176: Can AI Protect Patients? Forensics, Pathomics & Breast Cancer Insights

What happens when AI becomes powerful enough to diagnose—not just one disease, but entire fields of medicine at once?
In this episode of DigiPath Digest #33, I break down four new PubMed abstracts shaping the future of digital pathology, clinical AI integration, federated learning, and multidisciplinary cancer care. Across every study, one message is clear: AI is accelerating, but human oversight defines its safe adoption.

Below are the full timestamps, key insights, and referenced research to help you explore each topic more deeply.

TIMESTAMPS & HIGHLIGHTS

0:00 — Welcome & Opening Question
 How far can AI safely scale across medicine—and where must humans stay in control?

4:10 — AI in Forensic Medicine: Accuracy Meets Ethical Limits

Based on a systematic review, we discuss:

• AI advances in personal identification, pathology, toxicology, radiology, anthropology.
  
  
  
• Benefits: reduced diagnostic error, faster case resolution.
  
  
  
• Challenges: data diversity gaps, limited validation, lack of ethical frameworks.
   📌 Source: PubMed abstract on AI in forensic disciplines
  
  
  

10:55 — Confocal Endomicroscopy + AI for Pancreatic Cysts

Researchers trained a deep model on 291,045 endomicroscopy frames to detect papillary and vascular structures in IPMNs:

• 70% faster review time
  
  
  
• More consistent structure identification
  
  
  
• A step toward scalable “optical biopsy” workflows
   📌 Source: IPMN / confocal endomicroscopy AI abstract
  
  
  

16:40 — Federated Learning in Computational Pathology

A comprehensive review of FL for:

• Tissue segmentation
  
  
  
• Whole-slide image classification
  
  
  
• Clinical outcome prediction
   Key takeaway: FL can match or outperform centralized training—without sharing patient data—yet still struggles with heterogeneity, interoperability, and standardization.
   📌 Source: Federated learning review
  
  
  

22:15 — The Lucerne Toolbox 3: A Digital Health Roadmap for Early Breast Cancer

A global consortium of 112 experts identified 15 high-impact knowledge gaps and proposed 13 trial designs to integrate AI across early breast cancer care:

• AI-based mammography screening
  
  
  
• Personalized screening strategies
  
  
  
• Digital knowledge databases
  
  
  
• AI-driven treatment optimization
  
  
  
• Digitally delivered follow-up & supportive care
   📌 Source: The Lucerne Toolbox 3 (Lancet Oncology)
  
  
  

28:50 — Big Picture: AI Expands What’s Possible—but Humans Define What’s Acceptable

We close with the essential takeaway echoed across all four publications:
AI is getting smarter, faster, and more integrated—but clinical responsibility, validation, transparency, and multidisciplinary alignment remain irreplaceable.

STUDIES DISCUSSED AI in Forensics — systematic review examining applications & ethical barriers

1. Confocal Endomicroscopy + AI for IPMN — hi

Transcript

00:00:03

Good morning, my trailblazers. Now, I do need your We start with audio check. Let me know if you're here because I just switched microphones. And there's always a threat that it's not going to be working. So, if you don't hear me, let me know. And if you hear me, also let me know. Um, and I see you're joining. Thank you so much. So, it's 6:00 a.m. in Pennsylvania. Uh, let me stop sharing for one second. I have a couple of different things to share with you. So, let me know if you

00:00:41

hear me. You should. And if you don't, we're going to switch the microphone real time. But before a few more of you joined, uh I wanted to start with a couple of tools because it's the end of the year and you know everything happens uh like you evaluate your year at the end of the year and you're thinking okay I was thinking okay perfect thank you so much for the feedback that you guys hear me well and I was evaluating okay what did I do what could I share with you and obviously if you're on my email list,

00:01:20

you get the updates already. And if you're not, you can get on it right now. I'm going to share the QR code with you to get on my list. H, no, not this one. This one, it's the book. But when you get the book, the free PDF through the QR code, you're going to be on my list. Um, and I'm going to be sending out an email about these couple of tools that I found super super helpful this year. And one of them I'm going to show you. It's the It's a pin. Is it focusing? Yes. So, I have it on

00:02:00

the bracelet. It's an AI powered dictation device that I take to conferences and I talk to it a little bit like a crazy person. I talk to my hand uh because it takes notes and it just saves me so much time. Then you have an app on the phone and uh you can leverage these notes for a lot of things. Um then the second thing that I discovered this year is a tool that uh a friend shared with me. Uh actually I heard about it in a video first. It's called Whisper Flow. Whisperflow is something that lets you dictate stuff

