158: Multimodal Magic AI’s Role in Lung & Prostate Cancer Predictions

What if AI could predict cancer outcomes better than traditional methods—and at a fraction of the cost? In this episode, I explore how multimodal AI is reshaping lung and prostate cancer predictions and why integration challenges still stand in the way.

Episode Highlights with Timestamps:

• [00:02:57] Agentic AI in toxicologic pathology – what it is and how it could orchestrate workflows.
• [00:05:40] Grandium desktop scanners – making histology studies more accessible and efficient.
• [00:08:03] Clover framework – a cost-effective multimodal model combining vision + language for pathology.
• [00:13:40] NSCLC study (Beijing Chest Hospital) – AI predicts progression-free and overall survival with high accuracy.
• [00:17:58] Prostate cancer prognostic model (Cleveland Clinic & US partners) – validating AI-enabled Pathomic PRA test.
• [00:23:35] Thyroid neoplasm classification – challenges for AI in distinguishing overlapping histopathological features.
• [00:34:49] Real-world Belgium case study – AI integration into prostate biopsy workflow reduced IHC testing and turnaround time.
• [00:41:03] Lessons learned – adoption hurdles, system integration, and why change management is essential for successful digital transformation.

Resources from this Episode

• World Tumor Registry – A global open-access repository for histopathology images: World Tumor Registry
• Beijing Chest Hospital NSCLC AI Prognostic Study – Prognosis prediction using multimodal models.
• Cleveland Clinic Pathomic PRA Study – Independent validation of AI-enabled prostate cancer risk assessment.
• Grandium Scanners – Compact desktop scanners for histology slides: Grandium.ai

Transcript

Aleks: 00:00:02 - 00:01:24 Good morning, my digital pathology trail blazers. Welcome to the 25th or let's say 25th and a half Digipath Digest because yesterday I went live and it didn't go so well. I'll tell you about it in a second. Hi I'm Dr. Aleksandra Zuraw. I'm a toxicologic pathologist and a very big digital pathology enthusiast. And the goal of this series is to go through digital pathology literature. Welcome back. If anybody was here yesterday and is joining today again, so much love to you because what
00:00:43 - 00:01:53
happened yesterday was that I looked at the description of my live stream and then I looked at the papers thatI prepared and um unfortunately they didn't match. So um yeah, I I basically ended up giving a few updates and ended the live stream. I do have the same clothes almost. I don't know if I had the same earrings and I don't even know if they match to my red jacket.but it doesn't really matter cuz I want to show them to you anyway because I have them available in my
00:01:19 - 00:02:18
online store. So, if you're interested in looking at them,you can do that at the end. I'm going to give a QR code and when you're joining, let me know where you're tuning in from. Andwhile people are joining, I'm going to give a few updates. So, I alreadytold you what happened yesterday. A little bit of a I don't know. It wasn't really a disaster, but there were people waiting for me to do this live stream. And I don't think everybody's going to come today.but
00:01:50 - 00:03:05
if you are here and you were here yesterday, let me know in the comments. I'm going to send like replay, but if you're here, stay with me. Um so, back here. Let me take my notes. On Thursday, I'm came back. I came back from STP, the Society of Toxicological Pathology annual meeting andit was a very nice meeting. Was great to meet um the the people I know and it was I think my fourth meeting in a row. So now I start feeling like I hey I meet friends because I do. Um but couple of things on the digital
00:02:27 - 00:03:27
pathology front and I see you're joining. So, let me know in the comments where you're tuning in fromwho you are because I want to recognize you so much for coming here on Saturday morning. Like, who does that?people who kind of screw up on Friday like mehave to come on Saturday, but also amazing digital pathology trailblazers from different places in the world. So, let me know which is your place, where you're tuning in from, what time it is. And going back to the update from STP um
00:02:57 - 00:04:18
couple of discussions, couple of presentation one of my presentations um interesting um workshop discussion where I was invited to be that I was invited to be part of was about agentic AI. So Agentic AI and AI agents agentic comes from this word agent and AI agents are AI programs that coordinate other AI programs. And so there was a you can still hear me right. There was a workshop organized by the company called Histtophy. And Histophy wants to helpdrug development help bring drugs faster to
