Visiopharm Tech Talks Session 1

I think we are live now.

Let’s just wait a little bit for people to join.

All right. Good morning then.

Good morning.

And good afternoon, evening to everyone that is in different time zones today.

And I want to welcome everybody who is here for the first Visiopharm Tech Talk. So, this is a series of webinars that is going to happen every quarter and it’s designed for you guys to learn what is new in the software, new updates, learn tips and tricks with our experts. And also this is a good space for you to ask questions.

So with me today I have our technical sales specialist Dan Winkowski, who has a lot of experience with the software and can answer all the questions that you have.

And his presentation is going to take around twenty minutes and then you have a chance to ask questions after that.

But of course you are welcome to write as you go in the chat.

We did receive a lot of questions when you subscribed for this talk, and Dan is going to try to cover as many of Dan during his presentation.

But feel free to ask the questions again in the chat if you didn’t get them covered in the talk.

What else I have to say? I have a question for you actually. Are you subscribed to Visiopharm’s newsletter?

Because I think a lot of people were not and you are not receiving information about what is new, about events like these, webinars and so on.

And we want everybody to get access to that. So there’s a link here popping on the screen now, feel free to subscribe and yeah, you also get information about new product releases and yeah, everything that we are doing for you.

With that we can start with some polls today.

We just want to learn a little bit more about who you are and what you are doing.

So let me just put on the screen now.

First question: What best describes your study area?

I’m gonna give you a minute to answer that.

Alright. I think we can move on to the next one.

Which area best reflects the majority of your work?

Okay. Next one.

Are you aware of the quick start ups?

Let’s see if you guys are subscribed to the newsletter because we talk about them quite a lot there.

Okay, well, good news is that Dan is talking about this today. So for the ones of you who haven’t heard, you are in good hands. And now final question.

Have you clicked on the what’s new icon in the top banner of the software?

That’s a new thing that we launched. I don’t know when then, but, yeah, think people don’t really didn’t see that.

It’s pretty recent.

Alright. With that, we can move on to the presentation, I guess.

Thanks a lot. See you in yeah. In twenty minutes.

Welcome everybody. For today’s session, we’re going to touch on a few topics. I’m going to start by highlighting recent platform innovations and updates. Next, I’ll demonstrate how you might be able to work more efficiently in the platform by describing some tips and tricks.

Later, I’ll reveal a coming soon feature that’s expected in the release later this year.

And lastly, we’ll wrap up with a live software demo. We have a lot to cover, so let’s get started. First up is innovations and updates.

Our developers have been working behind the scenes to make AI training on our platform essentially bigger, better, and faster.

Some of the core platform improvements that they’ve made have resulted in increased training speeds of up to 5x. This improvement is in part due to the fact that we’ve added support for the latest hardware, specifically the latest and greatest GPUs on the market. Think that five thousand series graphics cards from Nvidia, Blackwell Architecture, those types of units we support them currently. So that’s a big improvement.

Another improvement in our AI training is that you can now combine multiple channels as input to AI models. This is really helpful in fluorescence where the colors are already separated for you by the instrument, but it also works on color deconvolutions. And together we feel that these improvements will produce more robust generalizable models much more quickly than before.

So you might be asking, how do I find all this information? We’ve added a What’s New button in the ribbon that points you to a video description and details of what the improvements have been made in the platform. There’s a lot more in each release, so I encourage you to visit the What’s New button and plug into the full release notes on your support portal for all the details.

Next up, we’ve added a new improved drawing tool. This redesigned toolbar makes annotating easier than ever, fully supported by user requested keyboard shortcuts. We’ve replaced the old drawing wheel with a conventional toolbar that’s visible by default. It’s also supported by keyboard shortcuts, as I just mentioned.

It also works in both bright and dark modes within the platform. And overall, it’s just a much more intuitive experience for all users. And this matters for at least a couple of reasons. First, when you’re annotating training data for AI models, you want speed and precision and this design provides that.

Second, when you’re selecting regions for downstream analysis, you want the tools to stay out of your way. And this toolbar lives right down the bottom of the viewer window with has a relatively small footprint. And so it’s really something that it doesn’t overwhelm your workflow and really provides a lot of benefit and value.

So that’s the new and improved drawing tool. It’s faster, more flexible. We’ve been receiving a lot of good feedback from our users that have access to it. And so I encourage you to get it in the latest release.

