and making applications.
So here is where the magic comes. So in Cytomine, the viewer, it just a kind of front end that you may have for your solutions. You can have front end. We can [00:18:24] say that a python developer, an AI developer, just need to have information, patches, mask, alpha alpha mask, a coordinates of annotations to train these models to be efficient.
And when you run its model on the image, you have to push back to the servers, the class you have done ,the segmentation you have done, the identifications, accountings, everything. So what we have done is Cytomine has an open API. So every component of Cytomine is speaking together.
The db, the database, the image servers, the viewer, et cetera, are speaking together within the open API. And we have made a python client. So if you install on your computer, the python client of Cytomine, you will be able to get connected through python to a Cytomine instance. Ask for the annotation made by this user at this date on this collection of slides for this proposed, and it will be sent to you, the tiles, the classifications, annotations, a coordinate, et cetera. You will be able to make your computational pathology at the end. You may send back [00:19:24] to the servers, the results to allow the pathologist to see it through the web viewer. So both communities, pathologists on one side and AI.
Aleksandra: Dig you just build the perfect tool where the pathologist know any.
Gregoire: I didn’t say that, but that’s why we are generally. I hope guest is chosen by communities. It’s that it allows the both communities to work in the, in tools that are, fits and needs and that what we were doing at university because as AI developers and researchers, we need to have access to libraries in python to make some, a lot of developments. And the pathologist with whom we were collaborating, just need to see the image and verify in their own context if what is has been generated by the AI really feed the context in medical information and et cetera. And they also need, because it was supervised learning at this time. They also need to annotate and to edit the annotation.
You have made a podcast and a short video how [00:20:24] to have efficient annotation for deep learning, for example. So you need to have good annotations. You have to correct them to go back to validate them. So we built a system to peer review annotation. So for example, we have. 10% in the project, we decide that they cannot see what the team’s, , colleagues are doing. They annotate on the slide, and then after, let’s say two months, we show to each other the annotations of the other and we cherry picks the best annotations after editing. And we, what we are building what we call the reviewed collection of annotations and only distribute collection of annotations will be used to feed an AI model, for example. So we have processes to be sure.
Aleksandra: We need to be talking about this because, people need to learn about this, about an efficient process. one question or one interruption. So just to be clear on what I was confused on, there is a web, Web access to it, right?
Gregoire: Yep.
Aleksandra: Cloud or server is just where the thing [00:21:24] is hosted.
Gregoire: Yep. The backend.
Aleksandra: Yeah, backend. So front end is the web. I put the whatever.
Gregoire: Yep.
Aleksandra: Google address, url,
Gregoire: yep.
Aleksandra: Whatever. I go there. I annotate.
Gregoire: Yep.
Aleksandra: As a pathologist. I don’t have to do anything else then, like take my mouse or whatever, tablet, and do my annotations.
Gregoire: Nothing to understand.
Aleksandra: And you.
Gregoire: You just need to have the URL of the Cytomine you have access to, and just to remind your username and password. And even if you have a doubt, you take your tablet at home and you get connected to your Cytomine, and you see your image in emergency if necessary, and so on. So for example, for the student, it’s the same. They can access the slide. Anytime from anywhere. So when I was a student in veterinary medicine, when we had histology classes, you and me, you went, we were forced to be a 10 hour in room C in front of the microscope number 34.
Aleksandra: Slide.
Gregoire: Have the collection of slide, number 53 now is it just, open.
Aleksandra: We we’re like drawing them trying to take pictures.
Gregoire: Yeah, now the students are [00:22:24] in pyjama in their bed they open the browsers and they work whenever they want. Because when? When, yeah.
Aleksandra: on smartphone. How about I learn some histology?
So then, me as a pathologist, I don’t care about the backend. I go in, I draw the programmer or whoever, knows how to code and has to code for this, build AI solutions. They don’t have to deal with my coding inefficient browser for dummies, they can do their python coding on their end. How they usually do.
Gregoire: Yep.
Aleksandra: This feeds into where I can later, I dunno where.
Gregoire: This is. Yeah. This is during the, mostly during the model training or AI development process. And, at the end you have the pathologist with one day wanted to have themselves, the tools that the developers have. So at the end of the process, we have a system that the developers can release the application within the Cytomine with a system which checks a version so they can release different version, and we indicate which is the [00:23:24] latest release, which is duplicated, the one we can use.
