The Future of Cancer Diagnosis: Digital Pathology and AI

Summary

This episode of the Inference show by Turing Post, dedicated to Breast Cancer Awareness Month, features Akash Parvatikar, an AI scientist and product leader at HistoWiz where he leads the PathologyMap platform. The conversation explores the current state and future of AI in cancer diagnostics, with a particular focus on digital pathology - the process of digitizing physical glass slides to enable AI-powered analysis and remote diagnostics.

Parvatikar explains that the field is currently in a transition phase where pathology labs are being digitized while AI tools are being developed in parallel. The challenges are significant: whole slide images can be several gigabytes in size (equivalent to 1-2 tennis courts if printed at 40x resolution), requiring specialized patch-based processing approaches. A major bottleneck is the lack of high-quality annotated datasets, compounded by significant disagreement among expert pathologists on diagnoses - different hospitals and regions may diagnose the same sample differently.

The discussion covers the timeline for adoption (2-3 years for more FDA approvals), the economic case for digitization (digital storage is cheaper than physical storage over 25 years), and the human role in AI-assisted diagnostics (humans will remain “in the driver’s seat” like with Tesla autopilot). Parvatikar emphasizes that AI must “speak the language of pathologists” to gain clinical approval - heat maps and gradient-based explanations are insufficient without pathologist-friendly feature descriptions.

Highlights

”One Whole Slide Image Occupies 1-2 Tennis Courts”

yt-dlp --download-sections "*5:20-6:20" "https://www.youtube.com/watch?v=IJvh0Ws1_i0" --force-keyframes-at-cuts --merge-output-format mp4 -o "IJvh0Ws1_i0-5m20s.mp4"

“If you take a one image print out at a 40x resolution it occupies anywhere between 1 to two tennis courts.” — Akash Parvatikar, 5:35

”The Same Sample Could Be Diagnosed Differently Around the World”

yt-dlp --download-sections "*10:45-12:00" "https://www.youtube.com/watch?v=IJvh0Ws1_i0" --force-keyframes-at-cuts --merge-output-format mp4 -o "IJvh0Ws1_i0-10m45s.mp4"

“Your ground truth data might itself be imperfect wherein there might not be a gold standard ground truth for that particular image. There could be a doctor from NYU giving it a particular diagnosis. But if you take the same sample to west coast, they might diagnose it differently. If you take it to Europe or India, it could be entirely different.” — Akash Parvatikar, 10:45

”AI Must Speak the Language of Pathologists”

yt-dlp --download-sections "*23:52-24:50" "https://www.youtube.com/watch?v=IJvh0Ws1_i0" --force-keyframes-at-cuts --merge-output-format mp4 -o "IJvh0Ws1_i0-23m52s.mp4"

“If the AI model doesn’t talk the language of the pathologist you don’t have an explainable AI framework. Sure it’ll give you papers or maybe patent but it will not get a clinical approval.” — Akash Parvatikar, 24:08

”Digital Storage vs Physical Storage for 25 Years”

yt-dlp --download-sections "*22:00-22:50" "https://www.youtube.com/watch?v=IJvh0Ws1_i0" --force-keyframes-at-cuts --merge-output-format mp4 -o "IJvh0Ws1_i0-22m00s.mp4"

“Just imagine if there are like 12 million patients, 12 million glass slides across 25 years that has to be stored within the hospital setting. You would pay much much more for the real estate cost versus for a digital cloud storage.” — Akash Parvatikar, 22:03

”US Facing Large Shortage of Pathologists”

yt-dlp --download-sections "*16:09-17:00" "https://www.youtube.com/watch?v=IJvh0Ws1_i0" --force-keyframes-at-cuts --merge-output-format mp4 -o "IJvh0Ws1_i0-16m09s.mp4"

“US is facing a large shortage of pathologist and in the coming decade the gap will widen even more.” — Akash Parvatikar, 16:09

