DPP sponsors:                            

How to approach colon cancer with supervised deep learning image analysis w/ Rish Pai, Mayo Clinic

Subscribe via RSSGoogle PodcastsApple PodcastsSpotifyStitcherDownloadPlayer EmbedShare Leave a ReviewListen in a New WindowSoundCloudSubscribe on Android

How to approach colon cancer with supervised deep learning image analysis w/ Rish Pai, Mayo Clinic

This episode is brought to you by Aiforia. Thank you Aiforia 🙂

Today you will learn how Raish Pai, MD, a busy, practicing pathologist from Mayo Clinic developed a complex supervised deep learning tissue image analysis model to quantify visual diagnostic features of colon cancer and in the process developed a model that can predict clinical outcome.

He used the deep learning-based tissue image analysis platform – Aiforia.

The quantified features included:

  • Stromal immune cell Infiltrates
  • Immature stroma
  • Tumor-Infiltrating Lymphocytes
  • Mucin
  • Different growth patterns
  • & many others

Watch on YouTube

This Episode’s resources

digital pathology resources

Transcript

Aleksandra: [00:01:14] Welcome my digital pathology trailblazers!

This episode is brought to you by Aiforia. Aiforia is a deep learning-based image analysis platform for pathology and tissue image analysis so thank you very much for bringing this episode to us and today my guest is Dr. Rish Pai. So let’s dive into it!

Welcome to the podcast, everyone. Welcome Rish. How are you today?

Rish: Great. Thanks Aleksandra for having me.

Aleksandra: So Rish is an MD pathologist and he is a practicing pathologist. He’s also doing digital pathology, but not only digital pathology. He’s doing image analysis and usually. Practicing pathologists and image analysis, it does not really go together. Usually you do image analysis in a research setting, and I’m gonna ask you about that, Rish.

But before we dive into it, let our audience know. Let our digital pathology trail blazers. Know who you [00:02:14] are and what was your first encounter with digital pathology and what did you think about it then?

Rish: Sure, yeah. So I’m a practicing pathologist. Like you said, in my specialty is GI and liver pathology, but I also do, general surgical pathology. I’ve been in practice for since 2009. So you know, a little bit of time.

Aleksandra: That’s when I graduated vet school.

Rish: Yeah, so some years under my belt. Um, yeah, I’ve always been interested in digital pathology and I have used it mainly in a research setting, kind of viewing, slides, working with other groups. Just basically reading slides digitally, not really doing much image analysis until, I would say two years ago when I thought about this project of using image analysis for colon cancer and some of the newer techniques we have now with neural networks, and deep learning, and tools that are a little bit easier for a practicing pathologist to use rather than, needing, a PhD in, computational tools. I [00:03:14] could use it just being a regular pathologist. So yeah, I’ve been doing, viewing slides, digitally for, over seven years. But, so I’m familiar with that aspect but the image analysis side has only been in the past couple of years.

Aleksandra: So how did this start?

When did you, how did you even find time as a practicing pathologist to do image analysis? Like what were the circumstances?

Rish: Yeah, a lot of bad things about the Covid Pandemic for sure. There was, a period of time where, the volume of material we were getting in the hospitals was a little bit lower, so there was more time, for pathologists, kind of, in April, May of 2020, when everything was really focused, on the Covid Pandemic.

So, during that time, I, thought, Well, let me see. I needed to I felt like I needed to embrace what was coming and what was coming was the digital revolution of pathology. Not just viewing, and reading slides remotely and digitally or, but using the tools that are available, because the [00:04:14] slides are now digitized.

So, that’s what I had a little bit more time to think about, this project.

Aleksandra: I’m laughing because I’m like how was that? You woke up one day and decided today I’m gonna embrace that digital revolution.

Rish: It was, it was a very spur, spur of the moment thinking, Ah, what the heck? Let me give it a try, your expectations are super low, but then you start doing it, you get a little excited. You see the potential, um, which I wanna encourage other pathologists like me, who don’t have a background, in, data science.

There is options , for for us to dabble in this. Not an expert by any means, but I like to dabble in it.

Aleksandra: So tell me about the project, Like why did you think image analysis is gonna be good for your project, and what were the tools that you were using, and also what does your algorithm do? I’m interested in that as well.

Rish: Yeah, so, my interest is GI and liver pathology and, a lot of, interest, in multiple aspects of that field. But colon cancer is something [00:05:14] that is kind of bread and butter to GI pathologists. It’s a really common cancer and we already know a lot about it. We know the features that are important in colon cancer.

