104: DigPath Digest #12 (AI in Pathology: From Retinal Screening to Global Cancer Diagnostics + my new AI Co-Hosts)

In this episode of DigiPath Digest you will learn about the development of AI models for glaucoma screening using fundus images, the use of AI in detecting metastatic deposits in colorectal cancer, and leveraging immunofluorescence data to reduce pathologist annotation requirements. 

Dr. Aleks also invited two AI Co-hosts and shared personal reflections on AI's role in the industry and invites feedback from listeners on AI-generated content.

00:00 Introduction to the Livestream Disaster
00:24 AI to the Rescue: Enhancing Audio Quality
00:38 Meet the AI Co-Hosts
01:04 Welcome to the Digital Pathology Podcast
01:30 Technical Difficulties and Audience Interaction
02:49 Exploring AI in Veterinary Medicine
04:34 Hybrid Convolutional Neural Network for Glaucoma Screening
07:49 Model for Detecting Metastatic Deposits in Lymph Nodes
11:23 Leveraging Immunofluorescence Data for Lung Tumor Segmentation
18:05 AI-Generated Content and Future Plans
21:37 AI Co-Hosts Take Over
32:42 Conclusion and Audience Feedback

TODAY'S EPISODES RESOURCES
📰 Hybrid convolutional neural network optimized with an artificial algae algorithm for glaucoma screening using fundus images
🔗https://pubmed.ncbi.nlm.nih.gov/39301801/

📰 Automatic segmentation of esophageal cancer, metastatic lymph nodes and their adjacent structures in CTA images based on the UperNet Swin network
🔗https://pubmed.ncbi.nlm.nih.gov/39300922/

📰 Retrosynthetic analysis via deep learning to improve pilomatricoma diagnoses
🔗https://pubmed.ncbi.nlm.nih.gov/39298885/

📰 Obesity-Associated Breast Cancer: Analysis of Risk Factors and Current Clinical Evaluation
🔗 https://pubmed.ncbi.nlm.nih.gov/39287872/

📰 Model for detecting metastatic deposits in lymph nodes of colorectal carcinoma on digital/ non-WSI images
🔗 https://pubmed.ncbi.nlm.nih.gov/39285483/

📰 Leveraging immuno-fluorescence data to reduce pathologist annotation requirements in lung tumor segmentation using deep learning
🔗 https://pubmed.ncbi.nlm.nih.gov/39284813/

📰 Bayesian Landmark-based Shape Analysis of Tumor Pathology Images
🔗 https://pubmed.ncbi.nlm.nih.gov/39280355/

📰 Globalization of a telepathology network with artificial intelligence applications in Colombia: The GLORIA program study protocol
🔗 https://pubmed.ncbi.nlm.nih.gov/39280257/

📰 Towards next-generation diagnostic pathology: AI-empowered label-free multiphoton microscopy
🔗 https://pubmed.ncbi.nlm.nih.gov/39277586/

📰 Sex differences in sociodemographic, clinical, and laboratory variables in childhood asthma: A birth cohort study
🔗 https://pubmed.ncbi.nlm.nih.gov/39019434/

Transcript

Aleksandra: 0:00

Who wants to listen to a disaster? This livestream, the 12 digit path digest was disastrous due to the collapse of the internet. And no, don't worry. I'm not going to be serving you. The disasters version, the disasters version is currently on YouTube, and I'm not even going to be sharing this with a digital trailblazer mailing lists because it was so bad. It was kicking me out. It was cutting off my speech. So for the audio version, I went through the whole thing. It's going to be edited. The audio is going to be polished by AI. Thank you, AI for helping me out so that you have a lot better listening experience, that those who are viewing. And today AI did not only help me make this podcast episode more. Listable I actually invited two AI co-hosts so, if you want to listen to how they performed. Can they stay on as official digital pathology podcast, cohost. Listen to the end. The first part is going to be just me and hast with the AI better audio. The second part are my two AI co-hosts. So let's dive into it.

1:05

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. Aleksandra Zuraw.