00:02:37

into your computer instead of typing and I'm a fan. I already am on the paid plan and everything, right? I'm a believer. So, um you basically speak to your computer instead of typing. And the third third thing or in America, this is the three um the third thing is uh an AI software for presentations. A software for presentations has saved me uh my life a couple of times this year because uh obviously every now and then I get to present somewhere. I have the honor of presenting and there is

00:03:12

going to be a next conference um that I will be presenting at which is digital pathology and AI global engage conference very soon. I'm going to tell you about it in a second. But the tool that saved my life couple of times for presentations, it's called GMA AI. And uh I'm going to put the links in the notes of uh this podcast, this video. Um but these three I wanted to share with you. And that brings me to the digital pathology and AI conference uh where I'm going to be speaking. So I'm pre

00:03:46

preparing the presentation uh right now. I mean, right now I'm talking to you, but uh today I'm going to be working on the presentation again. Um so, let me share who else is going to be there at the congress and you should be able to see. Now, let me make it bigger for you. Okay. So obviously me happy but a bunch of um digital pathology trailblazers Dr. Leon Pantanoids, he was a guest on my podcast. My colleague from TR River, Lee Bertrand, Monica Lambasani, who was a guest on my podcast and um few more

00:04:34

guests are going to be there. So, I'm super excited uh to join them again and uh also be a speaker there. I want to show you the agenda. Let's look at the agenda very quickly and then we're going to dive into the papers. I have four papers prepared for you today. Um, let's see. Can we do How do we do it? Every week I figure out how to do it, but it doesn't take so long anymore. So, let's start at the beginning. We have I will myself, okay, cuz this is my show. So it's me, but there's also

00:05:21

Arthur Lewis from Astroenica who I worked with uh extensively in the past and recently we started uh interacting again. Monica Lambasani who was a guest on my podcast. Um Orly Ardan who was a guest on my podcast or was she already? I think we didn't do the podcast but she was in a few videos already. Then there's going to be a round table discussion that one I'm going to be leading the other one Sebastian who is a talk path fellow colleague of mine. Um and where is lease? Lease is here. She is a keynote. Um she

00:06:03

has a keynote. Uh oh. Why did it disappear again? Are we sharing? Yes, we are. Can we do this? Yes, we can. Okay. So, Lee has a a keynote deploying AI in nonclinical pathology aligning innovation with industry needs. There are um there is a clinical track, there is a pharma track, there are different things happening in parallel. So, if you happen to be there, I'm going to be there. Welcome trailblazers who are joining the chat. I'm going to be there and I will love to say hi. Even if you see me

00:06:45

walking with the camera, just tap me on the shoulders, say hi. I know you from wherever. Maybe we know each other in person. Maybe we know each other from uh the digital space. H you are my trailblazers and I always want to hear from you in person. And now you're going to hear from me about the papers I prepared for you. H if you have any questions, drop me a comment in the chat. If you like it, drop me a comment in the chat. just let's um have this chat a little bit more lively. Okay. I

00:07:15

like when you send me messages. I just like it. And um [clears throat] before we start with our first paper, you know what happened today? I always have these live streams obviously early 6:00 a.m. and everybody in my house is asleep. And today I commented my microphone. I have this little note from my husband that is very sweet that he loves me. So that was a nice like pre Christmas thing that he did for me that um I just found in the morning. So now back to science, my friends. Back to science. And we're starting with

00:07:56

a systematic review about the evolving role. Let's check the tools. a systematic review about the evolving role of AI obviously in various fields of forensic medicine. I don't think we have covered forensic medicine yet in any of our papers. Uh obviously forensic pathology is part of it. This is from a group in Pakistan. And what these um authors say [snorts] this was uh they they systematically um reviewed the applications and impact of AI in forensic medicine. um and focusing on its role in mimicking human

00:08:44

cognitive processes, enhancing diagnostic accuracy, pattern recognition and operational efficiency across forensic domain. So they did this systematic search um and uh they included keywords uh they used PubMed, they used Google Scholar and they used AI in forensics uh also machine learning forensics analysis and they covered publications um in the not in the last but in the 10 years uh of publications from 2014 and 2024 um and they found 1,000 articles but Only 100 met the criteria. And I have a smiley face here. I don't know

