Digital Pathology Podcast

105: Artificial intelligence in pathology Part 1 - the presentation I gave on the day I got engaged.

• Aleksandra Zuraw • Episode 105

Share episode

Send us a text

In this episode, I celebrate another milestone of the Digital Pathology Place YouTube channel that was achieved thanks to you, my digital pathology trailblazer, reflecting on its journey since its inception in 2019.

I delve into the developments in digital pathology, focusing on the first video I ever published on YouTube about AI in pathology, highlighting trends, tools, and challenges in the field.

The video was based on a presentation I gave on the day I got engaged, so if you want to know the whole story listen in.

I explain key concepts like
- artificial intelligence,
- machine learning, and
- deep learning, and discuss
- How could AI eventually support pathology practice despite current challenges?

00:00 Welcome and AI Co-Host Feedback
00:19 YouTube Monetization Milestone
01:18 Reflecting on the First Video
02:47 Special Day and Personal Story
05:06 Introduction to AI in Pathology
07:26 AI Terminology and Concepts
13:17 Current Status of AI in Pathology
17:33 Challenges and Future of AI in Pathology
22:42 Conclusion and Call to Action
23:30 Updates and Future Plans

THIS EPISODE'S RESOURCES


Support the show

Become a Digital Pathology Trailblazer get the "Digital Pathology 101" FREE E-book and join us!

Welcome my digital pathology, trailblazers. How did you like the AI? Co-hosts let me know on LinkedIn or. Or via email. And if you haven't heard the AI cohost it's in the previous episode, episode number 1 0 4. At the end of the episode. But today. The digital pathology place achieved an interesting milestone on YouTube. We have monetized our channel. What does that mean? That means that now when play ads on the digital pathology place channel. Then they will pay us money. Which is going to be probably marginal But nevertheless, it is something that happened because of you. Because you, the digital pathology trailblazer are watching the YouTube videos. So thank you so much. And if you haven't checked the YouTube channel yet, please go ahead, check it out and subscribe because then you can get even more value. From the digital pathology information that I'm trying to spread out on the internet. So if you have not checked the digital pathology place channel on YouTube yet, please. Do it and subscribe. Because then you can get even more as I was checking all the settings for the channel. I came across the very first video that I recorded on the channel. It's about artificial intelligence in pathology. And. I looked at it. It was. In 2019. It was a presentation I gave at the conference. Actually, my husband was invited to. We're actually, my husband was invited to present. But I thought, oh, would they not be interested that my husband, isn't a clinical pathologist and I'm trying to invite him to the podcast. To share a little bit of his expertise about I didn't manage to do that yet. So stay tuned anyways. So he went to this conference and he was asked to present on behalf of the company who was with. And I thought, oh, would they not be interested in. It was for, it was a pathology conference. In Hershey, Pennsylvania. And. He asked the organizers. And they said, yes. Sure. So when I was preparing the presentation, I was really practicing. I also recorded a YouTube video and you can have a look at this with. And you're going to have a look at this year, the video. And see what the beginnings are. When I look at my videos now and the confidence. And the expertise I gained. I appreciate that there was a learning curve. Nevertheless this This content is still valuable. And I found that I thought I'm going to share. Glimpse from the past. Actually this happened on a pretty special day. Because I gave. No again. So I was recording this presentation. The sound is not great. Again, I was recording this presentation at the. Hallway of our apartment. My son was. Less than a year. All Dan. So I had to leave the apartment, go through the hallway. I sat in front of a window and I was recording. When you go through YouTube, the sound is not fantastic. I of course. Cleaned it up for you here. So I recorded before the conference, put it on YouTube and I am greeting all my zero youTube subscribers. Now we have over 3000. And then I think a week later I present. I was pretty stressed. And that was on a very special day. Because. It was the day I got engaged. After the presentation. We had our son there and when my husband was presenting, I was taking care of the son. Then I was presenting, he was taking care of my son. And then we went for a hike. We changed the from our fancy conference clothes and we went for a hike. And I was all relaxed. All fun. All good. And then my husband starts taking out stuff and I'm like, oh, that's so cool. We're having a picnic. I'm carrying my baby in the baby carrier. Just hiking ahead. Super happy that they can relax after this presentation that caused me a little bit of stress. And he takes out the blanket. And he takes out some food. And then he takes out flowers and I'm like, are you. Flowers. In your backpack and Until the last moment until the moment. He actually kneeled on this blanket, on the Appalachian trail. And. Ask me. If I'm going to marry him, gave me the ring and asked me if I'm going to marry him. I like had no clue that this is the day that I get engaged. So this video will always remind me of this very special day. And how digital pathology plays began. So enjoy. Dr Alex. From. Over five years ago. I let me know what you think. Learn about the newest digital pathology trends in science and industry. Meet the most interesting people in the niche, and gain insights relevant to your own projects. Here is where pathology meets computer science. You are listening to the Digital Pathology Podcast with your host, Dr. Aleksandr Zhurav.