Aleksandra: 1:31

Good morning my DigiPath Digest trailblazers. Welcome to the DigiPath Digest and let me know if you hear me because I had trouble before me out of my online stream again. Why are you doing this? Why are you doing this to me and we have already some people give me a give me a hi and let me know if You can hear me because I had this beautiful microphone that didn't want to work today and I had to change Last minute, so I would love you to tell me that you hear me you should and I want to tell you something. Yesterday or like this week or in general after reading a certain post that I'm gonna show to you it is recording locally in case I drop off. I will just upload this. Remember how I always say, oh, people are always asking this at the conferences. Will I lose my job with when AI will take over? So I felt kind of strange when I saw this. I know it's cutting off a little bit, but let's just ignore this glitch, and let me tell you about some important things. Dr. Candice Chiu, she's a fellow veterinary pathologist, and if there was anyone that I would think of as The rising star that's her. She's the rising star of implementing AI in veterinary medicine I had her on the podcast. This podcast is not yet published, but it's gonna be published on October 8th and She basically trained her own chat GPT. She teaches on AI on different AI tools and there was this post that she posted on LinkedIn telling people that you can now make podcasts out of anything and you just go to I'm going to share a different screen. You just go to Google Notebook. What is Google Notebook, Alex? I'm going to show you what Google Notebook is. It's called Notebook LM language model, I assume. And you can upload whatever you want, like PDF and other documents, and it's going to create a podcast for you. And I uploaded today's PDF.

3:53

And I will be.

Aleksandra: 3:54

Including this in the audio podcast version maybe like in addition to what we will be discussing in the DigiPath digest or replacing. So I basically gained two podcast hosts. So I wanted to let you know that this was the first moment where I felt, Oh, This like podcasting? Do you need real people still? I hope so. And if you are here live, let me know in the chat. And we are gonna dive into our DigiPath Digest. I think the internet's glitches are over. We're gonna start with Hybrid Convolutional Neural Network Optimized with an Artificial algae algorithm for glaucoma screening using fundus images. And this is from Journal of International Medical Research. The impact factor is 4. 046 So the objective was to create a decision support system for retinal fundus image based glaucoma screening, screaming, no, screening. And they combined computer vision algorithms with a convolutional network for fundus images and applied faster region based convolutional neural network and artificial algae algorithm with support vector machine classifiers. And they detected optic boundary, optic cup, and optic disc. And say, this there was a segmentation conducted. And they used three fundus image databases. There are different databases for fundus images that they trained it on. And the conclusion was that one of those networks, A A A S V M which is the artificial algae algorithm with support vector machines was the best

5:53

And could be.

Aleksandra: 5:54

The computer aided decision support that's going to be useful for

5:58

Optometrists.

Aleksandra: 5:59

I like these solutions because, so my touch screen is working, but my pen is not working, that's okay. See, it's time to buy a new tablet. I like this type of imaging that is in vivo imaging, so to say. Basically imaging that you can do while you have the patient with you. You have the camera you take a picture of the eye and you can run an algorithm. And basically this can be your computer aided diagnosis system.

6:28

And I have two super papers for you today.

