Digital Pathology Podcast

93: DigiPath Digest #2 (Why Digital Pathology Newsletter Audio Died and What I Am Doing Instead)

• Aleksandra Zuraw • Episode 93

Share episode

Send us a text

This is the audio version of the DigiPath Digest - Abstract review that I host on YouTube

Here is the video version if you learn more visually

Today I explain what happened with my "beginning of year initiative" to post an audio version of the Digital Pathology Newsletter sent out in an email form. 

In a nutshell: I just stopped posting it, you will find out why in this episode.

TIMESTAMPS

00:00 Introduction to DigiPath Digest
00:13 Challenges in Digital Pathology
01:31 Consistency and Sustainability
02:51 Abstract Review Process
04:25 Engaging with the Community
08:31 First Abstract: Molecular Classification of Breast Cancer
14:21 Second Abstract: AI in Breast Cancer Detection
20:53 AI-Assisted Pathology: Time Reduction and Sensitivity Improvement
21:36 Environmental Impact of Digital Pathology
22:23 Technical Difficulties and Viewer Interaction
24:30 French Authorities on Digital Pathology's Environmental Cost
28:45 Cephalometric Analysis: Digital vs. Manual Tracing
31:46 Exploring Undermined.ai for Scientific Research
43:05 Concluding Remarks and Future Plans

TODAY'S ABSTRACTS & 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. And today. We did the second edition of digit puff digest. I think it was better than the first one. And I wanted to tell you why. Digital pathology newsletter, audio edition that I started. And at the beginning of the year, Died and never like. Was resuscitated. And as the same reason why the abstract review for now, at least. Is going to stay in abstract drive you. Because. Doing it in the more. Sophisticated way is just not sustainable. So, Hmm. I'm going to link to the digital pathology newsletter stuff that. Digital pathology newsletter audio. That I published the beginning of the year and it was well listened. It was not that it had like a lower listening rates than the other episodes where I actually interview guests. And they did it twice and I was so enthusiastic. And then. I just didn't do it again. And I remember when I was talking to Heather Couture one time when she started her podcast. She says, oh, I'm going to be having it every two weeks. And my newsletter is, I don't know if it's every week or every two weeks, which by the way, she has a brilliant news diary about publications as well. What's new in computer vision and she reviews publications in text form. Anyway, and she tells me, oh, this is what I know I can do sustainably. And I said, oh, I want to do every week. And. If you listen to my stuff, you'll know it's not out there every week. It's whenever it can get out. In parallel to life and work and everything. So anyway, she is super consistent. So when I grow up, I want to be like Heather and be consistent. For now. This is my next attempt off on a consistency of showing up for you of discussing digital pathology. Innovation discoveries developments from abstract. And why am I keeping it? To abstract sometimes. There are going to be longer journal clubs. I'm, I'm thinking of opening this. For sponsorships as well. If there is a. Company that has a high impact peer reviewed publication. And they would like me to review it in full I'm open to doing that, but I treat it as an. MVP, minimum viable product, minimum viable. Initiative that's still brings value. So. I actually structured it better. Than last week and last week's episode, you can see on YouTube item, put it on the podcast. Well, I just didn't do it. And also it was less structured. And so we have a very simple structure. We started, so, currently it's it's at 6:00 AM. Eh, E S D. So I show up in the morning. Looking. I have to say suboptimal. There are times where I look better on camera than at six in the morning, but it doesn't matter. And we review abstracts and the abstracts come from a public alert that they set up and there's going to be a video of very soon. I already have it edited. So I will be publishing probably this week and sending to everybody who's online newsletter. So if you are not yet on the newsletter, please join. And you will also get my digital pathology one on one book for free. When you join my newsletter. So, and this is where I'm going to be communicating with you about everything. Basically communicate there about everything. So. I show up at 6:00 AM. People show up. And we talk about abstracts. And that's an MVP. And for now. It's going to stay like that. I show up. I took a little bit. Like a couple of minutes. To let people. Join me. Oh, we start on time. I took just a little bit, a couple of minutes, then we review abstracts. And I will aim for minimum three abstracts. So to review abstracts. To see what they wrote and to like