Aleksandra:

Hi, my name is Aleksandra Żura. I'm a veterinary pathologist and I'm passionate about digital pathology. I had the chance to speak at the annual meeting of the Association of Clinical Scientists in Hershey PA and I wanted to share the presentation I held there with all my YouTube subscribers, which is currently no subscribers, zero people because that's my first video, but with my future subscribers. Me tell you about artificial intelligence in pathology practice and how advanced we really are. We know that there is a shortage of pathologists in the world and AI would possibly save time of pathologists and they would be able to help more patients. So where do we stay? I'm gonna introduce you to the subject. We're gonna talk about pathology informatics terminology. What is the current status of AI in pathology? What are challenges in biomedical image analysis? What is the chameleon challenge in that? And what's a challenge design? Chameleon challenges is the most famous one in pathology. What tools are available, which are not? And what are the hurdles for applications in digital pathology? So let me start with some introduction. We are bombarded with artificial intelligence and digitalization of pathology on pathology conferences. The accuracy of the A. I. Models often comes very close and sometimes exceeds interim intra pathologist variability and the terms such as machine learning, deep learning neural networks. They increasingly are being used in pathology context. So the question is, are we approaching a new era where I will dominate pathology or is it still a long way off? Let me start with the terminology to understand it better. What is artificial intelligence? It is a branch of computer science about systems which can learn from data. So we encounter AI every time we touch any of our mobile devices. And examples here would be Google predictive searches. When we start typing something and to look it up in Google, we get a suggestion. And usually these suggestions are correct. And there, this is a AI product recommendations on Amazon. This is also artificial intelligence. When we buy something and we get suggestions of similar products, or for example, we get this message customers who bought this also bought that. And it suggests us different products, music recommendations by Spotify, based on the songs that we already liked. And Google Maps fastest route that is based on the current available traffic information when we are using it as a navigation system already. And based on the current traffic information, Google suggests a faster route. What is machine learning? This is a branch of AI that is based on the idea that computers can learn from data as humans learn from experience. And Then they can make decisions about data without human interactions. We have three types of machine learning supervised learning based on labeled data, unsupervised learning based on grouping unlabeled data points with similar features together, and reinforcement learning, where the user of the algorithm or the model gives the model feedback on what was labeled incorrectly. What is random forest? Random forest is one of the most popular supervised machine learning algorithms. It's a forest of decision trees. These do not tend to perform so fantastic in pathology. What is deep learning? This is a subfield of machine learning in which the models resemble neural networks of the human brain. And those models within deep learning are called artificial neural networks. And an example or a type of artificial neural networks are convolutional neural networks. And these have dominated the deep learning space. Computer vision is a field of computer science trying to mimic human vision. And patching is dividing large pathology images into small patches. And to better picture this, we need to know how big such an image is. And so a whole slide image is 15 gigabytes, more or less. What is 15 gigabytes? It's three, three hour long, high quality Netflix movies. Just one slide, one digitalized pathology slide. They need to be divided in smaller squares and these squares are called patches and the process is called patching. Graphical processing units are necessary for quick processing of pathology images. These are chips on the computer's graphic cards designed to rapidly process graphics. They're used in gaming, video gaming computers and are indispensable for fast processing of whole flight images. Computer aided diagnosis is when clinicians use computer defined regions to assist them in making the diagnosis. This is already in routine use in radiology. For the detection of breast cancer foci in mammograms and the potential use in pathology would be pre screening of lymph node sections in cancer metastasis. This is a tedious task, takes a lot of time, there are many sections of the lymph nodes and pathologists have to screen them under the microscope. It would significantly, save time if they could have them suggested by the computer and then just say yes, no. What is data augmentation? This is a way of getting more data when we don't have enough data. And on this example, on the left side of the screen, we have four mitotic figures. They have been slightly altered shifted a little bit, and we get this. From those four mitotic figures, we get 16 mitotic figures. We increased our data set from four points to 16 data points. And these are seen as by the computer as separate data points. So we don't have enough data. We can augment them to have more for the training of the system. What are probability heat maps? This is a color coded way of visualizing the classification results of the deep learning model. And on one end of the scale is corresponding to a high 100 percent probability of a feature in this case of the tissue being a tumor and the other end of the scale, the blue one in this case is corresponding to zero probability. So we can appreciate that those all red regions, is. Probably tumor and regions in between, and there is lower probability of tumor. To visualize how those terms relate to each other, the broadest term is artificial intelligence, and the machine learning is a part of artificial intelligence, where we have the patching, random forest, and deep learning is part of machine learning. We have Artificial neural networks, convolutional neural networks, as an example of them, we have probability heat maps and data augmentation. Computer vision is the discipline that uses all those models. GPUs, graphical processing units enable us to process whole slide images fast. And in the end, the algorithm can help us with making the diagnosis, and we have computer aided diagnosis. On conferences and in publications on LinkedIn as well, we are surrounded with all the AI news, but what is the current status of artificial intelligence in pathology? As of today, May 18, 2019, there is no FDA approved AI solution for pathology on the market. There are initiatives supporting and accelerating the development of AI applications, like the FDA Digital Health Innovation Action Plan, Digital Health Software Pre Certification Program, Pre Cert, and in different disciplines, AI already. Made it officially with an FDA clearance or approval like diabetes research in cardiovascular and brain disease treatment in radiology, but not in pathology yet and in pathology publications about A. I. All those publications come from R. N. D. Departments, either academia or industry, different companies, but it's research. So to foster AI and to encourage people to develop a solutions