00:03:37 - 00:04:49
patients by helping toxicologic pathologists with agentic workflow orchestration and and that's what we were discussing. Basically they wanted to know how can we help with this? Um, and there was a little bit of a gap because not everybody knew what this agentic AI is and even if they knew and obviously it was explained at the beginning um we didn't know yet okay wherecould it help us how could it fit in our work workflow what our workflow actually is I mean we intuitively know as toxic
00:04:16 - 00:05:23
toxicologic pathologists what we're doing but and it's in the SOPs and everywhere else but um not really like so much out there for people who are not embedded in it to understand it and to understand how an AI agent could help. So what I'm going to be doing this weekend, I'm going to be preparing a mini course about agentic AI. So if you're interested in learning what agentic AI is where it already has place in pathology because it already has place in um human pathologywe are
00:04:50 - 00:06:02
working on it or there are people working on it to um implement it in toxicologic pathology. If this is something of interest let me know put agent or a agent or course in the chat maybe course and and I'm going to prepare this mini course over the weekend for you and let you know about it. Um and then other than thatthere is a new sponsorfor digital pathology place Grandium. You guys know Grandium because they have these little cute very beautiful scanners that you can put on your desk and it will scan
00:05:25 - 00:06:34
your slides. Sowhat I'm working now on is just a basic hisystologology series and the grandium scanner is going to help me with that because I got a demo unit for um preparing a video that we are supposed to do together a demo video. So and I have the new version the four slide version. So now it's little scanner on steroids. It still fits on your desk.oh and I feel I see people are interested in the course. Fantastic. Thank you so much for letting me know. Andso the granium scanner, right?
00:06:00 - 00:07:05
You're going to see it. I have it in a huge box.it's in my other office. This is my YouTube studio. And in the huge box is the scanner. On Tuesday, I have a call set up to set up this thing properly. And I have a collection of slides that I want to slide. And let's see if I can stream that for you because I would just want to sit scan my slides, talk to you. So maybe we can do some Q&A me anything join justquestions in the chat and I'm going to be here answering and scanning your slides. And
00:06:34 - 00:07:45
those hisystologology videos are pretty popular. Um and you don't need to only learn hisystologology, study histologology, or work with tissue. You can also wear it. Let me give you the code for the earrings. It's in the store. I'm going to keep it up there for now. Let's see how it looks with our papers. Maybe it doesn't look great with the papers, then I'm not going to keep it. Let's see. Ah, it's okay. It doesn't really matter. It can stay there. So, if you're interested and there are other
00:07:09 - 00:08:37
courses in this um in this QR code as well, so feel free to scan and check. There is a a more general AI course that is called AI pathology makeover. So um oh and we have people from India joining. Thank you so much. Kudos to me. No kudos to you for coming on Saturday. Me I should have done it yesterday but long here intro. Stop long intro. Let's go to the papers. Let's see if my pen works cuz Okay, everything working for pathology and language analysis. So cost effective is the key word here. And
00:08:03 - 00:09:25
how can we make something like that cost effective? Is my camera moving with me? Hey camera, why are you not moving should move? But whatever. Um, this is a group from China. Everybody from China. Yes, everybody from China. So, and and the title of today's abstracts is multimodel magic. So there is going to be several papersthat talk about multimodality and this is one of them because we already start with vision language models. So vision language see if I can write better. Um so not just
00:08:47 - 00:10:23
vision, not just language but vision plus language andthey foster interactive conversation between AI enabled models and humans. It's I mean ch has the um vision creation already but ch is mostly um mostly language model but there are like visionadd-ons as well and clinic like challenges. Um, but these challenges, right? Because otherwise we're not going to apply
anything. Let me know in the chat where you're tuning in from. And yeah, especially the challengesare related to large scale training data