But what I find, what I found when I was a pathologist is just, it’s time consuming and difficult to extract those features to, to hunt for those features, and there’s some features that we don’t currently report that we know are important, but they’re just too tedious. And so the overall hypothesis was for me, can we use digital image analysis to extract those features in a quantitative way with the hypothesis is just better quantification of features that we already know are important

Like, you we know immune cells are important. We know tumor buds are important. We know the stromal ratio is important. Can we just extract those features in a quantitative way using image analysis? And does it correlate with outcomes, prognosis, recurrence, things like that. So it was actually a really simple hypothesis, just let’s just do what we already do, but in a quantitative way [00:06:14] and much less tedious, but of course training a model to extract those features is tedious, the annotations that you needed for a highly supervised method, which is what I use. So, um, I wanted to use, a deep learning method using neural networks. And one of the tools that I think

Aleksandra: Why did you wanna use that?

Rish: Yeah, you know,

Aleksandra: Before even diving into the tools? Huh.

Rish: Because it’s a powerful new method.

Um, I think, it’s nice to. You don’t predefine, like the certain, the pixel intensity, or the shape and size. You just circle the feature you want and then those, the features are kind of extracted and fit through a neural network to learn what it is that you’re seeing.

So, the traditional machine learning method of predefining, the parameters and, what does a lymphocyte look like and things like that. You, this was much easier in that you just. This is a lymphocyte. This is a lymphocyte. And then let the neural network know what features , make that a lymphocyte.[00:07:14]

So that, and it’s a powerful, technique for sure. Seems to be the future, for I think, really powerful image analysis tools. So, yeah, again, I don’t know I’m not a experts in neural networks. My superficial understanding is that it’s the way to go. It’s the technique that’s probably gonna be the most powerful and the one that will advance our field.

Aleksandra: And you’re the end user. So you’re speaking also from the end user perspective. I worked a lot with the classical computer vision methods, and what happens when you use those, um, handcrafted features, those, manually set thresholds, if it’s something pretty homogenous, it’s okay. And, you can, have good results.

But whenever it gets more complex, you always leave data on each side of the threshold. And you move the threshold to like optimize. You leave other part of data, other part or part of the image, you optimize for roundness of a lymphocyte. And those elongated that they’re moving or going somewhere they need to go, you don’t capture them.

So, um, that’s why deep learning I [00:08:14] think is gonna be the way to go as well. It probably, actually, I take it back. I think it’s gonna be a combination, um, of both approaches. But you wanted to go.

Rish: Certain methods, you certain methods, you don’t need the fancy, deep learning tools. You can just do it with, traditional handcrafted machine learning. Yeah.

Aleksandra: But you went for deep learning. Um, how did you evaluate your tools? What’s your, what was your tool and why did you choose this tool?

Rish: Yeah, so I used a Aforia, It’s a commercially available software, and it was something that we actually had at Mayo Clinic. A subscription too. We use it for education. It’s kind of a, an easy way to share, images, and if I’m giving talks somewhere else, I can pull up, the Aiforia hub, and share that.

But they also had the Aforia create tool, which, um, I didn’t know much about. And that was the meeting I set up way back when, to kind of go over what kind of tools are there. And it is an easy tool for pathologists to use. You don’t need to. You [00:09:14] know. It’s a non-coding. You don’t need to have any coding experience to, to do this.

And it’s basically a tool that allows you to annotate, um, regions that you want and you can do region detection, object detection, instant segmentation. Those kinds of things, you can build it in a layer structure so that first it identifies tissue, and then under tissue it identifies carcinoma and stroma.

Then when the, the stroma, you can have other things that you can identify underneath each layer. So that kind of made sense to me that that ability to do have a layered approach to the analysis. And, um, the ease at which you could annotate. And I ended up annotating over 24,000, um, regions and objects.

So, having a tool that was easy, , easy to do was really critical. Yeah.

Aleksandra: So what did you evaluate? You said that you visually evaluate certain aspects, certain regions, certain cells in colon cancer, but you don’t report. What did you quantify in your algorithm?[00:10:14]

Rish: Yeah, so, it started very straightforward. First let’s identify the carcinoma region and separate it from stroma. But then within the carcinoma region, we know that there’s various grades, solid growth versus grandular growth. So I used the tool to sub stratify those areas.

Signet ring cell cancer, kind of, get that out of and kind of quantify the percentage of signet ring cells, tumor buds, which is part of the tumor. It’s really small clusters of tumor cells that, that, that. Um, invade in our thought to represent epithelial-mesenchymal transition and bad outcomes. So that was another feature that I wanted to extract.