00:09:27

what happened with my eye. Sorry guys. I have a smiley face here because [laughter] I had this conversation with my husband about some meta analysis of something and I'm like, "Oh, how many publications met the criteria?" And it just reminded me of this conversation that we had. And he says, "Oh, yeah, that's true. you like analyze thousands of papers and there's like five or 10 that actually meet your criteria. So here we had more. We had hundred that meet the criteria after

00:09:55

screening for relevant study design and quality. And what were their results? Well uh first the AI applications were categorized in key domains personal identification forensic pathology which obviously is close to our hearts. Uh but also we like radiology and imaging, digital forensics and I well I mean I never heard of digit I I knew what digital forensics are but my personal connection to digital forensics is I actually have a friend who studied this and um and she does digital forensics in like all kinds of spectrum of digital

00:10:37

forensics not in not only in the medical space then toxicology forensic anthropology ology, anthropology, sorry. Uh, machine learning, deep learning, and neural network models demonstrated improved improvements in accuracy, reproducibility, and efficiency compared with conventional approaches. Um, for example, AI assisted imaging techniques reduced interobserver variability in post-mortem fracture detection. And then uh we had predictive models for postmortem interval estimation showed mean error reduction in up to 15%.

00:11:18

Um but obviously we have some challenges and the challenges are small non-representative data sets, limited external validation and ethical concerns. Have we heard that before? Oh yes we have. They probably have this as concerns in every single journal club, uh, every single digipath digest that we're doing. Um, do you have any ties to forensic pathology, forensic science? Uh, let me know in the chat. And, uh, yeah, the conclusion is that AI obviously enhanced multiple areas of forensic practice. It improved uh,

00:11:55

diagnostic capabilities, streamlined workflows, supported decision making. But we need wider adoption. It requires rigorous validation, standardization, and ethical oversight. Um, apologies for my like bad handwriting here because I actually want to share this with you. So, I need to I need to do a better job highlighting. Uh, and let me know in the chat if you are interested in getting this uh annotated PDF from me. H if you are then I'm going to share it. If you're not then, you know, who cares? We're not

00:12:32

going to share it. Um so yeah, what what do we need to do in the future? Obviously, um well, integrate multimodel data. So that that is for not only forensics but also for like for all healthcare. Um expanding data diversity and addressing legal and ethical implications. I love how AI connects pathology with other sciences like through this technology um and in general through technology I believe that we uh start doing less siloed work. So that's another uh advantage of these technologies that we're just discussing.

00:13:13

And now let's move on to artificial intelligence advances digital pathomics for confocal endomicroscopy diagnosis of pancreatic cysts. So confocal and endomicroscopy diagnosis. Confocal endomicroscopy diagnosis. That was an interesting one as well. And this is a group from Columbus, Ohio. division of gastroenterenterology, hepattology and nutrition and some other divisions um from that group. And so here, oh, and you guys are interested in the annotated PDF. Fantastic. This is so cool. Then I'm going to share it with

00:13:55

you. Thank you so much, Jason. Um so here endoscopic ultrasound guided needlebased confocal laser endomicroscopy also abbreviated as NC um needlebased confocal laser endomicroscopy. It was a new technology for me. It was the first time I learned about it but they did a cool thing with AI. So this provides real-time optical biopsies enabling diagnosis and risk stratification uh on intraductal papillary mucinous neoplasms and these are um in the pancreas. However, the clinical implementation obviously is hindered. Why is it

00:14:46

hindered? Because image review and interobserver variability. Hm. I uh feel like we've heard that one before as well. [laughter] Let me know in the chat if you agree. Uh so they decided okay AI is going to do a better job and being more consistent. They did AI models to accurately detect diagnostic structures. So this is um where will they be detecting these structures in a video. So they make a video they like put this confocal probe inside through a needle through a needle because it is needle based com focal

00:15:23

laser and the microscopy and they make a video right and I thought this was fantastic because obviously AI is leveraged for video in surgery and other spaces um of medicine and we don't talk so much about it because our modality and by ours I mean the pathology uh trailblazer modality is uh hisystopathology images, right? So we talk more about images than um other modalities like video. So they what they did they u diag they um had participants with definitive IPMN. So this intraductal papillary