Aleksandra: 6:30

I love them very much because so the first one that the glaucoma screening I like because it's not whole slide images Hall site images are amazing. You can do computational pathology. You can do a lot of things. Basically you digitized this glass specimen. You now can use all the computational pathology tools to to analyze it, to extract insights. Do whatever you want with it. But, They're huge, they require a scanner they're computationally still pretty expensive, meaning they're just big, and I mean there are methods that don't require analyzing this whole image. There is a streaming of the image where you like stream it like a movie, you don't have the whole movie. at a time to watch, you watch a piece at a time and you still know what's in the movie. So there are technologies like that, but still like you need a scanner. You need to digitize the analog glass slide with Static images in vivo images, first of all,, fantastic. Like you can immediately use computational pathology, computation, computer vision. Second, the best thing for applicability for being able to actually reach the underserved area are static images. And this is what this paper is about called model for detecting metastatic deposits in lymph nodes of colorectal carcinoma on digital non host light images. And I see people joining on YouTube as well. Amazing to have you here. Okay. Models for this model for detecting metastatic deposits in lymph nodes of colorectal cancer. This was published in Diagnostic Pathology Impact Factor 2. 6. And the main author, the first author is Dr. Talat Zera. And she was a guest on my podcast. Whenever this live stream ends, I'm going to link to the podcast with her so that you can learn more about her mission and how she is approaching digital pathology because she is working in Karachi, Pakistan, which is definitely a country that would be considered underserved area, underserved country. And what happened here, we are talking about colorectal cancer and this constitutes around 10% of global cancer diagnosis and death due to cancer. I share, okay. So colorectal cancer this is 10 percent of global cancer diagnosis and death due to cancer. And The study was there was a retrospective cross sectional study including digital images of glass slides containing sections of positive and negative lymph nodes. This retrospective study was taking pictures of glass slides and there was 165 previously diagnosed cases. They were selected from the hospital Aga Khan University Hospital from January 2021 to January, 2022. So it was one year. And the images were generated at 10 X. It wasn't 40 X. It wasn't even 20 X. It was just done at 10 XI think this was to mimic. the pathologist's workflow, and they were uploaded to an open source platform to QPath. And analyzed there, there was a deep learning model, Ensembl was applied, or Ensembl, was applied to the identification of tumor deposits in lymph nodes, and So we have static images, we have open source software, and we have low magnification, so we don't have huge images, so the computational requirement decreases significantly, and what were the results? The negative lymph nodes detected by AI was 78. We had 165 cases. And out of these 69 were true negatives. So that was 88% were true negatives and. 11. 5 were false negatives. That gives us the sensitivity of 89% and specificity of 83.1. And the odds ratio was 40 with a confidence interval of 16 26 to 98.3. And the conclusion is though it was a small study, its results were really appreciating and we encourage more such studies with big sample data in the future. I definitely encourage more of those studies because they are so low barrier to entry for digital pathology and they can help you leverage digital pathology without huge computational resources. The next paper that we have is very dear to my heart because of two things. First, it's about annotations. Like making it less cumbersome for pathologists. And the second, thing is it comes from the non clinical safety and pathobiology group at Gilead Sciences in California. Why am I excited about this? Because I am a toxicologic pathologist. I'm a non clinical safety pathologist. The first author is Hatef Mehrabian. The title is leveraging immunofluorescence data to reduce pathologist annotation requirements in lung tumor segmentation using deep learning. We are using immunofluorescence data to reduce pathologist annotation requirements. um, Why do we want to do that? Because annotations is the main bottleneck of training, Specifically segmentation models and we still work with annotations as ground truth. So we motivate pathologists to annotate at scale a lot and have consensus. And the best would be three pathologists annotating the same lesion and checking everything. That will not happen because we don't have pathologists who have spare time to do that, but still, that would be ideal for great models, right? So what can we leverage? And actually I was part of a similar project at my previous company. Where in the computer scientists were leveraging a specific immunosuppressive chemistry stain. Here we are leveraging immunofluorescence. So the technology is a little bit different, but basically. What the group here did, They create a large dataset of low cost, low accuracy pan CK, that cytokeratin based annotations. So we are talking about non small cell lung cancer NSCLC and PANCK is a marker that stains epithelial cells. PANCK is basically like a lot of different cytokeratins. There is more than just one cytokeratin and cytokeratin is a marker of epithelial cells. And if you stain those epithelial cells, then basically you delineated the tumor. It is still a part of many workflows, even though AI advanced models advanced on H and E, you can detect different things, but still delineating manually the tumor on images is very A bottleneck is still part of a lot of AI AI initiatives where pathologists are doing that. That's boring, I have to tell you. So anyway, they stained with pan CK and then they pre trained the model and determined the minimum required size of the expensive but highly accurate pathologist annotation data set. So here I have to pause for a second. We are using immunofluorescence. Sense, and that is not exactly like the cheapest technique to stain something, it's not easy. It's not just clicking a button and generating this out of thin air, you actually have to do it, do a pretty complex complicated procedure in the lab so they did IF and they compared this to the expensive pathologist time. So here is my things like, okay, yeah, if you don't need to do that pathologist time on top of the IF, then maybe you are lowering costs. So they say it was low cost and low accuracy, but this pre training was enough to Okay. have pretty pretty nice results and they pre trained the model and determined the minimum required size of the expensive but highly accurate pathologist annotation dataset. And PanCK pre training was compared to foundation models which I would need to know which ones and but that's okay. We can always go to the paper or you can use this. Uh, Google language, muddle notebook. That they showed you at the beginning. That actually can summarize it for us and give us a podcast without me having to do anything. other than uploading the file. So panCK pre training was compared to foundation models and various architectures were explored for model backbone and proper study design and sample procurement for training a generalizable model that captured variations in NSCLC and H and E was studied. This is a, an important sentence. Proper study design was studied. It's so important. Like study design can. Make or break your study, and if you have the wrong study design, then the data from your study are useless. So proper study design is key and they The H& E imaging was performed on 112 samples from three centers, two scanner types, different staining, and imaging protocols. This is fantastic because the, not only H& E is like more stable, it's a chemical stain, and even though it's more stable than immunohistochemistry, immunofluorescence that you have to optimize for image analysis if you want to do it right, but it's there is a domain shift, meaning if you have it in different centers, you scan it with different scanners, the models don't perform so well. So here they did insert it a lot of variability and that's what happened. So they had those different scanners and attention uNET architecture was trained using a large pan CK based annotation data set. 68 sample, total area of 10, over 10, 000 millimeters square. Honestly, I don't know how much that is, but 68 samples is not that much but that's okay. Now let's check the results. Followed by fine tuning using a small pathologist annotation as dataset. 80 samples, so a lot of samples, but the area was just 200 millimeters. Okay, so we had less samples for pan CK annotations, but the area was 10, 000 and here we just have the area is 200. 200 millimeters, so a lot less area annotated by pathologists. And they checked intersection over union between pathologists and and the detection of pan CK. Intersection over union is,