explain it into normal language. And like what actually happens. It takes me 10. Well, maybe seven minutes per abstract. So, We end up having these life streams. You'll see, because I'm going to include the audio version of it. And the video version is of course, on YouTube and also on LinkedIn and other platforms. The best one for sharing seems to be YouTube. So. If you can go on YouTube. Have a look at it to give it a like and share with somebody. I will very, very much appreciate. Medea the structure. I took a little bit, we review three abstract and I talk a little bit more about different things. And it can be books that I'm reading. It can be. Today we were discussing the software that. I'm exploring for literature research. And different things like that. So, Long story short, the digital pathology newsletter was fantastic. But it died because it was not sustainable. I am trying to keep this consistent and sustainable. So. Keep me accountable by common thing. And posting. Whenever I share these livestreams. And have a listen to what we did today. 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:

Good morning, good morning, good morning, good morning, everyone. 6 a. m. I have my coffee and we're gonna be reviewing abstracts and I see you coming in. I'm always so happy, especially today because yesterday was the 4th of July. Today is the 5th of July. Give me a comment if you hear me because I have a new setup. You can see this microphone. Let me know where you are tuning in from in the comments, in the chat, so that I know that you guys hear me. And we can proceed with our DigiPath Digest. I'm going to post a comment myself to say hi to everyone. Let me know where What time is it, actually? What time is it where you are? I want to know. It's 6 a. m. here. Um, and I see my comment, but I need to see something from you to be sure that my setup is working. I see the camera is working, but I don't, I don't know if the audio is working. Give me a yes for the audio. And I'm going to tell you what we're going to do today. We're going to review abstract and we have an agenda today. Official agenda. The agenda is You can see it. Oh, no. What did I do? Um, okay. Agenda. Okay. Yes, people are hearing me. Fantastic. Thank you, Nick. So, the agenda is, I'm gonna talk a little, then we're gonna do the abstracts, and then And then I'm going to talk some more because, um, if somebody doesn't want me to talk too long, then they can just leave after the abstract. So, um, because today is after 4th of July, I assume people in the U. S. are not going to wake up, um, and join us. So, I'm going to send them the recording. So let's just start. Let's just do it. And today we have the abstracts from June 29th. June 29th, which is last Saturday. So I have my PubMed alerts and I did a little bit of, uh, preparation. So I pre, uh, I pre marked my stuff here, right? So And pre mark the, um, the abstracts that we're going to be talking about. So, let's start with molecular classification of breast cancer using weakly supervised learning. And the keyword is going to be weakly supervised learning. This is something that was done in a group from Korea, from Seoul, Seoul, how do you pronounce? I would say Seoul in Polish, but it's Seoul, I think. Anyway, this Korean group, um, Did weekly supervised learning. So, um, obviously they did weekly supervised training for molecular classification, right? Molecular classification. And, but that's more or less what you do with weekly supervised training. So. They, um, because molecular classification is crucial for effective treatment, um, it's important, right? And they did weekly supervised, like usually people do weekly supervised because they don't want to do manual annotations. If you have ever done manual annotations, you know that it's not even consuming. At some point you get discouraged as the annotator and it's not scalable. Although there is, there are companies, uh, that are offering this as a service and recruiting different doctors for different specialties for pathology, including, but also for radiology. So that's what they did. Um, and, uh, they wanted to classify molecular subtypes of breast cancer and they have had two datasets, two whole slide image datasets. Sorry, something else popping up on my desktop, doesn't matter. Um, and those data sets were, were, were from one from one kg from Korea university. Let's see if I can make it bigger. Uh, and one was just hours TCGA that we know this is the cancer genome Atlas that everybody is using for different, different research in digital pathology. So, um, And they did a visualization with an attention based heat map, um, and reviewed the histomorphological features, which is a method that people kind of like reverse Um, I, I don't want to say reverse engineer. I want to say