00:09:42 - 00:11:15
as well as financial and computational resources. And here they proposea new model, a clover, a cost effective instruction learning framework for conversational pathology. Um, so this clever trains a lightweight module and uses instruction tuning while freezing the parameters of the large language model. Um, instead of using costly chat GPT, they propose a well-designed prompt on GPT 3.5 for building generation based instructions and then they fine-tune it with like an add-on module. So everything before stays and you train
00:10:29 - 00:12:07
just an add-on module and it's a lot less training. You don't have to unfreeze and retrain all the weights on of the main model. you do it um on a little module that is added on top of the main large constructed a high quality of template based instructions and and they used two benchmark data sets to check the stress strength of hybrid hybrid form pathological visual question answer instructions and um it. They they claim that this clover out outperforms baseline um that possesses 30 37 times
00:11:19 - 00:12:28
more training parameters and exhibits few shot capacity on external clinical data sets. So fshot capacity is basicallyzero shot you don't need to train a model it answers all your all your questions from the out of the box. few shot is you have to train it a little bit for a specific domain. So that's also an advancement compared to um like a lot of shot learning right when you have to retrain a lot. Um so that's that if you are just joining because I see new people joining let me
00:11:53 - 00:13:17
know where you're tuning in from. just put it in the comments and and also I just said thatthere is going to be a miniourse affordable miniourse on agentic AI that I will work on over the weekend. So if you want that let me know in thechat put agent courseAI agent and I'll get this delivered to you. Our next paper is disease patient level multimodally supervised learning. So cool words here patient level weekly supervised right so we're not talking um strong supervision with
00:12:40 - 00:14:35
annotations and we are also talking multimodel so we used different types of data um maybe text maybe um whatever right we're going to see in a second and I have here guests joining from Egypt fantastic what time is it in Egypt now are you like having lunch and listening to This. [Music] Ah. Okay. This one I think is here. No. Sorry. Do I have it? No. Sorry guys. I know that
my This I didn't want. This I didn't want. H is it this one? Yes. Yes, that's the one. Okay. So, my um apologies for that. What happened is
00:13:40 - 00:15:17
that my um I get my and when there's a lot lot of publications the last publications and then going back to our um patient level multimodel weekly supervised let's see what happened here and we can see that this is a group from China as well and they say that with rapid progress in AI and digital pathology prognosis prediction for NSLC non small cell lung cancer patients and has become a critical component of personalized medicine. So in this study they developed a multimodel AI model that integrated
00:14:36 - 00:16:19
whole slide images and dense clinical data to predict disease-free survival and overall survival with high accuracy for the NCLC patients undergoing surgery. Right? and they had 618 patients from the Beijing chess hospital and the model achieved area under the curve of 0.884 8084 a predicting progression and this this was for predicting progression um more or less the same1821 for in the test set right 0.7 for predicting progression for predicting death. Um, and then they were categorizing p patients into high-risk
00:15:36 - 00:17:06
low-risk groups and the model identify significant differences in survival outcomes. Um, so it also uncovered digital biomarkers associated with poor prognosis. So that offers additional insights into NSLC treatment. And they say that this model has the potential to revolutionize post-operative decision making and with a precise tool for predicting disease-free survival and overall survival. This may be one that I want to read in full. Okay. And it is 1:15 in Egypt. Amazing. Thank you so much for joining. And if
00:16:22 - 00:17:55
somebody just joined, let me know where you're tuning in from. So, this one I might want to read in full because I'm missing some information like um what what is this digital biomarker? Hm. I don't know. Let's see. We're not going to look for it now because we didn't look for it ahead of time. But, um this they're availablefor free. I download them just from the journal website. I click on PDF and um so let's this one I think we already did [Music] come on sorry guys too many tabs for me