17:45

a metric that compares how much, uh, two different annotations overlap. If they overlap perfectly, that's a one intersection over union and if they just touch, then it's almost zero and that's not good intersection over union. So this is what was checked here.

Aleksandra: 18:02

so here, we have a we have a first hand example. What happens if the internet goes down like digital pathology goes out the window? But I just wanted to thank you so much for joining me. On the audio version, There is going to be me trying out this notebook. Of the post from Candice Chu, Dr. Candice Chu that I mentioned at the beginning. She's exploring like all the AI tools that are available. And she found this tool Notebook LM, here, Notebook LM. So what I did with this notebook was, to put the PDF that We were just discussing into this notebook and it made a podcast out of it. And the podcast is with two hosts male and female host voice. And it sounds so realistic. They have a conversation. It turns the PDF document that is not written in a conversational style into a conversation. So I'm going to be exploring this in the audio version and I'm going to share it with you. If you are already on my mailing list, that's fantastic, you're going to get this. If you are not, you can go to digitalpathologyplace. com, download my book and that's how you end up on my mailing list and then I can give you all the cool things that I'm exploring about digital pathology, digital pathology literature, and pathology and AI, and all those cutting edge things. And in the meantime, you have a wonderful day, and I talk to you in the next episode. Okay, so here it is another 11 minutes of AI generated content. Check it out. I listened to all of it. It includes more than just the three papers mentioned previously. And I have this PubMed alerts set up for 10 papers. So it actually summarizes all 10 papers, but what I'm going to do as a little project, like additional podcast project, I want to take two papers. And make a separate AI generated podcasts out of them. Both of them are open source and they are about topics that they're close to my heart that help underserved areas. One is going to be the one by Dr. Talent zero. Zara from. Um, Pakistan Karachi. And the other one is going to be about tele pathology in Columbia, which was not included in the previous section, but it is included. The abstract is included. In that AI chat of my new AI. Co-hosts I guess I'm going to be experimenting with this and you let me know what you think. I'm going to be sharing this on social media. So whenever you see it, then you have an opinion on. Okay. Is it good? Is it bad? Anything you think about that you feel about it? Honestly today was the first time I felt, oh, AI is going to definitely take part of our responsibilities away and it might feel strange. So without further deal. Welcome my AI. Co-hosts. I feel, I, I feel like I need to give them a name. So it's going to be miss H and. Mr. E I guess H E are going to be discussing. The topics that I tried to discuss and let me know who's better. Which. Which. hosts do you prefer? Dr. Aleks Zuraw, Ms. H or Mr. E?