One they cannot use anymore. And as soon as you run it from the your web front, you’ll define yourself, your parameters that you want, to sort on the threshold, the numbers of image on which image, et cetera. On the database on Cytomine. We will know that Aleksandra Zuraw have run this specific version of this app on this day, on this data with this parameter.
So six months later when I will see the results and, oh, do you remember this dataset, it was so great with which application has been built, has been built? I just look at the Cytomine. Oh, it’s has been run by Aleksandra Zuraw 11th of, the 10th of November. And with these parameters on this collection of image, and I have my trustability, which is over there.
Aleksandra: We need to do a demo. So this is just, you know, podcast or discussion explaining this. But we need to meet again and you have to like show me on video so that I can show people, and we [00:24:24] post it somewhere. If anybody’s interested in the demo, put demo in the comments. So I will know that when we do the demo, you get the link to this demo. one more question before we go,
Gregoire: Yeah, sure.
Aleksandra: Into. Another important thing, big picture. We’re gonna talk about that, but one question you mentioned that this is also an image management system. Tell me how it’s an image management system and then tell me like with short steps, what are all the things that this software can do?
Gregoire: So what we call it a image management system is that for a Cytomine being able to provide an image to the web front and or to the Python libraries for AI and so on, we need an a system that can be store and read these images, but. what thing that we are facing is, as we all know, is that an image in digital pathology is quite heavy. So, we quite rapidly go into multi gigabyte images and sometimes even bigger, during the biggest.
Aleksandra: we have [00:25:24] some demo candidate. Dr. C wants a demo. So good.
Gregoire: Okay. So, for example, the biggest image we have faced is one 100 gigabytes. So, you understand mostly it’s one or six gigabyte or 100 megabytes, but you rapidly face a situation where your disc quite rapidly, thoughtfully. And then we make the systems that an image is a product only once, but can be shared in the. In a lot of different projects, and so we manage the right to the access to the image, so in Cytomine, by default, the image belongs to the users which have uploaded to it because there is absolutely nothing public within the Cytomine.
The assumption that we made it. Sensitive data. Once again, it’s data that has used in research in hospitals. It might be related to patients, so it’s sensitive data. So that responsibility to have an image into the platform belongs to the users that have added it, and only him can add this. Image into different project to allows other people to see them. [00:26:24] So that’s what we call an image management system. It’s to manage the right of the access and who can put this image in project or put it out of this project, and so on.
But something which is really important that never ever we transform this image. So everything which is done with inside mine is just stored at external data within database. So we never, ever transform the image. So if you publish an image into Cytomine and you come back six year later, you can build it back again. It’ll be exactly the same.
Aleksandra: Great. So this is, a huge, like, huge requirement both for medical and veterinary pathology, especially for toxicologic pathology, that nothing can happen to this image, there cannot be any modification if you’re actually evaluating your, this, for that diagnosis and all the. Manipulation in terms of image analysis, mask, whatever you need to do for domain shift robustness.
This has to be like [00:27:24] separate from the original one. So you guys have discovered and, okay. So you have been chosen as that provider, that software provider and that tool provider. No, I forgot to ask you again. What are the, all the things that this can do? So you can do annotations, you can, build AI tools. You can deploy AI tools.
Gregoire: Yep.
Aleksandra: You can manage images.
Gregoire: You can manage images, you can big application, which has, for example, data collections and data applications. So let’s say you have a project where you have, six thousand slides on which a lot of people have annotated for toxicology studies, for example. So, how many annotation I have with this category, how many annotation has been made within this time slot by this user from this location? Because main advantage, if you are, for example, a pharma company, you will have your, testing in toxicology with rodents and [00:28:24] so on in, in some countries, and maybe the scientist in another one, and maybe the drug development team in another country or another location.
Everyone can access to the same platform and the data are shared to, to everyone without having to move anywhere. So the slide remains on close, to the laboratory stuff, and every can work on them and do some computational, some statistical, et cetera, et cetera. So in Cytomine you can view, you can annotate, you can run applications, and this application can have AI inside.
And there is no, limitations in terms of technology of AI because, the strategies that we have. You can do your AI from anywhere, in the world. That can be get connected through internet to your Cytomine, because then you can choose your technology. You can link into your GPU for training and et cetera.