Now drawing annotations still takes time. What I’d like to show you now are some tools that we offer that automate that tedious part of annotating. And for a long time we’ve been able to combine images of different staging modalities to assist you in creating annotations. And this is a simple to use workflow that we refer to as Tissue Align. Now Tissue Align allows you to combine any image modalities into a single dataset to enhance your experimental options.

And now here’s where things can get interesting. With tissue align you can take for example an H and E and a multiplex IF image or any combination of modalities and align them directly in the software. This could be images with different resolutions, different stains, none of that’s an issue in our platform. Once you align them you can now fuse the images to create a true multimodal dataset and that’s what you’re seeing here on the right.

Here I’m displaying the cytokeratin channel overlaid on the H and E stains section and this can help facilitate segmenting tissue areas using chemical stains rather than morphological features alone. And as an added benefit for teams working with pathologists, Tissue Align can be very powerful. The pathologist is very comfortable with the H and E and they can make annotations on it. It’s a stain that they trust and those annotations can then be carried over or passed through to your fluorescence data automatically.

Similarly, as I mentioned on the previous slide you can use the chemical stains on the fluorescence data to guide your annotations of pests into your H and E. It’s really flexible, right? So you can use it as your transfer. You can use this bidirectional approach depending upon where you’re comfortable whether your pathologist is comfortable with H and E or more comfortable in the multiplex fluorescence space.

But basically using it as a transfer learning protocol and we’ll talk more about that in the upcoming slides. And this concept can also work at the cellular level. So here we have an H and E stained tissue and the question here is a pretty challenging one. Can you identify macrophages in this image?

And for most people lacking advanced training this is a real challenge. But what if you had serially stained the tissue and had stained for CD 68? You could find all the cells in the sample, identify specifically the macrophages based on the chemical stain and then transfer that labeled data to the H and E and train the AI. This is a really effective way to have a lot of confidence in your annotations and know that you’re providing high quality data for the AI to learn from.

Now you don’t have to take my word for it. We have publications that document this and here I’m highlighting just one of them. In this case the authors used a single stain IHC to identify specific cell types of interest and those annotations were then transferred to the H and E image and used to train a classifier using only H and E. This works really well and can translate to a wide range of applications as you might imagine, but really in the end what it does it builds your confidence in your training data and gives the AI the most clear picture of what the objects are that it should be looking for.

You might be sitting there and thinking well this is all really interesting but I’m new to the platform and how do I get started using it and I want to get started really quickly, I’m really excited about it. We thought about that and we’ve recognized that there are sets of image analysis tasks that are extremely common among our users and as a result each user should not need to reinvent the wheel to construct their workflow. So we’ve created a family of apps that provide an excellent starting point for any user. We call these quick start apps.

These apps are trained by our team and trained on a wide range of tissues and stating variation so that they are as robust as possible as an out of the box solution and can produce data immediately. As with all things Visiopharm, they are modifiable, trainable and customizable for your specific needs. In the event that the app underperforms for you, you can add to the training using your data. This typically takes minutes, not hours or days, and then use it in your workflow, Right?

If you need additional numerical information, you can set up a new output variable that’s specific for your project. The apps are available now and represent a wide range of applications. We continue to add to the Quickstart app family with a broader range of applications, but our hope at this point is that we’ve covered all the major bases and that these apps will allow users to hit the ground running within our platform with as little delay as possible. The app family covers standard applications, as well as IF applications and imaging mass cytometry use cases as well.

So I’d like to point out though at this point, a new member of the Quick Start app family, which is H and E tumor detection.

And this app, although it’s relatively new, comes with the software, no extra charge and it’s been trained across a wide range of tissue and staining conditions and that you can see here as shown with the breadth of tissues and stains that it performs well on. So the idea here is you can get started right away with designing your workflows and generating data that you’re looking for. Check it out and let us know how it works for you. We’re always open to your feedback and looking to hear how things are performing on real world data. So please keep us posted.

So last up, I’d like to transition to a preview of what’s coming soon. Everything I’ve shown you so far is available today or shipping very soon. I’d like to now give you a sense of where we’re headed. So our developers are working hard to improve our unsupervised analysis methods and specifically enhancing our clustering and visualization tools.