Mr. E: 21:37

Welcome back everybody to another deep dive. You know, we always bring you the latest and greatest in AI and digital pathology. And, uh, well, this month is no exception. We've got 10 new research abstracts hot off the presses just this past September.

Ms. H: 21:50

We're really talking about AI that's doing more than just, you know, automating tasks. These algorithms are uncovering hidden patterns. You know, they're giving us insights into these diseases that we've never had before. It's really changing how diagnoses are made and how we think about treating diseases.

Mr. E: 22:07

It's really incredible, you know, we've got AI teaming up with telepathology to bring those expert diagnoses all over the world. We've got algorithms that are analyzing the shapes of tumors and actually predicting how well patients are going to do. It's really incredible. There's even a glimpse into how AI is being used to understand really complex conditions like childhood asthma. It's really amazing.

Ms. H: 22:27

Yeah, it really feels like we're pushing the boundaries. of what we thought was possible, you know? Take the Joluria program in Colombia, for example. They're trying to solve this huge problem. How do we make sure that everybody has access to accurate and timely cancer diagnosis no matter where they live?

Mr. E: 22:42

So how are they doing that? That sounds like a really tough challenge.

Ms. H: 22:45

It is, but they're tackling it head on by using a combination of telepathology, which lets doctors share and analyze images of biopsies remotely, and they're combining that with these sophisticated AI algorithms. And what these algorithms can do is they can identify those cancer cells and even flag areas for the pathologist to review. So even in areas where they might not have the specialized doctors, they can get those expert consultations.

Mr. E: 23:08

So they're basically bringing the world's best doctors to everybody's doorstep using these virtual tools. I mean, that just seems like it's going to really level the playing field when it comes to health care equity.

Ms. H: 23:20

Yeah, absolutely. And speaking of seeing things in a new light, there's also some groundbreaking research on this technique called label free multi photon microscopy. Now, Traditional microscopy often uses dyes and labels to highlight specific cells or structures, but that can be time consuming and it might even mask some important details. So

Mr. E: 23:41

this label free approach, is that about kind of seeing those cells in their natural state?

Ms. H: 23:45

Exactly. Multiphoton microscopy uses the way that light interacts with tissues to create these high resolution images without needing those traditional labels. But here's where AI comes in. It can be used to analyze these super complex images, find those subtle patterns, those anomalies that our eyes might miss.

Mr. E: 24:02

It's like having, like, an AI co pilot helping us navigate and understand that microscopic world with this detail we never had before. Makes you wonder what breakthroughs are just around the corner.

Ms. H: 24:14

It really does. And, you know, it's not just about catching things early. Some of these studies are also using AI to analyze the shapes of tumors in ways that could completely change how we predict how well a patient is going to do, you know, their outcomes and how we can personalize those treatments for them.

Mr. E: 24:30

So we've got two studies that focus on lymph nodes and tumor identification. Why are those so important?

Ms. H: 24:35

Well, when we talk about treating cancer, accuracy is everything. Knowing the exact size and spread of the tumor, especially in the lymph nodes, is absolutely critical for staging cancer correctly. And that staging is what determines those treatment decisions. Do we use surgery? Do we use chemotherapy? Radiation? Getting that staging right is sometimes the difference between successful treatment and And, you know, a worse outcome for the patient.

Mr. E: 25:00

So it's really about getting those measurements right down to the microscopic level.

Ms. H: 25:03

Yeah. It's like having a microscopic ruler.

Mr. E: 25:05

Exactly. And the

Ms. H: 25:06

implications are huge. You know, imagine being able to stage cancer with even more certainty than we can now. That could lead to much more targeted treatments, you know, minimizing side effects and really maximizing how well those treatments work.

Mr. E: 25:19

So we're really talking about making personalized medicine a reality and it sounds like AI is a big part of that. But, you know, it's not just about cancer, these Abstracts cover a pretty wide range of applications for AI.