Key Points

• Digitization is the foundation (1:22) - Currently we are in the stage of digitizing pathology labs; clinicians still diagnose cancer under microscopes but there’s a large push toward digitizing glass slides
• Two parallel roads (2:03) - One road is digitization, the second is building AI tools to help assist clinicians; these will converge in the next couple of years
• AI as diagnostic assistant (2:58) - AI is being built as a diagnostic assistant tool to speed up the process and detect features that are highly missed
• Hospital adoption evolution (3:52) - Major hospitals like Mayo Clinic, Mount Sinai, NYU, and MSKCC evolved from using AI for high-level features to now analyzing low-level features like cell detection and spatial heterogeneity
• Massive image sizes (5:20) - One whole slide image is anywhere between a few hundred megabytes to a few gigabytes; printed at 40x resolution, it would occupy 1-2 tennis courts
• Patch-based processing (6:20) - AI scientists use tile/patch-based processing where they chop the image into multiple patches, run AI on each, then group results back together
• GenAI adoption is low (7:30) - Generative AI adoption for diagnosis is too low because AI cannot tell its failure modes or when it will get false positives/negatives
• Agentic workflows for quality control (9:05) - Agentic AI can be used in chunks of the workflow like quality control to ensure digitized images are high quality without pen marks or blurry regions
• Biggest bottleneck: annotated data (10:05) - The lack of high-quality annotated datasets is the biggest bottleneck; patients must consent, and there’s high disagreement among experts on annotations
• Regional diagnosis variation (11:10) - A doctor from NYU may give a particular diagnosis but if you take the same sample to Stanford, West Coast, Europe, or India, it could be diagnosed entirely differently
• PathologyMap platform (12:13) - An AI orchestration platform that allows users to run AI apps from multiple companies on the same slide simultaneously to compare performance
• Very few FDA approvals (14:15) - There are very few AI tools with FDA approval, and approvals are for very specific tasks, not general diagnosis
• Digital pathology is new field (15:05) - Digital pathology only came into being in 2017 when US FDA approved digital scanners for primary diagnosis
• COVID accelerated adoption (15:23) - COVID gave the field the right push because doctors didn’t want to come to hospitals for diagnosis
• Pathologist shortage (16:09) - US is facing a large shortage of pathologists and in the coming decade the gap will widen even more
• Spotify for precision medicine (19:42) - The goal is to create something like Spotify for precision medicine; Tempus is working on multimodal data integration
• Human role unchanged (20:20) - The human role will remain the same - like Tesla where you’re still in the driver’s seat but have tools to avoid accidents
• Economics favor digital (22:00) - US mandate requires holding physical slides for 25 years; digital storage is much more cost-effective than real estate for physical storage
• AI must speak pathologist’s language (24:30) - AI models that don’t talk the language of pathologists won’t get clinical approval; heat maps alone are not explainable AI

Mentions

Companies

• HistoWiz (0:41) - Where Akash Parvatikar leads the PathologyMap platform
• Mayo Clinic (3:52) - Has digitized 12 million glass slides; early adopter of AI in pathology
• Mount Sinai (3:55) - Hospital system using AI on digital pathology images
• NYU (3:57) - Digitizing all of its slides; has its own training standards for pathologists
• MSKCC (Memorial Sloan Kettering) (3:58) - Hospital using AI for pathology analysis
• Stanford (11:55) - Referenced as potentially having different diagnostic standards than NYU
• Tempus (19:52) - Building multimodal datasets and intelligence tools for precision medicine
• Google (18:51) - Referenced for Google Books digitization initiative as parallel to pathology digitization
• Iron Mountain (22:18) - Storage company mentioned for holding digitized data
• Tesla (20:32) - Used as analogy for human-AI relationship in diagnostics

Products & Technologies

• PathologyMap (0:41) - HistoWiz’s image management and AI orchestration platform
• Whole Slide Images (WSI) (5:20) - Digital pathology images that can be hundreds of megabytes to gigabytes
• GradCAM (24:32) - Post-hoc interpretability technique criticized as insufficient for clinical adoption
• LLMs (3:15) - Large language models being explored for moonshot diagnostic projects
• Python/MATLAB/R (13:08) - Programming tools users can use to generate annotations for upload to PathologyMap
• WHO Classification of Tumors (24:58) - The “blue book” for pathology with guidelines for diagnosing tissues