00:17:12 - 00:18:36
I just need one and go from up to down otherwise I'm overwhelmed with the amount of literature present on the on the screen. Okay. So, let's do the prostate um paper. I see people joining on YouTube. Is anybody joining on LinkedIn or are the comments just not going through? Um let
me know LinkedIn.raise your hand and let me know if you're joining on LinkedIn.and if I don't see the comments here then if you are I willI will see them on LinkedIn itself. Okay. So what happened here? We have of
00:17:58 - 00:19:37
course AI based digital hisystology classifier for prostate cancer risk. So this this session we're talking about risk stratification a lot and this was an independent blinded validation in patients treated with radical and this group is froming well mostly in the US Cleveland department of Wisconsin we have tish cancer institute mounts sina 9 New York Cleveland Ohio and Seattle. So this group is from the US and the purpose of this was or like the intro was that AI tools that identify pathologic features from
00:18:53 - 00:20:11
digitized whole slide images of prostate cancer and generate data to predict outcomes. and and the objective of the study was to evaluate the clinical validity of an AI enabled prognostic test. So again, we're we're having an AI enabled prognostic test and they call it pathomic PRA using clinical cohort from Cleveland Clinic. So this was a retrospective analysis. They took historical samples data um and they used prostate cancer whole flight imagesfrom patients who underwent radical
00:19:32 - 00:20:58
prostatctomy. My problem is always like they put these abbreviations at the beginning and by the end of the abstract I don't remember the abbreviations. Um but I guess it's an English as English speaker or first language English speaker. Anyway, so they did itthey 222 andthese people did not receive so had decipher genomic testing available. So data modalities and the host images were deidentified, anonymized and outcomes were blinded and the patients were stratified into high risk and low
00:20:26 - 00:22:01
Don't move. Why are you moving? Uhoh. Okay. I don't know what I just did. My thing is so thick. I made it thick, but I don't want it thick. Okay, that's I don't know what I did. Maybe this. I think I did this. Okay, we should be good. So high risk, low risk andonce they did that they um these categories were based on the some predetermined thresholds for pathomic prad scores h it was 0.45 for biochemical recurrence and 0.5 for distant metastasis. So they assigned scores to that and the resultssay
00:21:19 - 00:22:56
that the study included 30344 patients who underwent the radical prostatctomy with a median follow-up of 4.3 years and both the pathomic and decipher scores were associated with rates of
biochemical recurrence free survival BCRFC. Um so the pathomic score was over 45 versus lower than 0.45 sorry over 0.45 and the decipher scorewas over 0.6 versus lower than 0.6 six and they were 16 patients who had and DM was distance distant metastasis. Yes, here and 15 were in the high-risk pathomic group.
00:22:08 - 00:23:27
Both pathomic and decipher scores were associated with rates of metastasis free survival. And despite the low event rates for metastasis, the multivariable regression demonstrated that hypothomic score was significantly associated with distant metastasis. So this um whole slide imagebased score was associated high score was associated with metastasis. So um the conclusion is that these findings independently validate pathomic prad as a reliable predictor for clinical risk in the prostatectomy setting and
00:22:48 - 00:24:24
pathomic pride therefore merits prospective evaluation. So I hope they will do it and we can read it read about it in the nextjournal club or whenever they do it and publish it. So and then we have another one. This one is and do pathology and this group is from India. Look at somewhere. If you're just joining, let me know in the chat where you're tuning in from and what time it is for you. For me, it's 6:24. And we have I think two more papers. This one and anything else? I think this one and
00:23:35 - 00:25:24
this one. We have two. So, stay with me. Don't go yet. We're almost done. Let's talk about this one for um thyroid neoplasms. And I want to show you something. Let's see. for thyroid neoplasms because I know that thyroid neoplasms are there but what I'm talking about is the tumorworld tumor atlas let's say world tumor atlas is that it no world tumor ah shoot I should know word world tumor cancer cancer [Music] I don't know. World cancer. No, I will figure it out by the end. I'm