And what I’m showing here is a mock up of our hierarchical clustering with an interactive dendrogram on the left and a heat map and gallery views on the right. Together these tools can be super helpful as you might imagine for like cross multi omic investigations like linking protein and transcript data at the single cell level. We feel these approaches will provide enormous value in this space when you’re trying to make sense of super complex data. So stay tuned for new developments on this front and for any announcements for when it will be released.

Remember to check that What’s New button in the software. So that’s for now the product update. Now I’d like to switch over to some of these topics live in the software. Okay, so the first topic I’d like to demonstrate to you is our multimodal co registration capabilities and highlighting our image fusing feature.

So here we have two pieces of tissue that are serial sections, one stained with H and E, another is multiplex IF. Using tissue align we can bring these two images together with the click of a button. That button is right here, this auto align feature. And once these images come together, you’re prompted by the software whether you wanna fuse them or keep them as two different aligned layers.

That’s up to you and depends on your ultimate goals with the alignment. In the case here, I have fused them. And what we see when I click on this image, this aligned image, is I see a channel list over here that is a fusion of those two panels. A three plex on the IF image and then the red, green, and blue for the HNA.

So once the image is saved, can execute your workflow on the image using the color channels that are best suited for the task. It’s all there for you to use. And here is what that looks like. And so what I have here is one of the channels from the IF image turned on, and you can see if I toggle those on and off, we’re looking at the immune cells in the image and both from the IF and on the HNA.

And you can see there’s a nice concordance between them. And so you can use this fluorescence component to the image and use it to guide annotations for training your AI or for your analytical needs and feeling confident in the results that you’re getting. So that’s our image fusion capabilities. Next up, I’d like to refer back to one of those training improvements that I mentioned earlier in one of my slides.

So training on multiple channels and that includes color deconvolutions. Now suppose we wanted to create a superclass of membrane bound markers like CD4, CD8 and maybe some others and train AI to recognize these and improve the cell segmentation using biology rather than estimations of cell boundaries. Now you can do that in the software and here’s how. So we start off by creating a new app in which we would then go to our classification section and our advanced menu where we can then define the inputs that are going to be used for the training.

So if we click here we get this window and you can see here there’s a long list of the markers in the panel that match the assay over here on the right. As I said, we want to combine some channels here. So let’s start off by deleting all of them. We don’t need all of them.

And first we’ll include DAPI in there, but then for our next input maybe we want to combine our T cell markers, so CD3, CD4, and CD8. And maybe for good measure, and you can see all three of them as a single input, and maybe we also want to do something with our macrophages, so CD68 and CD163. And so here we’re combining those macrophage markers, here we’re combining those T cell markers and all of which will end up when we create our annotations and hit the train button, a more robust input to the AI so that it can learn more effectively what those classes are. Next up, I’d like to demonstrate a tip that I use very often using our Quick Start apps.

So what we have here is a small region that I’ve annotated and what I’ve done is I’ve run our nuclear detection app on it for Brightfield. And you can see here these blue labels show all those nuclei that are identified.

But really what I want to do is train the classifier to recognize tumor nuclei here and differentiate them from stromal nuclei. So I want to train a classifier to differentiate between those things. So how do we do that? So we can click on new app again and we’ll get a new dialogue here where we can set up some classes for training.

First class is always background, second class we’ll say is stroma nuclei, next class will be not surprising tumor nuclei. And then from our drawing tool here down at the bottom we can say let’s click on tumor and holding the control button down we can then change all these label colors from blue to red. And then the AI, once we hit the finish this task and feed it to the AI, will recognize these as tumor nuclei. And then we can say, let’s do with enough data and with enough training time, we can then push this train button and the AI will learn the rules that differentiate tumor nuclei from stromal nuclei and then we’ll have our classifier that does that for us.

Pretty quick, pretty easy, not a lot of annotations. You can imagine the time it would take to create all the annotations for these nuclei without this quick start app. So it’s really beneficial to your workflow in that way. So final thing is we talked about transfer learning and gave a couple examples in the slideshow.

What I’d like to do is just show you what that looks like here in the software. So again, we’ve taken our image and we’ve let’s turn these off taken our H and E image, the first three channels are red, green and blue to create that imagery that’s on the screen here. And then we also have our HyPlex panel here that’s listed with all the markers with metabolic markers markers and activation markers, functional markers. But what I have here is also CD68.