reverse explain, reverse explain. So they see what the model is doing, where, uh, in this case, they have an attention based heat map. It's like with, um, the dog where you classify dogs and cats. I'm trying to change my When you classify dogs and cats and you see that, uh, the model, you have a heat map and you see that the model actually is showing the dog's head. And that's where it took the information from, uh, that it is a dog in contrast to, for example, that was what's, uh, Anant Madabushi once told me, uh, either in the podcast or in the conversation, uh, that, oh, they were classifying husky, uh, Versus the wolves. And this model was so good. It was so good that it was too good to be true. So that when they looked at the attention heat maps, the, uh, husky dogs had, uh, snow in the picture. So the model was focusing on the snow. It's this like a red dot next to a tumor slide kind of thing. The model is gonna, um, focus on the things that are most easy to recognize. So, uh, when you see that the model is actually focusing on things that are morphologically relevant, which pathologist is going to tell you, and that's good because it makes sense. Um, so going back and checking, which is the best view. I guess this one. Yeah. So, um, attention based heat map and they reviewed histomorphological features. So these were the two KG and TCGA where their data sets and their results there, they did the area under the receiver operating characteristics and the value was 0. 749. And they had a challenge because they had an imbalance among subtypes. So class imbalance in general is a problem in those machine learning models. Um, so they merged the two data sets, they merged the data sets and the resulting model was improved. And also they were happy because the attentive patches correlated well, uh, with the results. Widely recognized morphologic features. And what were these features? These were, um, for triple negative subtype, triple negative site subtype was, um, oh, high incident. I say, Oh, because the focus on my screen improved high incidence of, uh, high grade, Tumor nuclei, there was necrosis and intramural tumor infiltrating lymphocytes, intramural, I mean, sorry, intratumoral. So in the tumor, in the tumor. And there was the other subtype, luminal A subtype, subtype, and this showed a higher incidence of collagen fibers. So these are things that happen in those cancers and they were happy because the model made sense. And this is fantastic. So moving on to our second abstract, we have clinical implementation of artificial intelligence assisted detection of breast cancer metastasis in sentinel lymph nodes. And of course, the, um, um, They have this funny name, Confident B, single center, non randomized clinical trial. Can you imagine? So this is a group from Utrecht, the Netherlands, Utrecht. They did this. They have a clinical trial. They have this. So, um, let me tell you about the, uh, detection of breast, uh, metastasis in, I need to find a key, key shortcuts for switching between because this is annoying for me and for you. I bet. So anyway, um, 2017 and 2018, um, 16. There was this famous chameleon challenge. Chameleon challenge was detecting breast, um, cancer metastasis in lymph nodes. And this is something that is happening on a regular basis when you do the diagnostic process of breast cancer. And these challenges, uh, there were different groups challenged, challenged to which, Algorithm, which group is going to make the best algorithm to detect those cancels? Give me a, uh, yes, if you've heard of this challenge. It's pretty famous. If you haven't, that's okay as well because you now have, um, but I promised I'm going to be talking some more later. So, but give me a yes, if you know chameleon challenge or give me a chameleon. However you want to spell it, the spelling is funny. So, going back to this abstract. So, they have something. What do they have? They have an AI assistant detection in a clinical trial. I think it's a big deal. Also, if you look where it's published, it's nature cancer. This is big deal, right? New Drift has this, um, has this clinical trial. So let's see what it did. Of course, assessing sentinel lymph nodes for breast cancer metastasis is a treatment guiding So treatment guiding, we do it to determine, uh, what the treatment is going to be and um, labor intensive, right? You have to look for a little tiny foci of epithelial cells, which are the tumor cells in the lymph node that is full of, uh, a lot of lymphoid cells, like a bunch of lymphoid cells, uh, together. in a sheet and then you are looking for an epithelium. So it's costly. I'm like, why is it costly? Because you have to stare at it a long time. No, it well as well, but it's costly because, ah, sorry, this appeared in my paper. It's costly because you do