Um, necrosis, musin, those are features that we currently, we often see in colon cancer. We don’t really quantify those, so I wanted to do that. And then within the stroma, um, there’s some interesting data that the different types of stroma have different meanings. So if the stroma has a rich infiltrative immune cells, that pretends probably a better prognosis than it fits [00:11:14] immature stroma.

So I wanted to to identify that. And the last thing was tumor infiltrating lymphocytes. These lymphocytes that are infiltrating the tumor epithelial cells. So, again, these are all features that we see under the microscope every time we look at a colon cancer. But how can we extract the, that data, from the image is a problem with our eyes.

And we can only do it in a qualitative way, like tumor infiltrating lymphocyte high versus low. And how many people really want to count manually. So, this is a perfect, Yeah perfect tool. I thought that, that or perfect way to, to employ AI to help us do this. So that, those were the features that I wanted to extract.

Aleksandra: So how many features did you have in total? I’m asking because for each feature you have to like train with annotation for this feature and then you stack them on top of each other. So by what you already said, it was a lot of work to design this algorithm

Rish: yeah. 15 features in, in total. Um, yeah. Yeah.

Aleksandra: Respect, [00:12:14] respect. So you did this, it worked. What was the framework like? Where did you have those samples from? How did you have samples with data that you wanted to correlate to?

Rish: Yeah, it ended up being a little bit more complex than I wanted at the beginning. Um, so there’s the set of images that you use to build the algorithm and it ended up being close to 560 tumors, but I had them scanned in different scanners and I wanted it to be, generalizable.

So the number of images was a little higher than that. So that was the images that I used for training the model for annotating the model and once, once, I felt that it was performing well. We validated against gastrointestinal pathologists that that I knew who who were experts in this field.

And so we did that whole algorithm validation step, which is extremely important. You wanna make sure that it’s, when it’s detecting something, it’s detecting it that other pathologists agree. And [00:13:14] then, so that was step one.

Aleksandra: I do have a podcast episode. Sorry to interrupt you. I do have a episode with Thomas Westerling-Bui, who is actually from Aforia. He was talking to me about the algorithm validation, so I’m gonna link to this episode as well. But yeah.

Rish: Yes. Yeah. It’s a useful tool that they have to validate as well. So once the algorithm was built and I felt that it was performing well and we validated against GI Pathologists. I then wanted to take that algorithm and apply it to a large cohort of tumors with the goal of simply first. Simply like, okay, when the tumor is called high grade by the algorithm, there’s a lot of high grade, the pathology report said that tumor was high grade.

When the tumor was called mucinous, the pathology report said it was mucinous. Simple things like that. Just as further validation of the algorithm. And then we know, um, a lot of molecular details of colon cancer. A lot of them get testing for microsatellite [00:14:14] instability, KRAS mutations, BRAF mutations.

So I, I wanted to see can I identify differences among these different features in tumors that are microsatellite high versus microsatellite stable. So that was like another part, the most important part at the, it ended up being the most important part. Can I use these feature? To say that this tumor has a high risk of recurrence, this tumor has a low risk of recurrence.

That was kind of the holy grail. Did I expect and that it would be successful? I hoped it would be, I hope, better quantification of these features that we know are important would be used, can be used to predict prognosis. But I was really gratified to, to show that it does. So those were kind of the three things that I wanted to do once the algorithm was built.

So it ended up taking quite a lot of time to build the algorithm.

Aleksandra: How much time?

Rish: Yeah. Um, I would say to build and validate the algorithm, um, probably six months to a year I had something workable at six months. [00:15:14] Um, but, I wanted to make it, the, sometimes pathologists, like to be perfectionists, so we wanted to make it perfect, but that’s another important concept I think.

It really depends on what your end goal is. If you’re for an algorithm, for example, if you want an algorithm to make a diagnosis for you, so you don’t even have to look at the slide, the bar is astronomically high, right? You cannot miss a case. But if you want an algorithm like mine, like can it do what we, can it do better than what we currently do?

Then the bar’s a little lower and can it you know in some of these things we can’t even, we’re not even doing now. So the bar is just like, it’s not so high. Um, so that’s that. These prognostic algorithms that I’m, that I created, the bar isn’t as high as a diagnostic algorithm where you’re identifying cancer and making a diagnosis using ai.