00:16:03

mucinus neoplasms um with this diagnose and they were selected from prospective studies 2015 to 2023 and then two observers labeled endomicroscopy images used to develop AI models to detect informative segments. uh and then performance was assessed with the area under the curve sensitivity, specificity and accuracy. And in 66 endom microscopy videos of IPMN, listen to this. 291,045 frames were analyzed. I have a smiley face here as well because it reminds me of like how many patches were analyzed

00:16:48

in the hisystopathology image. Um I mean we and my question is always okay like how many slides. So here we have uh how many videos 66 videos and a lot of frames were analyzed and uh what they um saw here it was 15.5% showing papillary or vascular structure and 84 uh.5% showing no structures. Um and they had four classification and segmentation models tested and they define classification as pattern recognition and segmentation as image division right they divide um the image into relevant uh areas

00:17:32

and uh they had this uh binary detection outcomes which was papila versus everything else and turnary I don't even know this word turnary but basically it says that you have three categories um that had papila vascularity and nonstructure so non structure there was no structure um and then they had this one classification model dino v2 vitg let's just abbreviate it and call it dino and this one outperforms everything with an area so for the binary outcome for the papila versus everything else um and

00:18:13

it achieved area under curve of 0.942 sensitivity 80.6% and um specificity 90.6% 6% and accuracy 89.3. And for the turnary outcome, uh they only did classification models. They didn't do uh segmentation uh because it was impractical to segment vascularity uh there which I think okay if it's impractical then let's do something else. And Dino, our friend Dino also outperforms uh uh everything else. Uh and it was detecting papila, vascularity and nonstructure. Uh you know with uh different uh metrics

00:18:59

that were over 80% uh and the conclusion like the important thing is here like what was the benefit of using this and detecting this stuff? Well, when you look at these videos, you need to evaluate structures, right? So there is no point in you looking at the frames that don't have any structures. So our dino model reduced the video duration to 1.85 minutes of high yield structured containing segments saving 4.27 minutes which is 70%. So it basically saved 70% of the time uh the evaluator would have to look at this

00:19:43

video. I think this is brilliant like efficiency pure because you don't like take any away from the uh from the I don't know where the pathologist or gastroenterenterologist I guess uh from the doctor right the doctor still evaluates this stuff but you cut out all the fluff out of this video and um obviously okay the question is how good is it and we have the metrics how good it is uh but the task is pretty like okay you have uh you you uh detect the structures and the frames with structures and frames without structures

00:20:20

and you just let the um let the doctor look at the frames with structures. So that's kind of parallel to uh what the well goal I don't know you know there is a discussion on that but the goal in pathology would be okay you screen the um the images for which ones are have some diagnostic um stuff and which don't and you like don't look at the ones that don't but a sorry a better analogy would be okay you find the patches uh and that's what's happening in the computer aided diagnostics and pathology as well

00:20:58

you find the patches that have something relevant and make the pathologist look at the patches and you know you spare them and if you have the diagnostic information there there's your diagnosis you don't have to look at the rest of the image h unless there are like some other things that you have to evaluate but most of the time you can uh point the pathologist to where the change is that they're supposed to evaluate uh boom slide evaluated so this they did the same for videos in this conf focal

00:21:24

uh microscopy and this is in vivo so I always am excited about the stuff that's in Vivo because you don't have to like well fine needle laser endomicroscopy. So it's not even a biopsy, it's an endomicroscopy. How cool. I like it. Let me know if you like it. Let me know what you think about it. And then we have federated learning. So for federated learning um if you want to learn more about federated learning there is a dedicated podcast which I'm going to find for you in one

00:21:59

second but um there was a podcast that I had this year I think uh and we talked extensively about federated learning and this is a method that lets you train models in our case obviously for pathology without creating huge aggregated data sets. Um, so let me just share which podcast that was. Give me one second. And I'm going to show you which podcast it was. Maybe I can just federate it. and not Federate Federate Court. We had a H. Why don't I see it? Let me know if you listen to this podcast.