This is a famed how do you detect the macrophage markers. And if we go into a particular area you can see first we’re doing a pretty good job the alignment where we have outlines of those fluorescent markers around the cells and good concordance with the nuclei and the H and E. But here we can also just turn on these labels and create an outline that says here are the nuclei and the pink ones are the ones that are CD68 positive and use that information to then train a new AI that is not tumor stromal nuclei. We’ll just click new app here again, not save this.

And here in our section here we’ll delete all the inputs and add in the red, green, and blue. So red, green, and blue.

And then use that training information on the H and E alone so that it can then perform that task on just the H and E stain sections. This is pretty cool way to utilize the combination of H and E and high plex immunofluorescence and really give you the benefit of all the information that’s available in these combined datasets. So that concludes our presentation for today. We’ve covered quite a bit Specifically, we’ve talked about platform innovations and updates and how to be more efficient in the platform by describing some tips and tricks.

We’ve also revealed a coming soon feature that’s expected in a release later this year. Finally, I just hope that today’s session has given you some practical insights and clearer understanding of how you can work effectively in the Visiopharm environment to really get to your endpoints and your results much faster and easier than you might have before today’s session. So look forward to the Q and A and yeah, hope you enjoyed it. Thank you.

Hayden, thank you so much.

Is a lot of new cool things going on. Thank you for sharing that with us.

We’ll start with some questions. I actually have some questions myself. I will start here with the first one.

Then if you had two highlights, one updates in latest release. What is the one that changes the game the most for people?

Yeah. I think for all the updates that I mentioned, the one that I would pick is the one that’s probably the most has the most value behind it is the drawing tool. Right? So it makes annotating really straightforward and really much easier than it was with the wheel.

It’s always present, you know, so users don’t have to figure out how to access our drawing tools. So now, admittedly, as a long time biz user with the drawing wheel, it did take some time getting used to it, but then transitioned away from the from right clicking to get access to those drawing tools. But now that I’ve recalibrated, it’s awesome. So, yeah, I would say the drawing tool.

Oh, cool. Thank you. So Jack Robertson is asking, what version does the GnomeWheel menu start?

It should be available now. So any of the current releases, I think, is the drawing tool should be should be accessible.

Yeah, it should be accessible today.

Okay. And then Agata, is there a cost to upgrade to this new version?

No. I think if you’re on an active subscription, you should be able to or, you know, SMS plan, you should be able to get this today.

Yes. We every time there’s a new release we send an email to all our customers and there is a link to download a new version.

Agatha, I’m not sure you are receiving these emails. We can connect with you later and make sure you get those.

So next question, Jamie, what version has all these updates?

As far as I know, the current version, if not, it’ll be released in August. So it should be end of summer.

Cool, pretty soon. And then Lourdes, serial sections are not exactly the same. Therefore most commonly the cells will not be exactly in the same place in two consecutive serial sections. How do you solve or recommend to solve this issue?

Yeah. That’s exactly right. Serial sections, by definition, could not contain all the same cells. The example I showed was, serial stains, so they do contain the same cells, that makes it a little bit easier.

But our tissue align technology does not need the same cells to be present on the on the two sections. So it doesn’t depend on on the exact same nuclei being present on the on the tube. So it’s really using the shape of the tissue and and morphology measures. So I think you should be okay with that. Of course, you know, if they’re staying present on one, then nucleus not present on another.

You can certainly define the rules that that you would follow for how you would treat situations like that.

But yeah, it is a challenge with serial sections for sure. Try to keep them as close together as possible would be my only recommendation, but it is a real issue.

Thank you. Shannon, is the new clustering going to be assessed in Phenoplex?

That’s a to be determined. It’s still in development. We’re still working through some of the of the details, so stay tuned. We’ll keep you posted on when that becomes available and where it resides.

Thank you.

You mentioned in the beginning of your presentation that we now have access to up to five times faster AI training. What does that actually mean for users?

Really, it translates to a boost in efficiency. And so in essence, what it’ll do is it’ll take less time for training AI, which translates into getting usable algorithms into a production ready state much faster, right? And so really in simple terms, AI workflows are gonna be developed faster and reduce time to results. So you just become much more efficient in that way.

Cool. You also showed automation in annotation. How much time users can actually save now?