And if you want it to be plugged into Cytomine, it’ll be isolated in Doctor containers. So you can put the technology you want, you have the insurance that there is no conflict with the other application within the solution because it’s isolated with Doctor [00:29:24] containers. .
Aleksandra: Fantastic. This sounds really cool. So now back to big picture. So I already talked about big picture on several occasions, but just to recap this, a big picture is a private public consortium in European Union, and the goal of this consortium is to build a huge. Image repository, both for a clinical, human pathology and for veterinary pathology. Mostly for toxicologic pathology to make a huge repository of slides where people can develop AI solutions on.
Gregoire: Yep.
Aleksandra: Or like the different stuff. But basically to have a huge repository and you guys have been chosen as the software provider. As the software, all this stuff is gonna be happening. So, like, why did they choose you? How did this happen and what are you gonna be doing within this big picture?
Gregoire: So why do we have been chosen? [00:30:24] Maybe we’ll have to ask to the management board of Big Picture. No, it’s a, it’s mostly
Aleksandra: I will do that.
Gregoire: Yeah. Mostly because I guess, we are server site and web based. And this is definitely an asset when you are 46 or 47 participants, , spread over Europe, contributing from slides, private, contributing from AI, et cetera. I guess that’s also because we can, you can run your AI by connect your scripts to the solutions without any complication and publish them after your AI was in the solution quite.
And also you can manage unlimited size of image because as it’s service sign, the limitation of number of image is only related to the size of your architecture behind. So I just have to correct that we are not the only solutions into big picture for sure. So we have one of the tools that has been chosen, and specifically the tool which has been chosen for viewing, annotate and run [00:31:24] AI by behind us.
There is a lot of tools that has been put in place to store this petabyte of data, to manage the repository, to manage the permission of each other, who has the right , to contribute, by giving some slide, by giving some metadata, by running some AI by giving some code to read the image and so on. So Cytomine is just a piece of a big puzzle.
Aleksandra: So, like you guys like are the hub that everybody else connects to and this then.
Gregoire: It’s Cytomine will base of platform that will be launched when you ask to see a Dataset. So for example, you want to develop an AI on, I don’t know, and, you want to, you go, you get connected to the repository.
You see, that within this repository there is a data set of image that may be interesting for your research. You ask to the. Consortium can I have an access to dataset? And they will say is, they say, okay, they will deploy your Cytomine just for you and within this Cytomine you will have access to [00:32:24] this image and you will be able to annotate them and to plug your AI to them and make your training and make your inference and.
At whatever you need. And at the end you will have the choice to say, okay, everything what I’ve done, I want to share it with the rest of the, ,big picture community. And you will have, you’ll be able to use Cytomine as a subscription tool. So now every annotation I’ve made, I want them to be, added to the repository of big picture.
So it’ll be a tool to see the data, to create some addition of data like annotation and classification and comments and whatever, and then push back them to the community of big picture.
Aleksandra: So, know Big Picture has a work package on building the tools, not only using the ones that are already built. What do you guys need to still do or are you ready as you are or what are.
Gregoire: We are never ready. There is always, [00:33:24] so, first of all you know that there is, let’s say with within brackets a. In the digital pathology against the image format. So each scan is vendors as its own for specific? Yeah. I will not take any position on any format, but there is a reason behind them.
Some are targeting on L hospitals or they claim okay it will be, some seconds to have an image scans. Some say, no. We are the best image in the world. We’re targeting research. So the scan will spend two hours if it’s necessary, but it’ll be the best image in the world. Some are saying, no, we want to have Z stack.
Some there is a lot of reasons who have a lot of different formats, each can, vendors as is own specialty, et cetera. But it, within the project, it’ll be too heavy to start to manage all these formats. it, there is a claim that every kind of image that will be submitted will be transformed to dicom.
So we had to meet the dicom support for Cytomine, that every kind of AI [00:34:24] application will have to respect this kind of format. To be able to be uploaded in some way. It would buy quite close of the one we already use in Cytomine. But to be able to enlarge the possibilities, we’ll need to have some extra features.
We’ll also have some requirements in terms of way of showing the results, it maps and et cetera, extract. So you have to expand, the collections of a way of showing the results of an AI, depending of the eyes that been, that have been built using this project etc. And also with the collaboration with the team dedicated to that to be able to support the metadata model that has been specified by this community.