00:23:19

And why don't I see it? It was Oliver Sana. Let me find Oliver. Yes, we have Oliver. Okay. Sorry for keeping you waiting, but uh that's the one here with Oliver Sana and um we talk we talked about swarm learning. It's a type uh it's even like more um sophisticated than federated learning and we were comparing and contrasting it to federated learning. Um and this is even more decentralized than federated learning. So going back to our paper means I have to on federated learning. Okay. So

00:24:27

let's stick to the literature. This is a publication coming from New York University at Buffalo. Sunny, sunny. Okay. Um, so, uh, AI is obviously powerful technique. I think I overused the word obviously. Let me know if you hear any other words that I overused. So, for 2026, I can make an effort to h use them less or if you like them. We just keep them. But let me know in the chat if I overuse something. I definitely overuse. Fantastic. Uh, amazing as well. However, uh traditional centralized learning

00:25:06

methods aggregating large and diverse data sets um they have privacy and security risk especially in healthcare. There is this concept of federated learning offers a promising alternative. And why is this disappearing on me? I don't like that my screen is disappearing on me. Apologies for that. That's the first time we're experiencing this. Okay, back. Um, so this is the federated learning is promising alternative uh by enabling collaborative meth model training without the need to share raw

00:25:56

data. So what is collaborative model training that the model kind of goes to the data instead of being trained on a bunch of data. So you can have this data decentralized in different institutions and then the model goes there and learns the weights and just comes with the knowledge so to say but not with the data and they examined the current state-of-art federated learning. They did a systematic review uh on federated learning in healthcare and they focused on image-based applications relevant to

00:26:30

computational pathology. Um and uh they focused on studies utilizing also computer tomography uh obviously whole slide um histopathology images magnetic resonance uh and then imaging computer tomography and posetron emission tomography. all different imaging modalities including host light uh hystopathology and um they evaluated also evaluated recurring technical challenges um system uh like system and data heterogeneity privacy preservation mechanism and communication efficiency. So this model

00:27:11

will have to and like in this project there is going to be a lot more communication uh than in a centralized project because it's going to be a bunch of different places where the data is located and you have to communicate like infrastructurally but probably also like in person somehow uh with them and it's a challenge. So uh the literature demonstrate a growing adoption of federated learning across healthcare applications. Um and uh the studies report promising outcomes for tasks such as patient outcome

00:27:46

prediction, disease classification and tissue segmentation. Um, notably, federated approaches often match or outperform centralized models in terms of accuracy while maintaining data privacy across institutions. Um, and in the case of computational pathology, it was feasible and effective, but we still have challenges, right? uh and the challenges we already mentioned them but it was data modality heterogeneity communication overhead slow model convergence um and there are several papers that propose novel federated learning

00:28:27

frameworks to address these issues uh but standardization remains limited so yes we have heard that one before as well but if it's a challenge it's a challenge um so Here the conclusion is obviously it was a review. So the conclusion is that it holds significant promise h enabling secure privacy preserving collaboration in healthcare particularly within computational pathology. H and we have these key challenges like system operability, data hogenity and model interpretability. Um, and they say that future research

00:29:10

should focus on developing scalable standardized federated learning infrastructures. And I have here a little question mark because I'm like really should future research focus on that or maybe future research should focus on something else. But obviously we're talking obviously. It's like the fourth time that I said obviously. Please catch me saying obviously the next time. Uh, anyway. So um yeah, why do I even have this thought? Because uh recently I interviewed uh guests from the big

00:29:44

picture, Julie Bulair and Yarun Fanderlac. They were there join me again after I think three years for another update on the big picture consortium. And the big picture consortium is no I wanted to show my face. The Big Picture Consortium is a an effort in Europe to aggregate a bunch of slides. So, um I'm trying to show myself in a different way. Just ignore my pauses. So, to to aggregate a bunch of slides. So and they had um some like thoughts of uh embarking on different initiatives and uh I think they mentioned federated

00:30:28

learning as well and they said no we decided that was not our priority. Uh so it's always a question of priorities uh for for research projects right now we have one more one more and then we're done. So don't leave stay with me at the Lucern toolbox uh digital health and artificial intelligence to optimize the patient journey in early breast cancer a multid-disciplinary consensus. So this is interesting because um Lancet this is in Lancet oncology and it's another consortium. So the big picture

00:31:08

that I just mentioned and there's going to be a podcast dropping around the 16th of December. So is that next week already or the week after? um the week after the DPA conference. Uh so this is a consortium as well this looser toolbox 3 and obviously there was obviously and there was a um okay my my resolution for 2026 uh stop overusing obviously there was a toolbox one and two so that's why we have now toolbox 3 h and this lousern toolbox 3 initiative addresses the pressing need for evidence-based integration of digital

00:31:49

health and artificial intelligence technologies in early breast cancer care and this is a multidisciplinary consortium and it uh one second. This is a multidisciplinary consortium that identifies and prioritize 15 crucial medical knowledge gaps across the patient journey from diagnosis to treatment and survivorships. So this is like a different approach different from what we usually review but they identified knowledge gaps and and they used a modified deli consensus process. This is a consensus process