Aleksandra: Yeah, definitely. I think the, industry is going away. So these questions, and [00:16:14] I think I pilot every other episode, these questions come up or will we lose our job? Because it’s gonna diagnose that whoever is involved in this, this is not the case anymore. We’re moving towards decision support system where we are still involved.

There is no like diagnosing for the pathologist anymore. Uh, I don’t actually know. There ever was like strictly, I think maybe this was something people were afraid of. Um, but yeah, definitely, you know, your involvement, you put a year of work into this. Um, this was, I know it was, um, in a framework of a grant that you had.

Are you gonna be using this now a year of your work? Are you gonna be using it? And how are you gonna be using it?

Rish: Yeah.

Aleksandra: I hope you will be.

Rish: Great question. Um, yeah, what was the end game? I really, it was more of a question, Can it do it and does it have value? And that’s step one, right? [00:17:14] And then step two is the hard one, clinical adoption, if it does have value. So I think I prove it has value. This algorithm, the features that we quantify, can better predict recurrence than what our current models do, um, using what we currently do.

So I think that’s, that was shown in, in, in the study. Clinical implementation is a whole different set of criteria and hurdles. Um, something I am not familiar with, but I think pathologists need to understand how to do it. Can it be done in kind of a laboratory developed test mechanism?

Not FDA approved. Kind of, I, I kind of envision this just like an immunohistochemistry test. We run this test, pathologist looks at the algorithm output. It doesn’t make sense, And then if it does, then it goes through the prognostic score that we can do through this.

And then a very simple report is issued to the medical oncologist who would say, Okay, this is a low risk tumor. , I can use that [00:18:14] little piece of data in my, in along with all the other data that I have to make a treatment decision. So we’re in the process of thinking about implementing it, but I think it’s challenges to that.

And, I’m very curious to see what happens in the next, few years as more of these tools become available. How are we implementing.

Aleksandra: What do you, what do you mean what happens? Are you like, are you moving in that direction or not quite yet. I’m asking because there’s, you have departments that are working on those proofs of concepts that end up as publications. And I know your work also ended up as a publication, and I’m gonna link to the publication in the show notes, but like in your organization, like you just said.

Okay, we did this and you have like gone beyond the normal visual validation. You took archive data, you correlated with the pathologist description. So you kind of did a little bit of multi-modality work where you extract information from the report that you know, what visual characteristics they [00:19:14] say they have.

And then you check, Oh, my algorithm is also, detecting this, like mucinous or um, whatever other attributes. So you did that. You did a lot. So I assume, well, I know you believe it, Your people believe it. What is the roadblock of moving towards you said you would go for an L D T, so something that’s lab restricted makes sense.

Rather than standardizing it for the whole world and, um, going the medical device route. But okay. Let’s say, yeah, it’s like IHC it has value. It can inform the treatment decision, what’s stopping you?

Rish: Yeah. What’s good question. I think we, we have interest in doing it clinically. And if I can talk a little bit about the Mayo Clinic Digital Pathology Program. It’s ambitious. Um, we started a few years back. Evaluating scanners and purchasing scanners. Now we’re in the process of digitizing the practice, so we are [00:20:14] scanning all of our slides and phase three, which we’re starting next year.

Is what you’re asking. Using tools like why did we digitize everything? Well, we digitize it because number one, workflow issues that, that, pathologists or different sites, you don’t need to ship slides. But the most important is, I think, um, using tools that enhance our practice. And AI is one of those tools.

So there is interest in using these tools. I think we’re. We’re in the process of figuring out how we can do it in a way that works. So, yeah I’m optimistic. I don’t know I mean, I’m not in charge of some of those decisions.

Aleksandra: The moment you implement, you’ll give me a call and I wanna interview again and ask you about the pathway.

Rish: How we did it?

Aleksandra: Exactly. Definitely.

Rish: Yeah. No I’m excited. I think the field of pathology is moving in an exciting direction and kind of what you’ve alluded to at the beginning, dooming gloom, they’re gonna be, we’re gonna be replaced. It’s. Definitely not that. It’s enhancing our practice, giving [00:21:14] us more tools, making our reports more useful.

I mean, how exciting is that? We, we can do stage and lift nodes and things like that, but now we can provide even more information. We, I think it’s gonna be a great and exciting time.

Aleksandra: Yeah, the pathology report goes out to the world, whether in a diagnostic setting where you know, everybody who has anything to do with treatment extracts information from there. Me or us as a pathologist, I don’t know what information is useful there for the people who are working with us down the road.