And also is a meta, is a storage systems that has been built within this community. Because we are used to deploy Cytomine for our customers on the own IT facility with our rules of storage. But within, picture with the numbers [00:35:24] of, partners that we have, we also have to take into account the own restriction, limitations, expectations, and so on.
So we will have to have to make some adjustment on how does Cytomine manage so big volume of image, so AI annotations permissions. It’s because the permissions will not have to be defined by our own platform. We will have to mimic the permissions that has been ordered and given by the big picture consortium to say, give access to these people to these data with this permission, and do not give them the permission to run AI, for example, or do not permission et cetera et cetera.
So there is a lot of, it’s not building new features, but it’s a lot of adjustment to make sure that the, solution at the end fits in need of everyone within this project.
Aleksandra: So it looks like there is no way around dicom as much as we, maybe some members of our community wanna defend themselves from dicom.
And like you say, there was a reason behind every format. I just hosted, recently at [00:36:24] chat with David Clooney who is the maintainer of the dicom format. And we had an extensive chat about this dicom for pathology because radiology there is, that’s a no brainer, but in pathology several people actually in the context of Big Picture, are looking into it, are looking also how to support this on their end if they wanna serve the pharma customers or the hospitals that are being part of this project.
Okay. So good. No way around dicom. So let’s figure out how to, work with dicom.
Gregoire: Yeah, but I guess it’s also, a pain that this, by the hospital sector because within the medical imagery it’s used to work with dicom since now, more than 20 years ago. And they have all the infrastructure already.
They have the culture of working with this format and this format has a lot of advantages. But yes, we have to phase a situation of one pathology images is very much bigger than the medic medical imaging one in terms of [00:37:24] size, in pixels, on the weight, and everything. So, sometimes I make a simulation and we can say that for one slide you sometimes replace 2000 x-rays within the pack servers.
Aleksandra: Yeah. I always.
Gregoire: The difference is really huge you know? So it means that the hospitals will have to manage not only the technology but also the storage. And so the technical revolution there is a long debate, but definitely we have seen that more and more people are speaking about dicom within digital pathology and mostly because sometimes they’re fed up about having so much different formats coming from so much different ways of acquiring media and so on, which has impact on how they store them.
The, yeah and to make them available for all the pathologists when there is a need for a patient care.
Aleksandra: So I know this very much from the computational pathology community where like a lot of resources is spent upfront on managing the [00:38:24] formats or on like, writing scripts to convert one to the other so that they are all the same and that you can deploy something that actually will have some value for the patients in the long run.
And they are fed up with that. They’re like, we can spend those resources better. We can spend those resources actually, On the end that has the potential to change something in the patient outcomes.
Gregoire: Yes.
Aleksandra: Or make stuff more efficient. We don’t wanna spend time managing your image formats.
Gregoire: And you have this struggle within the digital pathology workflow but when we you think a second about AI development we in the community, we can see that the, there were, there was a fear against AI a few years ago, but now what I felt when I going to into conference is that I see pathologies that really understood that AI can be a companion of with within their daily work. But if we want to have powerful AI, we’ll need to train them on the large [00:39:24] collection of image and you know that.
But if everyone can share the image, have images in different formats with different characteristics and that your AI tool within python and Script doesn’t have a free reader of a free way to read these images.
We will not be able to transit this AI so the problem there is we might face some AI only trained on these specific formats and so linking the way another to a format of image and scientifically, people will prefer to open the range of the different data sets that will be able to use to train the AI or on which they will apply the AI for better care.
So, it’s a part, it’s a debate. It’s a long debate. It’s not the role of Cytomine to take position in it. But what we’ve seen is that there is a lot more and more discussion within the community to go in the direction of dicom. And within the Big Picture project, it has been a decision that every slide [00:40:24] stored in the repository must be dicom.
Aleksandra: I love the name of their tool “The Dicomizer”, i love the dicomizer.
Yeah. So yeah, you say, it’s not your role to, you know express your opinion, but you definitely have to adjust to all the requirements.
Gregoire: Sure.
Aleksandra: And what you have been telling me are the requirements of a massive project. It’s like a six year, or a seven year.
Gregoire: Six year.
Aleksandra: Six year project that has like millions of investment, many many stakeholders, so I am super curious.