Ms. H: 25:31

Oh, absolutely. AI is not just limited to one area of medicine. We're seeing it being used to analyze those complex patterns in all kinds of diseases. Like that study on childhood asthma, for example.

Mr. E: 25:42

Right, and what's interesting about that one is it shows how AI can work with data that's not even visual. Right, like we're talking about lung function tests, environmental exposures, even genetic information, and AI can put all that together.

Ms. H: 25:54

Exactly. AI is really good at finding those hidden connections and huge amounts of data connections that we as humans might miss. And in the case of asthma, you know, AI might be able to find those early warning signs, or predict when someone's going to have a flare up, or even personalize their treatment plan based on their unique risk factors.

Mr. E: 26:12

It's like AI is giving us this deeper understanding of the disease itself, not just the symptoms. It's amazing to think about how this could change healthcare on a much larger scale.

Ms. H: 26:21

Absolutely. Now there's another study I wanted to touch on that really shows how AI can make pathologists even better at what they do. This one focused on pilometrichoma.

Mr. E: 26:30

Pilometrichoma. Remind me what that is again.

Ms. H: 26:32

It's a type of skin tumor that's benign, so it's not dangerous, but it's pretty common in kids. And sometimes it can be mistaken for other skin conditions. So that's where the diagnosis is really important. The key is being able to look under the microscope and identify those specific cells. Basaloid cells and ghost cells.

Mr. E: 26:49

And that's where the AI comes in.

Ms. H: 26:50

Exactly. They used a deep learning algorithm in the study to not only help diagnose pilometric coma, but to actually highlight those important cells within the biopsy image.

Mr. E: 27:00

So it's like the AI is giving the pathologist a guided tour of the most important areas to look at.

Ms. H: 27:06

Exactly. It makes the whole process smoother, improves accuracy, and ultimately means that pathologists can diagnose things faster and with more confidence.

Mr. E: 27:13

So it's not replacing the pathologist. Yes. It's more like giving them a really powerful assistant.

Ms. H: 27:17

Exactly. And speaking of augmenting expertise, there was that study that used a technique called retrosynthetic saliency mapping. What is that all about?

Mr. E: 27:27

Yeah, that one sounded really interesting.

Ms. H: 27:28

Basically the AI looks at the image and creates kind of like a heat map showing the areas that are most important for making the diagnosis.

Mr. E: 27:36

So it's like the AI is saying, hey, pathologists pay really close attention to this area. This is what's key here. It's like giving them x ray vision for pathology.

Ms. H: 27:44

Uh huh. I like that analogy. It really gets at what AI can do. It can help us see things that we might easily miss, even those trained professionals. And that brings up a really important point about data in AI.

Mr. E: 27:56

Right. These algorithms don't just magically work. They need to learn from data. And not just any data. It has to be high quality data that's diverse and accurately labeled.

Ms. H: 28:04

Exactly. And that's actually one of the biggest challenges in the field. A lot of these studies talked about how much work goes into creating these large and robust data sets.

Mr. E: 28:14

Like that study that used PANCK based annotations for segmenting lung tumors. Can you explain what that means?

Ms. H: 28:19

Sure. So PANCK is a technique where they use fluorescent dyes to stain different proteins in the tumor. And what that does is it creates this detailed map of the tumor so we can see its borders more clearly. And that helps with segmentation, which is basically outlining the exact boundaries of the tumor in the image.

Mr. E: 28:36

So it's like giving the AI a much clearer blueprint to learn from.

Ms. H: 28:40

Yes, exactly. And the results were really impressive. They were able to significantly reduce the amount of manual annotation that the pathologists had to do, which frees up their time to focus on more complex cases.

Mr. E: 28:52

That's huge. It shows how we can use AI to handle those tedious tasks and really boost efficiency in the lab.

Ms. H: 28:58

It's all about finding those win win situations where AI can complement what humans are good at and just make the entire workflow better.

Mr. E: 29:06

Now, I don't want to get too lost in the technical details, but there was one more study that Caught my eye. It talked about using something called immunofluorescence data to make lung tumor segmentation more accurate.