But the more information we can provide, the better. Also for, the treatment landscape, the treatment options, the drugs that are being developed are constantly changing. And those patients that, that, um, Their condition, their diagnosis was reported, I don’t know, a couple of years ago.

Could in a few years benefit from [00:22:14] the additional information. We don’t know that. So everybody kind of fishes for their information. The more there is, the better I would think.

Rish: Yeah. And there. Just thinking about that idea, um, there’s algorithms that are being developed and I think are super amazing using the weekly supervised approach using image tiles to predict something, right? You label the image tiles with. MMR, like these are MMR deficient tumors, and then you feed these tiles in and you know you can do that for, these are the tumors that do bad prognosis, and you can get a good model with that.

Another one, you can, these are the tumors that responded to certain therapy, and then you can build an AI model, right? Using that method. The method that I took was more of that highly supervised learning, right? Where I’m extracting features. The nice thing about the model that I think I, I developed is, Um, it’s agnostic to what you were looking at.

Like I can use those features to predict MMR. I can use those features to see if I can predict [00:23:14] outcomes. I can use these features to see if I can predict response to therapy. And so I think it’s gonna be a combination of the weekly supervised approach plus the highly supervised approach. Maybe together you can really come up and answer these questions that, that we want to know.

Will this patient, number one, do badly or, have a bad outcome? And then those are the ones we need to target with new therapies. Will they respond to that new therapy? Those kinds of things I think are really exciting and I’m hoping to test if this algorithm can predict patients who respond to, for example, immunotherapy.

Tumors that have a rich immune infiltrate, Um, maybe those are the ones we can target for immunotherapy. Um, yeah.

Aleksandra: Yeah. So currently there is an image analysis based test immunoscore, but it only quantifies, um, different types of lymphocytes. Your and staying with IHC, your algorithm was developed on H&E, right?

Rish: Yes.

Aleksandra: So [00:24:14] it has that advantage over immuno.

Rish: Ahahahaha.

It does require a decent H&E. But labs normally, their H&E are fairly good, but Yeah. Yeah.

Aleksandra: . So before we wrap up, I wanted to ask you about the grant and the framework that you work within about the, um, where those samples came from.

Rish: Oh, thank you for that. Um, so, I’m the principal investigator of the Mayo Clinic site for the Colon Cancer Family Registry. And that’s kind of, my interest in colon cancer is what led me to, to that grant. It’s a grant that’s been funded by the NIH for over 20 years and it has a wealth of data.

It has data on, over 9,000 patients, participants with colon cancer. And then it has data on family members and other people without colon cancer. So it’s this rich resource and so we, um, collected pathology material on these participants and we have tumor tissue and slides.

And so I was able to access our repository of [00:25:14] slides and digitize them. It ended up being over 6,000 cases. I ended up analyzing using this and the colon Cancer Family Registry was a, was an amazing resource that I tapped into. And this is a grant that’s funded by the NIH with the whole purpose of sharing data.

And so these images, um, are available with, it’s, um, we’re trying to figure out how we can make them available in a cloud-based format. So far that’s not been solved yet, but…

Aleksandra: But you wanna have them sharable?

Rish: Yeah.

Aleksandra: So let’s say if I wanted a couple work on an algorithm and have a hundred of colon cancer cases, can I like send you an email and ask, Oh, could I have hundred colon cases, images, or how does that work currently.

Rish: Yeah, so there’s, there’s, um, a website that we have a coloncfr.org, there’s a simple application process.

Aleksandra: Gonna be linking to it as well.

Rish: Yeah, we have a program coordinator that handles requests [00:26:14] for, from investigators to use our material. And there’s an agreement that, that we have that just states, you know what, you would include the Colon CFR in, in any publication that come out of it.

We wanna share resource. I think, I don’t know how many publications have come out of the colon.cfr data, but it’s over, it’s in the hundreds. I think it’sover 400 or so. So, I mean, we’ve shared our data a lot and that’s the whole purpose. We wanna share it. Um, there are, if we’re shipping hard drives and things like that, we need to figure out how that will work.

And unfortunately, I wish there was a better way us to.

Aleksandra: Let me, let me do one thing. Let me do a shout out to all the vendors that are dealing with this type of cloud computing cloud data sharing. If you guys have an idea how to share it. Then please contact Rish and, um, engage in those discussions.

Rish: That’s fantastic. That would be very helpful.

Aleksandra: Yeah, I mean, there are smart people [00:27:14] who deal with large data, so I hope they can help you. This is similar, like this initiative is similar to, um, another initiative that’s going on in Europe. It’s called a Big Picture, where there is an initiative, it’s a private public consortium. There is a huge image repository built for exactly this purpose.