Gregoire: Yeah. It’s a really, a great experience within this project is very enthusiastic.
Aleksandra: Yes. Okay. So we need to schedule a demo because I see like for me, for example, if I wanted to do annotations, I have to just like open a browser and go to this repository where I [00:41:24] should do annotations on, right?
Gregoire: Right. So the first step, you need to have a Cytomine install up and running, and as soon as it’s up and running, you just opened your browser, you put the url of the Cytomine, you have been provided to, and you put your login and password and you can to work.
Aleksandra: So can you like, can I get this ? we can take it offline, but I would like to, I would, let’s see maybe we can do some test thing or whatever.
Gregoire: No, indeed. We have a demo instance in the company so people are free to ask for a demo account and we will provide them a demo account on which again, get connected and tries a solution and to see if the solution fits their needs because you know, it’s cover the community of image analysis.
And when I say image analysis, it includes analysis by humans on image. It’s not only computation and analysis it’s covers some so many topics in, into [00:42:24] biology, into cytology, histology, pathology, et cetera. So the best way is to test it. So if you wanted to have your own, you can follow the documentation.
So you can go to doc.title.org and you have a freely accessible documentation with the user guide, how to install it and et cetera. How to develop your own python script.
documentation.Cytomine.org and, but you can also, if you don’t have the skills or don’t have the time or whatever, you can ask to the company with info@Cytomine.com and please provide me a demo, more account because I want to do this and this and test if you are compliant with this and this. And I will create your account and you can, you’ll be able to test a solution.
Aleksandra: Okay, so maybe I’m gonna do that. I’m gonna send this email and the next time we meet and you like, walk me through everything. Let’s see if we could do that because definitely people would wanna see it, see what the capabilities are and it’s just seems like a very good tool for an [00:43:24] institution, like a multi-site institution. And I think a lot of institutions are facing the challenge, okay, this site has this tool, but this site invested in a different tool. How can we meet, how can we work together?
Gregoire: Yes. And the biggest experiment we have made in this topic is University of Liège, and now as is launching a few days ago, the 12 edition of , so a massive open and light course on histology.
The first edition were in french. Now, this actual additions, which is on, open is also in English. And there we have from the different course, it’s thousand users spread as world that can get connected and explore the slides, make annotations, follow the exercise, and see the video from the teachers and so on.
So they can open the microscope, vitro microscope Cytomine whenever they want, from whenever they want in the world. And so yes, for a multi located company or you see that you, we are [00:44:24] seeing that more and more hospitals are joining force to make consortium it’s also the same in research if you want to have a grant for research, for example, it’s quite impossible to have it if you don’t collaborate with distant organizations. And more and more they ask you to have collaboration with university outside of your own country, , et cetera, cetera.
Aleksandra: You have like referred countries.
Gregoire: Yeah.
Aleksandra: If some partnerships from less privileged countries. Then you get extra points, and then you have.
Gregoire: Let’s say for example, we have a project where the university hospital here in the edge will provide the slide, but a computational part will be taken by a university. For example, MIT, they can work in the same.
Platform because it’s service side and the computational, developers can do their work within Python using the Python library, and the pathologists can open the web front and whenever they want, from whenever they were, they want at home or during congress or whatever to see the image, to annotate, to make the review and so on.
So yes, it’s a main advantage to be web [00:45:24] based. You can work from everywhere.
Aleksandra: I wanna test it. I wanna test it. I have this one video on youtube, like how to annotate an image scope and I wanna do a similar one. How to annotate in Cytomine it’s like, for example, five minute video or 10 minute.
Gregoire: You’re welcome.
Aleksandra: A lot of people are , looking it up. So I definitely wanna explore. Okay. Next step, we schedule a demo and I let everybody know that there is gonna be a demo. Thank you so much for joining me Gregoire.
Gregoire: You’re welcome. Thank you so very much to, to give us this opportunity to speak with you!
And it was a real pleasure. Before, during, and I guess after this podcast, you are really a shining people.
Aleksandra: Oh my goodness. Thank you so much for beautiful words about me. But we’re gonna be in touch cuz I think your tool is cool and I wanna explore it more.
Gregoire: Your welcome!
Aleksandra: Okay. Thank you very much.
Gregoire: You’re welcome. Thank you very much.