IHC and IHC costs money. So, um, you do it in morphologically negative cases. So the moment you don't find it, you have to do IHC. Um, and. This is a clinical trial, right? And in this clinical trial, uh, an AI assisted Workflow for detecting breast cancer metastasis in Sentinel, Sentinel lymph nodes was studied. When was it done? It was done from September 2022 to May 2023. So it's less than a year. We had 190 specimen, 190 lymph nodes in this trial. And we had, um, We had an intervention arm, let me use some, maybe green, intervention, sorry, that's okay. Intervention arm, 100, uh, people or 100, 000 of those, and the control arm. And what happened? Uh, in both arms, we used Holzlet images. So digital Holzlet images of H& E sections, uh, where you were assessed by an X ray. Expert pathologist. I think if you're a pathologist, you're an expert. Expert pathologist is even higher level. Anyway, uh, and, um, this pathologist was assisted by the metastasis detection app from Desiopharm. Here, let's clap and congratulate Visioform on making, like, they are in a clinical trial. How cool is that? And Visioform, you can use it, you can buy it. So you can have a tool that was in clinical trial, this app, and I don't know if they're selling this app. We will ask them. But basically, uh, in, in the, so in the non intervention arm, it was just visual, just the pathologist. And in the intervention arm, it was, nobody knows chameleon, by the way. Then give me a no so that I know that you're still, uh, listening to me. And so the, um, metastasis, dete detection app was in the intervention arm. And what were our results? Um, first we had a primary endpoint and it showed. Reduced adjusted relative risk. I don't really know what this metric is, but it's, uh, uh, reduced adjusted relative risk of IHC use. Um, honestly, I don't know this metric, but it showed reduced. adjusted relative risk of IHC use. Um, and it was 0. 68, 95 percent confidence interval. Uh, and for AI assisted pathologists, right? And then, uh, there was a cost saving. They actually calculated that it was 3, 000 less Uh, well, they saved$3,000, uh, not dollars, euro, because that was Rift. Um, it was, uh, and so they saved money with AI and the secondary endpoints showed that there was also time reduction and, but they didn't write the abstract. How much time reduction. Hmm. That's okay. And then 30% improved. Sensitivity for AI assisted pathologists. This is huge. 30 percent is a lot. Um, from the cases that you, so, so 30 percent didn't have to have IHC anymore because you already saw it on H& E with the app. That is big. I think this is, this is a lot. So, We have the, um, this particular paper that says, oh, AI saved cost and the pathologist was more efficient. And then we have this other abstract that I thought was interesting to include today because I don't think we talked too much about that. And I. I kind of have it in the back of my head. So the title of this abstract is the environmental impact of digital technology and artificial intelligence in the time of digital pathology. Um, and this particular one, uh, where's the rest of my, of my abstracts, I don't have it. That's okay guys. Let me find it in my, give me one minute. I'm gonna stop screen for a second and I'm going to search it for you. Okay. In the meantime, tell me what time is it where you are. Tell me what time is it, and where are you tuning in from? And I will quickly find my other abstract. Maybe I will not have my markings. It can be that I will not have my markings. No, I have them. Okay. One second. I have everything. I don't know why I didn't open. That's okay. I'm gonna open this one. Maybe it will work. Okay, there are people in Dublin, in Europe. I was, I was suspecting Europe today. I was suspecting a lower um, I don't know. Okay, is it working now? Yes, it is. Sorry guys, I don't know why this app did this to me, but we can present it right now again. Uh, share screen, and we're going to the entire screen, and we're going to our abstract. Apologies for this. And you guys are 11, 17 in Dublin. I assume in the UK it's the same. Thank you. Fantastic. Thank you for joining me. Okay. So we have our abstract and we said, now we just said that, Oh, clinical trials said you can save 3, 000, 3, 000. And, uh, it's all fantastic. 