People

• Akash Parvatikar (0:31) - AI scientist and product leader at HistoWiz, leads PathologyMap platform
• Ksenia Se (0:24) - Host of the Inference show by Turing Post

Surprising Quotes

“If you take a one image print out at a 40x resolution it occupies anywhere between 1 to two tennis courts.” — 5:35

“Your ground truth data might itself be imperfect wherein there might not be a gold standard ground truth for that particular image. There could be a doctor from NYU giving it a particular diagnosis. But if you take the same sample to west coast, they might diagnose it differently. If you take it to Europe or India, it could be entirely different.” — 10:45

“If the AI model doesn’t talk the language of the pathologist you don’t have an explainable AI framework. Sure it’ll give you papers or maybe patent but it will not get a clinical approval.” — 24:08

“US is facing a large shortage of pathologist and in the coming decade the gap will widen even more.” — 16:09

“Just imagine if there are like 12 million patients, 12 million glass slides across 25 years that has to be stored within the hospital setting. You would pay much much more for the real estate cost versus for a digital cloud storage.” — 22:03

Transcript

0:00 When will AI meaningfully change the trajectory of cancer diagnostics? There are very few AI tools in the space which has got FDA approval. How do we standardize the quality and the diagnosis of the data? Do you think we will have Spotify for precision medicine?

0:24 Hello everyone and welcome to the inference show by Turing Post and today I’m speaking with Akash Parvatikar an AI scientist and product leader working at the frontier of digital pathology and computational biology. He leads the pathology map platform at HistoWiz focusing turning complex histopathology data into insights that accelerate cancer diagnostics and precision medicine. This is a very exciting episode for me because we decided to air it and dedicate it to breast cancer awareness month. So, thank you for joining me and let’s talk about this very important topic.

1:02 Amazing. First of all, thank you so much for inviting me to your podcast and I’m super excited to be here and just discuss some of the pressing things and what’s happening in the field of cancer diagnosis. Why is breast cancer so highly misdiagnosed especially in its early stages? And how is AI sort of helping bridge the gap?

1:22 That’s actually one of my first questions. Let’s start with the big question, big picture. When will AI meaningfully change the trajectory of cancer diagnostics? What I believe is that like right now we are in the stage of digitizing the pathology labs and as you know even now the clinicians diagnose cancer under a microscope but now there’s a large push towards digitizing those large volumes of glass slides and that is what will sort of open the doors for bringing in intelligent tools like AI to better detect the cancer features. So right now we are at the transition phase of going digital from physical but then I think we are running on two parallel roads where one road is towards digitization and the second road building AI tools that will help assist the clinicians in a positive way to better diagnose. So what I believe is that very soon maybe in the next couple years it will converge to a point wherein you have all your glass slides digitized and you have your cutting edge AI diagnosing features which might have taken a lot of time or which might have been misdiagnosed early on.

2:29 If you think in terms of patient outcomes what is early detection better precision what breakthroughs are real right now and what’s still aspirational? Right so I think right now where the field is to get the adoption from the clinicians who are the domain experts for cancer diagnosis. So the AI is being built for the existing literature that doctors have been using for quite some time and there are some features which takes a lot of time. There are some features which are highly missed. So right now AI is being built as a diagnostic assistant tool to speed up the process and also detect the features which they do every single day on a routine basis. But then we are also seeing some moonshot projects wherein there are some AI based on LLMs and deep learning wherein it’s being built to look at the features which previously the doctors might not have paid attention to but it’s a very important prognostic marker. So but those are all long moonshot projects but right now to get the quick adoption let’s keep it simple and try to build AI models that the doctors can use in their day-to-day practice.

3:38 And when you say that the doctor is using it for quite some time, how long they have been using AI that you observed and what’s changed from the initial “what’s this” and how they think about it now?