00:24:31 - 00:26:13
going to I'm going to search for it. Um, what I wanted to show you was the There's a repository of host images from different cancers with diagnosis and um it's tumor sorry if you know let me know in the comments world tumor registry h wrong world tumor registry okay apologies for this world tumor registry I want to show it to you Um probably screen share this one. [Music] Let's see. Yeah, world tumor registry is something thatwhere you have this like tumor collections andthyroid is one of
00:25:36 - 00:27:09
them. Um, okay. Let's accept our cookies. And you can search by diagnosis. And I did read the terms of use. And you can find different thyroid cancers here. So, let this load. Oh, and I have people joining from Algeria. Amazing. And it's 11 a.m. there. Okay. Um so these thyroid like if you have not heard about this website WTR world tumor registry I'm going to put it in the chat and you let me know if you can see the comments in the chat. I just put a very long link. Don't be afraid of
00:26:22 - 00:27:31
it. It's me it's me sharing it.world tumor registry. You can search by diagnosis. You can search by subtype. And like did you notice that I only had to so they are asking me for cookies. I can can I deny I can even deny cookies cookies. Um so I didn't have to register. I didn't have to um put my email anything. I just had to acknowledge that they use cookies which how shall I put it everybody uses right? And you can look at these thyroid cancer and basically like see case information
00:26:56 - 00:28:22
here. um where it geographically came from like super diverse collection of tumors and the thyroid gland is the first organ where they already have a lot of them. Come on. See, I haven't been here for quite some time. So, yeah. Anyway, that was a little bit of a digression and [Music] Why didn't I see? Okay. And I have guests from Nigeria. Fantastic. Africa is over represented today. Thank you so much Africa for joining. And so I bet there are some tumors from Africa as well in this repository. But now let's
00:27:43 - 00:28:45
go back to our papers. We just have two more, my friends, and then we're free to have a fantastic weekend. But it's so funny. I see some more people joining. So whenever you join, let me know where you're tuning in from. And I see Africa and YouTube are over represented today. I don't know if LinkedIn is not going through or maybe LinkedIn people are actually having a weekend and not doing this on the weekend. So um here the three-stage AI assisted approach for accurate differential diagnosis and
00:28:17 - 00:29:50
classification of NFTTP and the thyroid neoplasms. And we're going to see what this NFTP is. This is so-called a non-invasive follicular thyroid neoplasm with papillary like nuclear features features nift nftp and so this was a recently introduced term h and this marked a pivotal shift in the classification of encapsulated follicular variant of papillary thyroid carcinoma. Oh my goodness. Wow. They have some Uhoh. My thing stopped sharing. Why did it stop sharing? What happened? Let me check.
00:29:03 - 00:30:31
Stop and window share. H. When I draw it stops sharing. No, it's working now. Okay, let's hope it keeps working. And I see more and more people joining. Okay guys, whoever is just joining, that's fine. Fantasticfor you to be here and let me know where you're tuning in from.are you from Africa? Because we have a bunch of guests from Africa. Maybe you want to counter with some other country. But going back to our thyroid cancersthat have verysophisticated acronyms and theythis reclassification from
00:29:47 - 00:31:06
the malignant to low-risk neoplasm categorysignificantly reduced over treatment but its hystopathological diagnosis remains challenging. So they introduced a new term the non-invasive follicular thyroid neoplasms with papillary like nuclear features and um they need to like figure out what to diagnose like that.okay more people from now we already have this person from Nigeria. Thank you so much for joining. So why is it challenging? because there are overlapping features with other thyroid lesions and also interobserver
00:30:27 - 00:32:10
variability is always always always an issue in pathology diagnosis. So let's see if AI can help us overcome these key limitations of histopathological evaluation andthe preliminary studies are promisingbut the AI models capable of efficiently disting TPfrom to be developed. So there is not yeta solution for this. This three-stage hierarchical pipeline patterns and nuclear features and prioritized models were trained using a lot from 130. So I operate Does this operate in how many sections?