and challenges in biomedical image analysis have been established. And what are these? These are competitions aiming to compare new and existing algorithms in biomedical image analysis. Thank you. And researchers are very strongly encouraged to take part in those challenges in order to promote and contribute to AI development. The participants try to solve a stated problem on a common data set, and they can use a solution of choice and are required to publish their results. So the most famous challenge in pathology was chameleon challenge organized in 2016 and 2017. And this and other challenges are gathered on the comic platform standing for consortium of open medical image computing. And they are called grand challenges. They are the most famous in biomedical image analysis, but there are also other platforms that are hosting them. And these include Coda, lab Coval, or Virtual Skeleton. And let me tell you about the communal challenge. What it was about it, as I said, was one of the most famous challenges in pathology. Maybe not in biomedical image analysis or different ones that I'm not aware of, but in pathology. This one was the most famous one. It was hosted by in the Netherlands by, the diagnostic image analysis group and department of pathology offered the Radboud University Medical Center in Nijmegen. And as I said, there were two additions, Chameleon 16 and Chameleon 17 in 2016 and 2017. And almost 100 submissions for both of these challenges were done both from academia and from companies. And the problem here was detection of breast cancer, metastasis, and whole slide images of lymph nodes. The participants were working with the same training and test set of 1, 399 hematoxylin and eosin stained lymph node sections. And they were free to use any image analysis method to best solve this problem, so they didn't have to use deep learning. But it just so happened that the winners used deep learning. So this performed the best. This is how the challenge is designed. Not only the pathology ones, but the other ones as well. A meaningful task needs to be defined. In this case, that was breast cancer metastasis detection in lymph nodes. Representative data needs to be gathered. In this case, it was labeled whole slide images. A reference standard needs to be defined. That was the diagnosis of the pathologist and their annotations. and a discriminative evaluation metric needs to be determined. And in this case, the consensus of the algorithms with the pathologist input was measured by Cohen's kappa and the participants are required to write a peer reviewed pathway. So what are the tools that are available and what are not available? There's no FDA approved AA solutions for pathology, like I said, but Why is there not? We will see later. But for research use only, image analysis software companies are starting to incorporate AI modules in their solutions. where a pathologist or a user can annotate target structures to train the model and later the model detects those structures automatically. For example, the pathologist or the user annotates glomeruli on one slide or in one part of the slide and then the algorithm detects the rest of the glomeruli on the other part of the slide or on the rest of the slide. These solutions are currently available from different companies like Indica Labs, Viziopharm, and Aphoria. And these tools incorporate random forest or deep learning modules for image analysis, but as I said, they're for research use only. Nothing for clinical use so far. And why not? Why is there something AI based in other disciplines and not in pathology yet? So there are some hurdles that are pathology specific, and these include lack of validated tools, as we already said, lack of labeled data. We said that the machine learning need supervised machine learning needs a large amount of labeled data. This would mean that the pathologists instead of looking at sites would have to go and start annotating whatever structures they want to have detected. And it's not going to happen because they are supposed to diagnose patients and they use slides for that, the non digital slides. So lack of labeled data, high complexity of histological images. In comparison to radiology or cardiology, the complexity of histological images is much, much higher. We have different colors, textures, tissues, organs, and deaths multiplied by each other contributes to this high complexity of histological images. The dimensionality of pathology diagnostic problems is also high. And the example that we had the detection of neoplastic foci in a lymph node section is just one binary problem that was used for the competition. But the dimensionality of Primary diagnosis in pathology is a lot higher. There is a different pattern recognition involved and incorporation of many different processes visible on the slide and different data available about the patient. Another problem or a hurdle is that pathologists are the gold standard or they are generating the ground truth. So the algorithm can only be as good as a pathologist and also pathologists tend to differ. Algorithm is gonna be close to what one pathologist says, does it mean that it's a good one if another pathologist says something else? So that's a hurdle. And the affordability of computational power and storage space is a problem. Those GPUs that are required for processing these huge slides cost a lot more than a normal PC, normal computer. That's a hurdle. And we need to store somewhere those digital slides, enormous amounts of data. So that's another hurdle. So to sum up, AI is affecting our lives every time we touch a mobile device, phone, whatever, computer. It has officially entered various medical fields, but in pathology it still remains restricted to researchers only. There are many pathology specific hurdles to overcome, and the most important being enormous size of pathology images and complexity of morphological patterns. The regulatory pathways have been already paved by other disciplines, which is good. And the FDA officially supports the digital health initiative with their digital health innovation action plan and digital health software pre certification program. So to answer the question that we ask at the beginning. Is it gonna happen? Yes, it's gonna happen soon. A. I. Is coming soon. It's we're not quite there yet, but it's coming soon. So pathologist should prepare. They can prepare by going digital in their labs by recognizing where A. I. Will add value. So it may not add value to every step of the pathology workflow. It's crucial to recognize where it would and apply it there. And they can prepare by educating themselves. There are courses available, one course for example, from digital pathology association, an online course that pathologists can take and they should proactively engage with a initiatives and beat the challenges like the chameleon challenge or collaborations, cooperations with academic partners, with industry with different companies. Thank you very much for listening to me. And I hope you liked it. If you liked it, please click subscribe to do my channel. And if you have any questions, don't hesitate to write an email and go and check out my blog and digital pathology consulting dot com, where there is an article about this presentation as well. And I will also. Put this presentation for download if you would like the slide. Thank you. Bye Here the references in case you want to go back to the primary literature or and Follow the links for the things that I mentioned.