00:32:32

that has uh defined um ways how to reach consensus and they had 112 unique members from 27 countries and 16 medical societies, trial groups and patient organizations. And these knowledge gaps included AI based mimography screening, personalized screening strategies, digital knowledge databases, AIdriven treatment optimization and digitally delivered monitoring and supportive care. And then and it's small guys. You need to tell me that you want to dig. And so then they developed 13 trial designs in the population intervention

00:33:23

[clears throat] comparison and outcome format to address these gaps. And uh they achieved consens consensus or majority vote in 98% uh of statements. And the recommendation emphasis precision medicine, patient centered care, interdisciplinary collaboration to improve outcomes, efficiency and equity in breast cancer care. Um, and by presenting a road map for actionable trial trials, the Lucern toolbox sets a foundation for advancing digital health in breast cancer care. So this is to me it's a different approach. uh you you

00:34:06

basically are like are are identifying okay where do we need more work in these areas and then how to contribute to closing those knowledge gaps um which is pretty close to my heart and that's what I'm going to be well not exactly that but uh a similar topic is going to be the topic of my digital pathology and AI congress in Europe it's going to about uh how knowledge empowers uh the advancement of uh digital pathology. How uh it all starts with learning about the topic and anybody can do that and you

00:34:46

obviously you are doing this as well. Uh because you are joining me every week here. Uh this is our 33rd Digipath Digest. We have met 33 times in 20 25 I think or did we start last year already? I would have to check. Um you can let me know in the comments when we started and if you have been in the first ones and how they compare to the last ones. I hope I have improved. But honestly I didn't watch the couple of first ones because it was I was intimidated and I didn't know if you're

00:35:28

going to be showing up if you like it. And uh I always have people in the morning, a group of people, some are regulars and then I always have people viewing it uh as a recording like as their lunch routine as their like somehow of some some kind of um digital pathology knowledge acquisition routine which I'm extremely grateful for. H and if you want to learn more or if you are just starting your digital pathology journey then of course there is the book digital pathology 101 h and the book is

00:36:04

being updated like for real the my goal for December is to update the book I already like updated the outline uh the the main updated chapter is going to be the chapter on AI uh which has changed a lot since 2023 the other aspects did not change that much, but there is an update coming uh and hopefully I can launch it with a splash in January. So, if you grab this book right now as a PDF, you're automatically are going to get the updated version. So, if you don't have it yet, get it. And then you're

00:36:40

going to end up on my mailing list and get all the digital pathology content that I'm creating, invitations to these live streams. And um probably in January or maybe like after Christmas uh a list and uh affiliate codes to these tools that I mentioned at the beginning including this pin that helps me. Come on pin focus. Why is pin not focusing? Yeah. No. Um it takes notes without taking notes because it takes verbal notes. So if anybody is interested in the PDF of today's um digit path digest, let me

00:37:23

know in the comments even if you're watching the recording. Ah [gasps] one more thing, one more little thing. So we are still in the like tail end of the cyber week I think. H and we do have uh a code a discount code uh in the digital pathology store. uh you can find uh courses there. You can find the audio version of the book and different things and also our ear the earrings. Um so the for the earrings if you're it has been gifted several times and if you're thinking of getting this

00:37:59

as a gift for somebody for Christmas then you would have to let me know today. You would have to buy them today because uh on Saturday I'm going to Poland. So, I'm going to be broadcasting or writing to you from Poland uh starting next week and then I'm going to be going to London from Poland. So, today and tomorrow are the two last days that I can actually ship it to you. Um and if you order it after that, but the discount code, let me give you the discount code. Um let's see what is our discount code here. And if

00:38:36

[clears throat] you already have it, let me know in the chat what the discount code is. Like I should know. Okay, it's easy. It's cyber 35. Easy discount code because the discount is 35%. So I'm putting it in the chat discount for the shop for the store. Anything uh that you want from the store is going to be discounted by 35%. And um so if anything resonates or if you want to get uh these earrings that they always put at conferences and I'm going to have it in have them in London as

00:39:17

well. Um you would have to get them today cuz then I'm traveling and it's not going to arrive before Christmas. Thank you so much for joining me today. I am honored to see people live here every morning. You are amazing. You are my trailblazers. So, a lot of positive vibrations for the preh