3:52 Right there are hospital systems in the US like the Mayo Clinic Mount Sinai NYU MSKCC wherein early on they were using AI on the digital images to look at high level features like tumor necrosis stroma and all that because of the maturity of the AI algorithms in the space. They’re now using it for low-level features. It could be cell detections, types of cells and not just that the spatial heterogeneity among these features. So they are trying to ask much more complicated features, complicated questions even beyond feature analysis like can the AI sort of map out the understanding the spatial tumor heterogeneity within the tissue sample and what prognostic markers can I get from doing this. So I think we have taken few steps ahead to what they were using it previously and right now but as I said the features that some of the hospitals have been using in the past we want to push it to more hospitals at a nominal cost in the US and outside.

4:50 You are an AI scientist so you know the machinery of AI pathology how it works. So AI pathology isn’t one model predicting cancer or not. It’s slides and pictures and research the volume of diagnosis multimodal inputs. So what is the hardest about working with whole slide images? Can you really go deep into the machinery how AI works there specifically?

5:20 Yeah, absolutely. So I think that’s a very important question that you asked. So when we are dealing with wholesale image one image is anywhere between few hundred megabytes to few gigabytes as I’ve said in the past as well that is if you take a one image print out at a 40x resolution it occupies anywhere between 1 to two tennis court. So that is the volume of image that you’re looking at which is very different from a cat versus dog or scene detection which is much at a much lower resolution and the doctors typically detect at a 40x resolution. So the magnification is quite high. So again this comes with the problem that we’re trying to find a needle in a haystack problem from a computer vision perspective but also when you talk about memory constraints and all that now we want to load this entire multi-gigabyte image or multi-hundred megabytes of image onto a server and run AI at scale. So there are a lot of parallel processing that has to go in place to run your AI model.

6:20 But typically in the machinery what we do is we do tile patch-based processing where we chop the entire image into multiple patches. We run the AI on the entire patch images and then we group it back. So again this is a very standard routine that we AI scientists follow and differently I think in the field of pathology and I think outside the field as well it’s very important to understand the data. Yes, I might be a great computer vision scientist and I might have built amazing models for other types of data. But when it comes to pathology, it’s very important to understand the features that you want to look at and how does it differentiate from one category to the next. Some sort of a literature review or sitting with the pathologist is crucial because at the end of the day you’re building the AI for the pathologist and to some extent it should be done by the pathologist too not in terms of coding but you should have gotten significant expertise and communication from early on when you’re building the AI and I’m more than happy to dive into detail if you want me to.

7:22 Yeah, that’s interesting because where you explain it’s more on the computer vision side but then there is also generative AI now and how that plays out.

7:30 Right I think again the adoption of generative AI for diagnosis or diagnostic decision assistance might be too low right now because we are in the stage wherein the AI is not able to tell its failure modes like it doesn’t know when will I get a false positive or a false negative especially when you’re dealing with these large LLM models and all that. Now if you take the few shot learning approach where you train it on a smaller data set well curated captures the diversity of the sample there is a possibility that once you build an AI on that it can actually tell you where the AI does good where it does bad the kind of data the nature of data and all that and that was the core of my research during my PhD where I built an explainable AI framework on a subset of data but it could actually tell not just where it performs well but where it fails. I think again it’s taking baby steps towards adoption. You cannot come up with a model which is gen AI based and try to entirely replace the pathologist and they are not going to buy it. Neither will the patient have an acceptance to be diagnosed by such an AI agent as you would call but it has to occur linearly at least for now. So we are in that linear phase of growth.

8:45 You said the word agent. Would you say that agentic workflows are implemented or have been implementing in the workflow of practitioners or it’s still too far for that?

8:57 It’s still too far for pathology. But I think there are multiple pieces within the pathology workflow. Like one thing is you do the biopsy. You send the slide for processing. You digitize the slide. Next is you have a workflow for quality control. You want to make sure that the image that has been digitized is of highest quality without any pen marks, blurry regions and all that. Now you can incorporate an agentic workflow to make sure the quality is standardized across the entire cohort and is adaptable or amenable for the AI that has to come at the next stage. So I think you can take parts of the workflow and create agentic AI models for that. And I think we are talking workflows for standardizing the quality of pathology data. But I think we have to envision this as more like chunks and not the entire piece like for diagnosis. So that’s way too far out.