00:31:31 - 00:32:36
And they did 134 sections from 125thyroid. And why do you need to why do we need need to um like do patches? Because these images are huge and do cute models. Um you always run them on little patches on little squares and thenone after another. Um and um there are new approaches to that as well where you can map the dependencies between patches. Um but let's go back to the paper otherwise we're going to be sitting here till Sunday. So they had these nodules representing follicular
00:32:04 - 00:33:39
nodular disease. Follicular adenoma dominant nodule if infollicular nodular disease invasive vptc that was the um encapsulated follicular variant of papillary and thyroid carcinoma. um and classic and infiltrative follicular subtypes of PTC and the external validation revealed good accuracy at the overall patientwise and classwise level. However, it showed limitations in the differential diagnosis ofthese two things. thethe the newintroduced term the non-invasive follicular thyroid neoplasm
00:32:54 - 00:34:17
np and vptcapologies to all thyroid pathologists for me butchering yourabbreviations and um so this was an expected challenge due to overlapping nuclear features and the absence of incorporating the assessmentof the tumor capsule. Um so basically what they are saying we need to improve and we need to do a rigorous validation. So I am not surprised because if there are overlapping hisystological features of something with something else that are difficult for human, it is going to be
00:33:35 - 00:34:43
difficult for AI unless we have some additional informationthat can give us like can give AI the competitive edge or give AI basically more information to distinguish it not just based on the hisystological features. So let's say there are like some suspicious features and then you take other modality of data genomic data whatever data integrate it and then you have a model that's more powerful than just visual assessment but the visual is going to be challenging anyway so not surprised needs further
00:34:09 - 00:35:29
improvement but they did the work so congratulations this was a group from India fromdepartment of pathology all India Institute of Medical Sciences New Delhi and some other groups. Okay, the last one I have it separately. I love this one because it is kind of like a guidance how to implement digital pathology. If you're just joining,let me know where you're tuning in from. And before we dive into this last paper, a couple of things that I said during the live stream. So there's going to be a course
00:34:49 - 00:35:55
on agentic AI. This is like the new frontier of how AI can help pathologists, how AI can help in medicine. Um, so if you're interested in the Agentic AI miniourse, it's going to be small, affordable. I'm going to be developing it this weekend. If you're interested in that, let me know in the comments. Put AI agent course or a AI agents, something like that. Um, like Ahmed said. Um, and I know in the comments that this is something interesting for you. If you like these beautiful earrings of mine that don't
00:35:22 - 00:36:42
match my jacket, h in the left upper corner of the screen there is a um QR code for the store where you can get them. And nowthis is our last last last paper and I see people joining still. This is so cool. Hello India. We are u most of our people today are from Africa. So fantastic tosee somebody from India and this review practical implementation of AI in a nonacademic non-commercial pathology laboratory real world experience and lessons learned non-academic non-commercial what kind of
00:36:01 - 00:37:07
lab is it then like it's either academic or commercial I don't knowand if you're interested in the course justput the word h course or something that I um what you're interested in because I'm together with my team we're going to be going through the comments. So put AI course, AIagent course and there is an AI course in the store as well called a pathology AI makeover. That's also a relatively affordable course that you can have a look at. Um so yeah, what is
00:36:35 - 00:37:48
a non-academic non-commercial laboratory? Let me know what that is. Hobby laboratory. No. Um I should have checked it ahead of time. So this is a group from Belgium from Adverb and Gant Belgium and the goal here, excuse me, the goal here um wasto show what they did with their AI integration. So um as pathology departments transition toward digital workflow, the integration of AI is anticipated to become increasingly common. Is it so common? We don't know. But some people go for it and this
00:37:21 - 00:38:43
particular lab did. So this study aimed to describe their real world implementation and impact of AI integration in routine pathological diagnosis especially focusing on prostate biopsy. Uh, and this Oh, okay. So, non-commercial is a hospital like duh, Alex, of course. Sorry for this gap in logic for a second. Okay. Thank you so much for letting me that you're interested in the course. It's going to be done over the weekend and delivered to you as soon as I have it ready.so methods and results AI tool for