Well I mean the answer there is really it depends on the complexity of what they’re trying to train their model to detect. It could be quite a bit if you use these quick start apps. So in the example I showed, annotating those individual nuclei could be really time consuming.

Running the app and changing the labels was really fast, It did it in just a minute or two. So in essence it could really translate into quite a bit of time that the users get back with harnessing the power of these quick start apps and then just making some minor changes.

So yeah, it could really be a benefit.

Great to hear.

Mirian, is it possible to somehow group pictures into one or a group of sorts? Say well plates for cells where you have three hundred separate pictures and analyze it as one?

Not so sure what I if I understand, but I think with well plates, we could probably if they if they’re already separated, we can analyze all those images. Or if you wanted to bring them together into a single image that’s stacked, I guess you could do that as well with tissue align.

But then if you wanted to recreate a whole slide image, I don’t know. Our software would not be able to do that today. So, yeah, hopefully I answered your question, but if you do, okay, great. Thanks.

Thanks Dan. And for someone who is new to the platform, what is the fastest way to go from zero to meaningful results?

Oh, that’s quick start ups for sure.

So they provide an excellent starting point for general analytical endpoints. Right? So they are robust and tend to perform well for most of the projects we’ve tested them on and our users have applied them on.

So it really reduces time to results, right? If you need better performance, as I mentioned, or additional information, you can always make those improvements.

And if you’re looking to save annotation time, as I also showed, you can use them as a starting point and create new content that may not be available in the quick start up family. But quick start apps are the way to go. Right? So that’s that’s my recommendation.

Thank you. And how do you get these apps? How is it on the latest release?

This comes in the share?

Or We’re I think we’re making that easier as well. But for now, I think you’re gonna have to go to your support portal, and there’ll be a download for your quick start ups. But that’s gonna be made easier in the somewhat near future where it’ll be right in the software.

Cool. Caleb, how many images can be aligned at one time with tissue align?

I think a personal record of mine is nine to ten. Right? So but if they’re serial sections, obviously, you lose information the further you go away in that stack. If they’re serial stains, then kudos to you for being able to do nine different washes. But, yeah, theoretically I think there’s no limit, but yeah, about nine or ten is what my personal record is.

Cool.

Miriam has another question. Can I let an analysis run and simultaneously annotate in a different picture not included in the running analysis?

Yes. Absolutely. Yep. So running analysis in the background is sort of what we refer to as a batch process does not prevent you from using the software in, you know, in doing whatever you wanna do in terms of annotating or or creating new content.

Problem.

Good and just so everybody knows this session is recorded. You can rewatch later and also if you have other questions feel free to reach out to us or to Dan directly. We will share the recording via email probably early next week and yeah I think we are almost ready.

How often these meetings will occur? So we are planning to do this every quarter.

So the next one probably will be either in September or October.

So stay tuned to that.

And of course we have the August release which will come with a lot of information, we are recording some videos with our R and D team which they will go through everything that is new and you can watch and download the new version.

Rob, are there any sources of tutorials already available?

Yeah. So we do have a we refer to as a learning management system. It’s a self paced system that you can go through some videos and and tutorials.

The way you can get access to that, if you don’t have access to it already, would probably be to email, send an email to supportphysiopharm dot com requesting access to the LMS. So that would be your best way to do that.

Thank you. Emmanuel, is there a tool being developed for multiple length thickness measurements for elongated labels?

Say I want five equally spaced thickness measurements for a banana shaped object.

Sure.

So the answer is yes. We’re aware that we’ve that users want would like to do this, and we are working on ways in which we can make those measurements in a in a more regulated manner than what we do today. Yep. So it’s coming soon.

Great. Boo, I have been on LMS. Are there any other videos available that are not posted there?

Not I don’t control the content there, but we can sorta if you send an email we can sort of talk about what kinds of content you want, we’re looking for and maybe post that. We’re updating the LMS and the videos all the time and But if there’s something specific, let us know and maybe we can create some content for you.

Cool.

Lovely engagement. Yeah. This is great.

Yeah. And, yeah, thanks everyone for joining and for engaging with us. Thanks, Dan, for all the nice insights you shared here today, and we look forward to the next session.

So keep looking at your email for the recording. Don’t forget to sign up to the newsletter to get news and see you next time. Thanks everyone.

Bye.