9:50 So from what I heard what the bottlenecks are explainability and let’s call it hallucinations or false replies. Is that all what are other bottlenecks that you see and when will we deal with them? So I think the biggest bottleneck that I see even during my PhD and research is the lack of availability of high quality annotated data set especially in the field of pathology. Why? One I think the patient has to be consented to make it available for research and we’re talking about pathology images could be breast colon liver any organ but not just that there is a high degree of disagreement among the domain experts while creating those annotations. Even in my PhD when we got the annotations for the breast sample for the pre-invasive we took the consensus diagnosis or majority voting among the three expert pathologists who are generating the diagnosis or annotation.

10:45 So you can imagine that your ground truth data might itself be imperfect wherein there might not be a gold standard ground truth for that particular image. There could be a doctor from NYU giving it a particular diagnosis. But if you take the same sample to west coast, they might diagnose it differently. If you take the same sample to Europe or India, it could be entirely different. So I think that is the biggest bottleneck and that is what the field is struggling right now like how do we standardize the quality and the diagnosis of the data and not just that the biases on the data itself. So it’s heavily dependent on the hospital system whom you collaborated with in training your AI model. But once you train it, how adaptable is it to patients outside US or outside your hospital or even outside the region. So that’s the biggest bottleneck.

11:35 Do you mean it can be that specific? Yeah. So the thing is if the AI model was trained from the data from say NYU right now NYU has its own training for the pathologist where they come together and they look at some features for diagnosis but it’s not gold standard for all over the world. If you take the same image to say Stanford they might diagnose it differently. They might give some other category. Now your AI might completely fail because their ground truth might be entirely different for that particular sample. And we are talking about pre-invasive that is where the maximum confusion lies. So I think that’s the biggest problem in the field of pathology right now.

12:13 That’s very tricky. So you building pathology map is that something that you address with this platform? So the platform that we’re building right now at HistoWiz is an image management platform wherein it will allow you to manage access and analyze your high-quality images. But in addition to that we have also created an AI orchestration platform that allows you to ingest AI apps from multiple companies and the user can test for example they can run say breast analysis from four different companies on the same slide at the same time that can tell you how the AI is performing and the intricacies and all that. We want to give the decision making to the hands of the user in what AI tools they want to run. So that’s why we didn’t make it exclusive to one or two AI apps but opened it up and made it non-exclusive open so that we can ingest multiple AI apps or even upload your own annotations that you might have generated using Python, MATLAB or R onto the platform and we are one of the first ones in the world to do that.

13:12 Is it more for medical specialists or for everyone? It’s mostly for preclinical and clinical pathology research. By pre-clinical I mean it’s for the drug discovery companies for clinical trials where they want to understand how the drug works on human data on mice data and they can work on that or clinical data wherein they might have a patient 5 years ago ran a breast biopsy they want to upload it to a digital platform because accessing the physical slides might be cumbersome and might take a lot of time now they have a digital platform wherein it’s easy to access easy to share and they can also compare their clinical notes so that’s what we’re trying to do with our orchestration platform called pathology map.

13:51 What about regulations? How is that treating you because it’s such a sensitive topic? Absolutely. So at HistoWiz we are GLP compliant. So we are in pre-clinical research. So we don’t run the AI for diagnostic patients but mostly for historical patients or animal models. So for us the regulations are slightly relaxed but for the field itself I think there are very few AI tools in the space which has got FDA approval. But if you look at it and I’m more than happy to attach few resources to the comment section or description the FDA approval has been obtained for a very specific task. It’s not the AI for say prostate diagnosis but observing high important regions within the image. It’s very specific and still the industry is struggling to get FDA approval for the apps that actually makes the most sense in the diagnostic setting.

14:40 Let’s go back to what we started with about cancer awareness month, breast cancer awareness month because I want to ground this technical side in the human side. Have you seen moments where AI has already made difference for patient or pathologist? Well, I think yes to an extent but all of it is in pilot phase right and I think to give you a sense the field that I’m in which is digital pathology it came into being only in 2017 when US FDA approved using digital scanners for primary diagnosis so it’s a very very new field before that radiology has all been digitized and they have been using AI tools for that but pathology it’s at its inception stage and COVID actually sort of made the field go really fast it gave it the right push because now the doctors didn’t want to come to hospital for diagnosis. So more and more hospitals are digitizing.