Ms. H: 29:18

Ah, yes. That study was interesting because it showed how AI can actually combine different types of data to get even better results. So instead of just using those standard images, they also included data from immunofluorescence staining, which, as we talked about before, highlights specific proteins within the tumor.

Mr. E: 29:34

So it's like giving the AI a more complete picture of what's going on.

Ms. H: 29:37

Exactly. By combining those different data sources, they were able to create an algorithm that was even more accurate at identifying the edges of the tumor.

Mr. E: 29:46

And having that level of accuracy is so important for making those big decisions about treatment. Like whether or not surgery is even an option, or how much radiation might be needed. It really shows how AI can help personalize those treatment plans based on a much deeper understanding of each person's tumor. You know, it's amazing how much information we can get from just one tiny little biopsy image these days.

Ms. H: 30:08

I know. It's really incredible. And it really speaks to how far we've come in such a short time in digital pathology and AI. I mean, just a few years ago, we could only have dreamed of. This level of detail or making these kinds of predictions

Mr. E: 30:21

feels like we're on the edge of a revolution in health care here Like any revolution there could be challenges that come with all this opportunity

Ms. H: 30:28

Oh, absolutely as we go further down this road with AI driven diagnosis We've got to keep the ethics in mind, you know, think about potential biases We have to make sure that these technologies are developed and used responsibly We need to always be asking ourselves. Are we building systems that are fair? Are they transparent? Are they accessible to everyone? Really, really important questions, and it makes you realize that this isn't just about the tech itself. It's about collaborating, having these open dialogues, making sure that AI is serving humanity in a way that benefits everybody.

Mr. E: 31:01

Couldn't have said it better myself.

Ms. H: 31:02

Yeah.

Mr. E: 31:03

It's about taking the ingenuity of these algorithms and combining that with the expertise of our healthcare professionals. Because at the end of the day, it's about giving them the tools they need to make the best decisions for their patients. You know, one thing that struck me looking at all these abstracts was just how global this research is. We've got studies from Colombia, from China, from the U. S. It really highlights how this is a worldwide effort.

Ms. H: 31:25

It's really encouraging to see because when you think about the biggest challenges in healthcare right now, improving access, personalizing treatments, really understanding these diseases at a deeper level. These are global issues that need global solutions.

Mr. E: 31:38

Exactly. It kind of gives you a sense of hope. It's a

Ms. H: 31:48

good reminder that even when the world feels divided, science and technology have this power to connect us and drive progress, and these abstracts, they give us this glimpse into a future where location isn't a barrier to quality health care.

Mr. E: 32:01

A future where we're diagnosing these life threatening diseases faster than ever, tailoring treatments to each individual patient, and really pushing the boundaries of what we thought was possible.

Ms. H: 32:12

That's the future we're all working towards. And it's a future I'm really excited about.

Mr. E: 32:16

As are we. Well, we've covered a lot of ground today. Everything from AI powered microscopes to personalized medicine to the ethical considerations of it all. And remember, This is just the beginning.

Ms. H: 32:28

Oh yeah, the field of AI and digital pathology is just exploding right now. The possibilities are really limitless.

Mr. E: 32:33

So stay curious, keep exploring, and join us next time for another deep dive into the world of digital pathology. Until then, happy diving everybody.

Aleksandra: 32:42

Okay, it's me back again. Your host, Dr. Aleks. What did you think? If you've listened to the end, you are a true digital pathology trailblazer. And I definitely want to hear from you what you thought about the AI generated content. I think it's impressive. And I will be experimenting with this. I don't know how I will incorporate it. It's definitely leverage. And I love tools that give me leverage. And that very much makes me think. So initially it will told you, I felt strange about this, but now I'm thinking that. AI taking over so many things, it kind of gives us permission to be unapologetically ourselves. So that's what I'm going to be leaning towards. Let me know what you think about it. Are you scared? Are you excited? A combination of both. It's a combination of both for me and I'm going to be actively. Figuring out how to use this tool. To leverage the promotion of digital pathology. I'm looking forward to hearing from you and I talk to you in the next episode.