00:38:02 - 00:39:07
analyzing prostate biopsies was integrated into the department's daily workflow by embedding it into existing laboratorylimbs laboratory information and reporting system and um there was a short adaptation period. I think this is super crucial to have an adaptation period for whatever new technology is out there. It's so funny the towards the end more and more people are joining. Let me know where you're tuning in from. Um, and I think, okay, I see comments from Facebook. I don't see
00:38:35 - 00:39:35
anything from LinkedIn. I think LinkedIn is not going through on my streaming software. But that's okay. I'm going to respond to everyone on LinkedIn directly whom I don't see here. But if you're on any of the other platforms, let me know where you're tuning in from. And they had this adaptation period to let people adapt, which is fantastic. I think that should be part of any um technology transition. Andwhat what they measured they measured the improvements and the one that was uh
00:39:04 - 00:40:20
most notable was reduction in the number of IHC tests required indicating more confident primary diagnosis. How cool is that? That you basically can reduce the number of follow-up tests because upfront you have higher confidence in your primary diagnosis and it's like a direct cost reduction. Everybody is asking like oh how is this AI going to um give return on investment here? This is one of the ways it can h yield return on investment. you don't need to do ancillary tests or you you have to do
00:39:42 - 00:40:56
less of them, right? And there obviously are expensive. So there was also a decrease in turnaround times for biopsy evaluation and highlighting improved efficiency. And the conclusion is that the first year experience of integrating AI into daily pathological practice demonstrated tangible benefits in diagnostic efficiency and workflow optimization. However, they of course had challenges. Of course, we always have challenges. Um, for example, uh, one of them was adoption by the staff and system
00:40:19 - 00:41:41
integration hurdles. These are these are the two main things that that always happen with new technology. So, um, change management, adoption uh, and integration, right? [Music] because there are so many okay I'm like covering my handwriting which is not that beautiful anyway so let me just take it awayintegration so there are so many systems in the lab already right and there is like a workflow that's workingwhich is fantastic h and then you introduce a new tool so obviously system
00:41:03 - 00:42:30
integration is going to be a challenge Um and the lessons learned that theytalk about here was that um that they talk about here will be a valuable guidance for anybody who wants to do AI in their lab and they emphasize the importance of strategic planning training and system compatibility. So this was our last paper. Thank you so much for joining me. Um, one last well, first of all, shout out to everybody who came here on Saturday. You are amazing digital pathology trailblazers. Um, I really, really,
00:41:45 - 00:43:06
really appreciate you. Um, I was your host of this live stream, Dr. Alexandra Zurf, a toxicologic pathologist and a a strong digital pathology advocate and enthusiast. Couple of resources that I want to point you to other than the the course out. If you like read reading you the digital pathology 101 book especially if you're new starting and there is a QR code at the bottom of the
screen this is the one to scan for the book andso this is the paid version that is available on Amazon but there is a PDF version that is
00:42:27 - 00:43:49
absolutely to version start with it. First thing I and if you like physical books this is available on Amazon also.we are working and by we I mean myself and my team we are working on updating the AI chapter of this book. So if you have the currentPDF version of this book, you will automatically get the updated version. Sogo aheadscan the code and be on my email list so that I can deliver the updated version to your inbox. If these earrings are something of interest for you or for
00:43:16 - 00:44:43
somebody who you like or you think that wearing pathology or wearing histologology is as cool as doing pathology, these are available in the store in the DPP store. This is the QR code in the upper left. We also have different versions. We have the is it focusing? multi-ucleated gentel. And we have this these guys. These this is also crypts that look like daisies. And this is something you can give as a gift, as a summer gift or a conference gift. I always wear them to conferences. So that being said, oh, somebody is
00:44:00 - 00:44:51
interested. Fantastic. Thank you for scanning. Uh, that being said, I appreciate you so so much for coming on Saturday. I will not make the mistakes of not double-checkingthe topic of my live stream next Friday because I feel so sorry that the people who were here on Friday didn't actually get the live stream. So, that's on me. I'm going to do better next time and I talk to you in the next episode.