15:35 The field has evolved in some in terms of again digitizing but when it comes to diagnosis I think real world cases we are still little far from that but still it’s all in the pilot phase it’s not in a real world setting because again you want to be so sure that the AI is doing the right thing and if you deployed it into a hospital like you as a patient will you go to a hospital where there’s AI in place or will you go to a hospital where there is no AI so there’s lot of questions to be asked even for the hospital to bring an AI product into market. But having said that, AI is making strides in the workflow optimization. As you might already know, US is facing a large shortage of pathologist and in the coming decade the gap will widen even more. So AI is definitely helping the current pathologist to improvise their workflow. It could be quality, it could be these features which takes hours together for them to diagnose like few low-level features like mitosis, lymphocyte counts and all that. AI is doing a very good job to identify these features and quantify that. So they’re using that parts of their existing knowledge to come with the diagnosis together. So in that way yes we are seeing a good change.

16:40 We should have probably started with that but still can you give me a one sentence description what digital pathology is so everyone just understands what it is exactly.

16:52 Yeah absolutely. So digital pathology is all about digitizing your pathology glass slides which was always looked upon like something that you will see under a microscope for primary diagnosis. Now digital pathology utilizes these high throughput image scanners that scans your physical glass slides puts it on a big screen very high resolution and it allows integrating your image management tools softwares and all that. You can compare it to the radiology which has been digitized like a decade ago and similarly the pathology labs right now the routine workflow is that you do a biopsy from the patient you put it on a physical glass slide you mount it send it to the histology processing and then it goes to a pathologist office and they have a microscope in their office they look at it under a microscope do the diagnosis but just imagine having one step forward wherein you have a scanner or a camera that sort of digitizes the physical glass slides and now the doctors are able to look at those high resolution images and only you can imagine the powers of digitization like one thing is it’s very easy to access second thing is very easy to share for example the patient could be in rural parts of US or rural parts of India but still an expert pathologist sitting in New York or California can diagnose that patient there are very large advantages to going digital. I’m just highlighting few of them. But these are the reasons why it’s becoming so popular right now.

18:21 Can you give me your timeline when we will see the main milestones, breakthroughs?

18:29 I think to take a step back, Mayo Clinic has digitized 12 million glass slides. NYU is digitizing all of its slides. That’s the first step, right? So again to draw parallels, you can think of it as before Spotify existed, we all had cassettes and DVDs to watch movies, to listen to songs and all that. But now who uses them? It’s all digital, right? Even with physical books like Google when they launched the Google books, they actually pared out the entire book library and made it all digital. So there was a huge initiative to make it all digital and then they built tools on top of that like there was Kindle and all that. So right now we’re in the stage of digitizing the pathology slides. I would say in the next 2 to 3 years we will be seeing more and more FDA approvals for diagnosis or maybe not just that diagnostic assistant where there will still be a human expert who is the pathologist they will be using all these tools just so that they don’t have any bottlenecks or roadblocks which they might have had previously but I think in the next 2 to 3 years we will definitely see either more cases being diagnosed by the pathologist more access to high-quality pathologist all over the world through this and not just that within a single case the accuracies will also increase because now they all have all these tools to avoid accidents.

19:42 Do you think we will have Spotify for precision medicine? Absolutely. That’s the goal. Yeah. Do you know of anyone building it? No. But there are some interesting initiatives by Tempus AI. You might have heard of them. There’s Tempus who are building multimodal data sets and intelligence tools wherein they are taking a pathology report a radiology report blood samples and genomic analysis all together and how well can you drive the precision medicine by profiling the entire patient across multimodal data. So Tempus is doing a fantastic job in that space.