But obviously the more images the better. And everybody who has an idea how to solve it logistically so that everybody can benefit from those images, go ahead and contact Rish or on LinkedIn or wherever.

Rish: Excellent. Thank you.

Aleksandra: Okay. Thank you so much for being my guest. I appreciate you very much, and I appreciate everyone who is listening to this, all my Digital Pathology Trail Blazers. Thanks for staying ’til the end.

This week in north America is the Thanksgiving week, so I want to thank Aiforia for sponsoring this episode and I want to wish every single one of you, beautiful holidays but [00:28:14] before we started getting ready and start preparing to meet our loved ones,

I wanted to let you know about a special Thanksgiving offer that I have

for those of you, those of my digital pathology, trailblazers who are starting their journey in tissue image analysis, and are not pathologists.

So, if you are part of this group or, you know, someone who is part of this group, keep listening.

I have prepared a limited time specially discounted online course offer. And the course I have prepared for you is called “Pathology one-on-one for tissue image analysis. So if you want to finally understand tissue pathology and do better image analysis on pathology images. This is for you.

So here’s what I’m doing. I am hostinga special BETA cohort of the “Pathology one-on-one for tissue image analysis” course. What does this mean? The BETA [00:29:14] cohort are those of you who want to co-create this course with me.

So the course is partially ready, and you can already purchase all the modules that are available. So if you want to join the BETA cohort, you will immediately have access to everything that’s already there to the existing course modules. Then what we’re going to do together. Uh, we are gonna be co-creating the remaining live modules.

And they’re going to be tailored to your specific tissue image analysis needs.

So by joining this bet that cohort, you can take an active role in the development of the course by providing feedback for the future content. And also by just interacting with me during the live sessions.

But that’s not it. Everybody who joins the BETA cohort this week. So the, this is a limited time offer. [00:30:14] The offer expires. So the discount expires on the 27th of November. So we have one week. And to take part in this.

So everybody who joins me now, who joins me this week will have access to the course will help me co-create the course and we will have an exclusive private Facebook group for the participants of the course. In addition to the course to this course -“Pathology one-on-one for tissue image analysis”, you will also get all the recordings from our recent virtual computational pathology event “Bridging the gap between computer science and pathology”.

Those videos are going to be available to you in the membership area. And you will also have access to audio. So if you want to listen to those lectures on the go while you’re driving while you’re doing your chores and don’t want to look at our faces anymore. You can just listen to this because you’re going to have [00:31:14] the audio of all the lectures from that life event, that’s going to be yours forever.

And then the coolest thing that we’re going to do together from the moment that the BETA cohort starts, we’re going to start monthly educational pathology sessions for tissue image analysis specialists. So, if you want to join me every month for a webinar,

Where I will be answering your questions, working with you on your tissue image analysis problems – we’re going to be doing this every month, and every course participants will get invites to those sessions and we’ll be able to join whenever you feel like you need some pathology support for tissue image analysis problems. So if this is something that speaks to you that resonates with you, that looks and sounds good to you. Here are the four steps to do. In the description of this episode, there’s going to be a link [00:32:14] to join, to join the BETA cohort. And to join the better cohort there are four simple steps .First you have to purchase the course with a discount CODE and I can already tell you what the discount code is.

the discount code is BETA2021. And no, it’s not an old code. I just, at some point when I was setting up this code, I forgot that we are already in 2022. So it’s BETA -B E T A 2021, and you’re going to see it in the description of the episode as well. And you can buy the course with this promo code. It is a $300 discount.

So that’s step one. Then you watch all the available modules. You watch them, you can binge watch them. You can start them the moment you buy the course. Then what I’m going to do I’m -going to send you a questionnaire via email to fill in and give me feedback on what did you like, what did you not like about the existing [00:33:14] modules and what would you want to learn in addition to that? And after I have gathered your feedback, we’re going to be having live sessions, a weekly life session. To address the needs that you have and to finalize the pending modules.

You will have lifelong access to the video recordings and I’m also gonna make them available in audio. I’m a podcast host, so of course I’m a big fan of audio. So each of the modules will also have an audio version.

So, this is also something you’ve been waiting for. I cannot wait to see you in the course.

Go to the episode description and click on the BETA cohort link to learn all the details and have the descriptions of all the steps that you have to take so that we can see each other life very soon. And I’ll talk to you in the next episode!

related contents

image: digital pathology club membership
Join Our Community