30 percent improved sensitivity, but I have to check 3, 000 euro like per case or per what? Because if it's just 3, 000 euro, that's not too much. That's okay. But then we go and the French authorities say that we need to be mindful about the environmental impact of all this. It's not at zero cost, right? So, um, this is, this is a group from France and. I have my special comment. Let's see if the comment opens, but basically what they tell us is that this is the future. We're doing digital pathology, but actually we need to be mindful about this not being free. It's not a free lunch, right? There's never a free lunch. I wish there was. So, um They say, of course, they bring, uh, up the global warming, overstepping of planetary limits, threatening human health and the functioning of the healthcare system. And this functioning of the healthcare system, I think, uh, it's probably not specific to France. I've been reading several books about us healthcare system and that it's like semi collapsing. Um, but anyway, so the French government's, uh, Delegación Mixta. Is saying that, um, in spite of everybody saying, Oh, digital, uh, pathology pro, uh, this is the materialization. So like, let's say we can, um, maybe skip IHC so we don't have to, uh, use so many reagents. Uh, we don't have to make so many antibodies or different things like, right. We can predict molecular properties, properties from tissue, all that stuff. Actually, it is. Very material industry. You say it's very material and it generates greenhouse gas emissions. It also consumes water and mineral resources and has social impacts. Um, and the, the digital sector is impacting, uh, at every stage of manufacturing equipment, use of it and, and of life equipment, and it can only be recycled in a very limited extent. Um, and. That they say that we need to understand consequences and, uh, phenomena such as rebound. Let's see if it's going to open my note. It might not, but basically I looked up what the rebound effect is, but it is, uh, the effect that, um, that they're the lower the energy costs and the higher the efficiency of the energy. The higher the consumption of it. So, actually, like, you don't save anything. I, I don't know, you can probably like, um, compare it to the accessibility, like the cheaper the food is, the more people eat and the more obese there are, and then you have a problem. So here, I don't know if it's a good, good analogy, but basically everything is getting more efficient, cheaper. And then you start using it more. So actually the cost is the same. So that's what they are saying to be mindful of, which I think we should be mindful of. Right. So they ask us to do the implementation of a sober, responsible, and sustainable digital pathology. How are we going to do this? I don't know. Maybe they said in this paper, so we would need to read the full paper, but Um, I don't know. I think we are still like at the very beginning. Everybody wants to make it happen. So This is at least not that much talked off out of those ten abstracts. There was only one of this I mean, it's a niche topic And I have a lot of abstracts, but we're not going to do all of them. I think we're, I will usually, uh, just restrict myself to three and then I'm going to show you something else. Uh, so one was, let me find it, uh, here, this one, digital versus manual tracing of Cephalometric analysis, a systematic review of meta analysis. This is not strictly digital pathology. This is actually cephalometry is, uh, measuring a different, like, morphological, like shape of the head in, um, in x rays or in tomography. So this is cephalometry. Um, and what they did, this is a journal of personalized medicine. I checked it's a 3. 4 impact factor. So I was compared to my toxpath journal where all the toxicologic pathologists published their stuff. And toxpath maybe is too, anyway, not like super high, but it's super niche as well. So I don't, I don't. Say it's like worse than higher impact factor. It's just super niche. But what happened here? And actually, the point of this abstract is not this cephalography, sorry, cephalometric analysis, because as pathologists and digital pathology professionals, we might not care. But, um, what they did, they did, uh, because this was a meta analysis, they did a systematic search using keywords. Took keywords, right? They did digital and manual and cephalometry. So they wanted to have all the keywords in that's why they use and, uh, operator. And they searched a bunch, bunch of different, uh, databases, right? Some Medline, every everywhere scope was everything science tech. They searched, um, and they had results. Their results was okay. We found 20 studies and. Between the 20 studies were between 2013 and 2023, and that conclusion is. There is a trend, sorry, the, the, the, it revealed trends suggesting that, uh, digital tracing may offer reliable measurements and specific cephalometric, of specific cephalometric parameters, efficiency, and purity. I love this. It revealed trends the digital trains may offer. So when I was studying for my boards, um, part of this is keeping up to date with the, the veterinary pathology literature. And, uh, we had kind of like criteria when to read the paper, uh, and study from it and when not to read one. And when you would identify these words, um, trends