Thank you so much for listening to the end. That tells me you are a real digital pathology trailblazer. This video, was posted five years ago and it has 3000 views on YouTube. This is one of the most viewed videos. Of course it has been there for the longest. And there was a whole chunk of what happened in the last five years missing, like all the generative AI foundation models, chat GPT. And all that good stuff. So of course there is an update video that I posted five months ago. And this was has almost 2000 views already. So I'm going to link to both the older version and the new version if you feel like comparing. How I was presenting them versus how I'm presenting now. In those over five years. My YouTube channel is telling me that there are 510 videos. It's a little bit. Elevated, because at some point I connected the digital pathology podcast. So all the audio podcasts are on YouTube as well. When Google decided to merge Google podcasts with YouTube, we migrated to YouTube. So it doubled the number of videos, but definitely over 370. Native YouTube videos. So I decided to create a structured curriculum out of those videos, they were being created more or less on demand when I was hearing from you. Oh, I would like to have a video on that. I would have liked to have a presentation or if I would give a presentation, there was an option to record. You're going to find things like that as well. Recently, I started adding conference vlogs to give you a glimpse of what's happening at the conferences and a life live commentary on what's happening. So I decided to create a structured curriculum. And create. Course based on. Selected and curated YouTube videos. I am still working on putting it together. But if you're interested in checking it out, In exchange for feedback and testimonials, I have a waiting list, for 100 digital pathology trailblazers who will get access to this course for free. So if you're interested in test driving the course, giving me your feedback, whatever you think about it. You just let me know. And if you love it, if you would give me a testimonial, then this course is going to be for you for free. I already shared it with the digital pathology trailblazers from my mailing list. And several of them signed up. There are still a few spots left. For you, the podcast listener but before the. free. access waiting list, Gets shared on social media. I decided to share it on the podcast. So for you who listens to the end. This is an opportunity to get access to this course for free. And let me know what you think. Check out the link in the show notes. YouTube course Waiting list, sign up and whenever I have ready, it's going to end up in your inbox. And I talk to you and the next episode.