20:20 Well, if we extend your timeline a little bit and think about some ideal situation, how you see it, what will be the human role in this workflow with AI? Human role will be the same that we have seen right now. Like for example, I think of this as your Tesla. If you would ask where would be the human role when you designed the Tesla, it will still be the same. You’re still in the driver’s seat. We are still very far from going away from the driver’s seat and letting the car drive by itself. What I think is that it will still exactly be the same where it was previously except that you will have tools that will avoid accidents. It will give you computer vision tools or AI tools that will help you detect features which you might have missed and it might help you access previous reports from tens of years you might run the AI on the entire 10 year repository of slides and you want to compare clinical notes. Okay, there was this patient 7 years back. I diagnosed this patient to be this. Can I quickly take a look at the report? If you want to go physical, it will take a hell lot of time and it’s unimaginable to go into the physical building, look at the glass slide, access the report and all that. So I think we will definitely see a very large improvement in how you collate the data and use it for better diagnosis but still the human part will still be the same but with much more tools at hand.

21:36 Do you know the economics behind it? Is it supported by the numbers the development of this area? Yeah, that’s a fantastic question. So I think we’re all talking about the ROIs and patient reimbursements and all that. So there is an active conversation in this sense. But I think in the long term there is a mandate in US that you have to hold the physical glass slides for I think close to 25 years. So just imagine if there are 12 million patients, 12 million glass slides across 25 years that has to be stored within the hospital setting. You would pay much much more for the real estate cost versus for a digital cloud storage and you can still hold it in places like Iron Mountain and all that where they can hold this. So I think storage in the long run is much more optimal when you go digital. That’s one advantage and it’s also the case signs. So say a pathologist might be signing out maybe 30 cases with physical digital they might be doing 50 and that will sort of compensate the shortage of pathologist and more number of patients who are coming to get diagnosed can get it done. So but still I think this is a continuous conversations with the healthcare organizations in US as well.

22:48 Yeah that makes sense. Let me ask you what are you concerned about and what excites you the most about this world that you help build with AI? For pathology I think I can answer the second question like what excites me the most is again for me it’s telepathology because for me there might be patients who might not be part of the entire cities where the core specialization or specialty doctors might exist this field allows them to get good care by digitizing the slide remotely sending it to the expert pathologist I think there’s this communication platform that happens even for second opinions, right? So there might be a doctor in New York, but they have a friend in California. They want to quickly get a second opinion. With physical, it might takes a lot of time, but with digital, it can happen really quickly within minutes. So I think that excites me the most wherein you’re trying to establish a communication platform for the pathology data and the experts through the domain. So that’s very exciting.

23:52 And what scares me is the AI developers who don’t have domain knowledge and are building AI models for the space and trying to go for the post hoc interpretabilities like the grad cams and hotspots and all that. It’s truly not explainable if the AI model doesn’t talk the language of the pathologist you don’t have an explainable AI framework. Sure it’ll give you papers or maybe patent but it will not get clinical approval. So that scares me because I have seen several models where they just show some heat maps and there is gradcam and all that and call it as an explainable AI model but it’s not unless you can tell me what features are being detected and are those features actually pathologist friendly. You can’t develop an AI model and give some deep learning features and when you take it to the doctor they have no idea what this feature is. It has to start small. You have to talk the language of the expert.

24:43 That’s a very important point. Yeah. Thank you. My last question will be about books. Is there a book that shaped you your approach to AI medicine or life? And it can be a book that you read when you were a kid or a recent book.

24:55 Since we are talking about the pathology and being more technical here. One book is the blue book for pathology which is the World Health Organization classification of tumors. It’s a very understated book. It’s called the WHO classification of tumors and they have one for every organ and there are multiple volumes for every organ. So for breast there are I think close to 12 to 15 volumes wherein in each volume they tell in detail as to these are the guidelines that has to be followed in diagnosing the breast tissue. So I followed that throughout my PhD and especially if anybody wants to get into a pathology field. It could be any organ. I think go and read that and try to understand the data really well. That would be one book that has stuck with me in my early PhD days and I’m sure it will be useful for others as well.

25:46 Thank you so much Akash. That was deeply technical and also deeply human because you talk a lot about how actual doctors and pathologists work with the technology. Thank you so much.