and, um, it would be like, no way I'm reading this paper. But what I wanted to show you, um, is, uh, because they did the keyword research, I wanted to show you a new tool that I'm exploring. I actually, um, talked to the people who, uh, designed this tool. So let me share a different screen and I'm going to find you this tool. You can find it, uh, as well. It's called undermined. ai. Undermined. ai. It's, um, You have to make an account to actually use it, so let me share. Don't go away, I'm sharing, I'm sharing. Hit the screen, and we're going to share this tab. Okay, Undermined. Sorry, this content is no longer available. Am I streaming? Guys, let me know if I'm still streaming. I confused my tabs here. Um, I confused my tabs. I'll find myself. Okay, I am streaming. Sorry. I am. I am. That stuff happened. Sorry. Uh, anyway, so this undermined, I want to show it to you. This is like a different, uh, it's not really PubMed or, um, I'm so confused with the terms. Anyway, let me focus on this. So, um, the, there is, um, they call it the radically good scientific search. So scientific search in, uh, Literature, I would call it literature research as opposed to keyword research. And that's how the founders, uh, are also positioning it. So, um, search now and, uh, I already have an account, the account. So when I look at price, it is pretty raw and what, and, um, at the free one. So I have some searches, but let's, let's, uh, I will check this stuff. Yeah, search now. Let's, let's search now. Um, for example, for, uh, what tools have been tested, what AI tools, what AI tools have been tested for, Uh, detection of lymph node metastasis, and I'm making spelling mistakes in breast cancer. Let's see what it says. So it's going to tell me, um, what that it's understands what I'm looking for. Uh, uh, And it tells me to, like, have a better query because this is like a large language model based, this is AI based. It searches the databases with an AI model so that it's more, um, semantic search rather than keyword search. Um, well, I'll just repeat my stuff and let's see what it tells me. Um, many AI tools. Can you specify? If you're interested in AI tools tested in histopathological images, radiology, radiologic, radiologic scans, or other types of medical imaging. I love it. Okay. I'm interested. I'm interested in histopathology, pathology. Okay. And it gives you like, uh, this I want to research articles that have tested AI tools for detection of lymph node metastasis in breast cancer using histopathology images. Like, uh, I don't even know what keywords I would use, probably AI, lymph node, metastasis, but here I can just, um, just ask a question, right? So, um, because it's a large language model, AI based, um, thing, it takes some time to, uh, generate, um. A bunch of different papers that are relevant. I'm going to show it to you in a second. Um, and let me make myself, let me show you this in a different way. Actually, let me show you what it's doing. It's basically not doing anything. It's, it's, it's doing its search. Um, thank you for, for the confirmation. Uh, and, uh, I want you to. To wait, and it says it typically takes two to three minutes to generate deep systematic search. So what I would use this for, um, so now I have the PubMed alerts and actually I'm going to be publishing a video how to set them up. So I'm going to show you, if you guys are on my list, then you can make this email for sure. And if you want to share it with other people, I would very much appreciate it. So keep receiving those. And, um, this week I was already, uh, more prepared, right? I had my, uh, my abstracts pre marked. I had 30 abstracts picked. So, um, my way of, um, of getting better is just going to be doing it. every week in the morning, maybe later. I don't have to do it in the morning because yesterday we finished, uh, 4th of July celebrations at 11 and I'm, I was going to sleep and I'm like, okay. And then I remembered, no, I need to do the journal club. I need to do the abstract. So, uh, during the night I was waking up not to oversleep. I'm like, there's going to be people waiting for me. I don't care how many, uh, I need to show up. So. So, I did that, I pre read the abstract, it, um, it was, uh, pre marked. The next thing I want to do is maybe, like, um, highlight which journal and, like, what impact factor the journals have, or something, like, to rank them in one way or another. I might be doing it, or not, uh, because I might just go by personal preferences and, uh, you know. You go with it or you, uh, have your own alerts, uh, which there is gonna be a video how to do it. And you can also put your own keywords. The keywords I put in were, um, artificial intelligence and digital pathology. So everything that has artificial intelligence and digital pathology. Sometimes, uh, you know, last week we had the plant pathology and now we had, uh, radiology and Paper included, that's okay, right? But that's keyword search. And I want to show, um, the semantic search. What I asked those founders is to, um, show side by side, the results from the undermine. ai, uh, Google Scholar and PubMed, and they say, well, we actually did it and they have a white paper was this runs, um, I'm going to show you the white paper, um, that they have a figure, um, About how is it different from the other keyword, uh, based searches, but what I would use it for is, is, um, exactly like for a systematic literature search that is more based on a scientific question. Or when I am starting a project and I want to know, okay, what has already been done? So for example, in this case, okay, it's, it's ready. Okay, let's have a look at it. Um, in this case, let me do this, and I have to go. Let me stop actually. So in this case, um, our question was, uh, search topics. I want to find research articles that have tested AI tools for the detection of lymph node metastasis and breast cancer using histopathology images. Um, and. Oh, I have a lot of, so what happens? It has this, um, it tells me so far I've clearly analyzed 100 almost promising papers and I found 52, 57 that are relevant, meaning that's a lot when I look for talks about stuff, you're like, 5 to 10 that were relevant, which is probably 61 percent of all that exists. Like if this is already 61%, I would probably be done with my search. But basically, um, let's see if we have the one that we actually discussed, the one from Utrecht. Um, let me check which, which, who had, uh, who was the first author? Uh, C. van Let's see if we have Fandoyevet, um, I don't know. But anyway, what you get here is like a topic match is this 100 percent match, then a year. Okay. We can only take the last two years and, um, we can go by year. So was there, yes, this is our paper guys. It found it. Uh, yeah, so PubMed versus Undermined, uh, we at least have a tie. Uh, it's, it has 99 percent relevance and it's very recent. And then the other ones. So when I was talking to the founders and I was like, okay, so I did a similar test, like we did a little test. Okay. I had some. stuff from PubMed and I'm, uh, asking a question and I'm checking, Oh, is it there? And it's there. And they said, well, you found it, you like knew what you were looking for, but actually if you have a hundred percent score, then, uh, all the other papers are also relevant. So here they have lower score. So basically I found how many, one, two, three, four, five, six, seven, eight, nine, that are relevant and it ranks them by relevance. Um, Which they have like an algorithm and this one is so it still zero citation in their database. The new ones have zero citations, but they're super relevant. So that is what we have here. And we will meet. next week to discuss more abstracts. Feel free to let me know, uh, when you're watching this as a replay or, uh, just when you receive emails from me, uh, send me some, uh, papers that you want to have reviewed, um, at some point. So what I'm going to be doing, I'm going to start. So for example, now, um, there was this. Um, whenever there is a commercial entity, a company involved, I'm going to ask them if they want to sponsor a live stream, uh, about their paper. So we might be doing that. Uh, and that's it for today, guys. You have a wonderful day. Thank you so much for joining me. Thank you so much for staying. Stay until the end and I will show up next week with new abstracts and figure out a way to do it better. You can give me feedback. I got some feedback after last week's, um, last week's live, uh, what to do better. Some of the feedback, so, um, some of the feedback was, Oh, go more in depth into one paper. I might be doing this sometimes. Um, for me, The idea is to show up and give like this quick win quick value that also does not require me, uh, to prepare a lot in advance because I've, I'm going to be doing this every week, especially in the morning. Uh, there is no way I can, and yes, last week I already like after the last live stream, I was like, Oh, let me put the presentation. Let me put all those images. Let me like do it better. Okay. And I started building a PowerPoint presentation and then I'm like, no, that is not the point of this particular, uh, activity. The activity is to show up, go on a regular basis. And this is kind of a bite size information. Uh, and. That's how I'm gonna keep it. It will evolve, you know, if we have sponsors that have relevant papers, you know, and it's um, Then we're gonna be talking about them. That's okay as well and I'm gonna let you know ahead of time Um, thank you so much and I talk to you on the next live stream in the next video and in the next email