Fabien Beckers on the future of AI in medical imaging
Today’s radiologists face a deluge of data, and their work can be tedious and error-prone. But should humans even act as radiologists? It’s becoming clear that computers and humans working together are better than either alone. Harry’s guest this week is Fabien Beckers, CEO of Arterys, a startup creating products at the intersection of the cloud and AI in medical imaging.
Please rate and review The Harry Glorikian Show on Apple Podcasts! Here’s how to do that from an iPhone, iPad, or iPod touch:
1. Open the Podcasts app on your iPhone, iPad, or Mac.
2. Navigate to The Harry Glorikian Show podcast. You can find it by searching for it or selecting it from your library. Just note that you’ll have to go to the series page which shows all the episodes, not just the page for a single episode.
3. Scroll down to find the subhead titled “Ratings & Reviews.”
4. Under one of the highlighted reviews, select “Write a Review.”
5. Next, select a star rating at the top — you have the option of choosing between one and five stars.
6. Using the text box at the top, write a title for your review. Then, in the lower text box, write your review. Your review can be up to 300 words long.
7. Once you’ve finished, select “Send” or “Save” in the top-right corner.
8. If you’ve never left a podcast review before, enter a nickname. Your nickname will be displayed next to any reviews you leave from here on out.
9. After selecting a nickname, tap OK. Your review may not be immediately visible.
That’s it! Thanks so much.
Harry Glorikian: Welcome to the Money ball medicine podcast. I’m your host Harry Glorikian This series is all about the data-driven transformation of healthcare and life sciences landscape. Each episode we dive deep through one-on-one interviews with leaders in the new cost-conscious, value-based healthcare economy. We look at the challenges and opportunities they’re facing, and their predictions for the years to come.
The tech and radiology communities expect artificial intelligence to transform medical imaging, providing better services at lower costs. For example, if you’re getting an MRI an AI program can improve the analysis leading to better treatment. Today’s radiologists face a deluge of data as they serve patients, the work can be tedious making it prone to error. The remarkable power of today’s computers has raised the question of whether humans should even act as radiologists.
The reality is that the two working together are much better than one or the other alone. Imagine what would normally take 45 minutes to do by a human alone can be done in a few seconds when applying these technologies. My next guest Fabien Becker’s is the CEO of Arterys. A medical imaging start-up that is creating products at the intersection of cloud and AI in medical imaging. They are one of the companies at the forefront bringing a web-based medical AI imaging analytics platform powered by AI to the market.
Fabien has led the growth of the company from 4 co-founders to a team of 100 today. The company has become a pioneer in cloud-based medical imaging software, offering the first FDA-cleared, end-to-end cloud infrastructure for medical imaging. Fabian’s vision for the company is to accelerate data-driven medicine by building precision medicine tools, based on the consistent quantification of medical image features in combination with molecular, genomic and patient history data. Fabien holds a PhD in quantum physics from the University of Cambridge, and a Master of Business from Stanford University. Fabien, welcome to the show.
Fabien Beckers: Hi, very happy to be here? Thank you for giving some time today.
Harry Glorikian: Fabien tell me about Arterys.
Fabien Beckers: Yeah, Arterys so we created a company six years ago, now it’s working since then and the vision and the mission that we have is really to try to spend some healthcare, and to make it truly data-driven and we really realized today that the main issue clinical diagnosis as a whole and especially AI in medical imaging comes from the fact that I think today’s infrastructure is really not what you call in a pre-internet mode. You have all these computers lying around try to make sure they’re connected and therefore thin positions don’t have all the power and all the right tools. That they need to really be enabled as we think they should be and could be.
And so, when we go out the rest is instead of building more workstations and more computers that we deploy on-site. We are trying to bring the image into the web, and we think we can bring it to power the Internet’s really helped position, and so instead of being a local machine. It just uses a browser you just use your browser to get images. The data into the cloud while preserving data privacy and then therefore you have in the background an enormous and limited amount of computation to do the work that you can do.
But then now in second rather than minutes or hours and so yeah they basically prepare the case for the physician and the who can override and change anything they want. So but it still saves enormous amount of time up to seventy percent in some cases, and then the report goes back with an illegal workflow to whatever reporting structure or whatever dated is long. And it can be shared with anyone they want also because it’s a web-based platform. So by doing that we think we’re going to equalize and democrat size quotes of care around the world, and enable relates you have better accuracy and better productivity a lower cost for healthcare and AI in medical imaging.
Harry Glorikian: So you started in cardiac and now I’ve you know you’re moving to other areas within oncology. But what are the ultimate goals of the company.
Fabien Beckers: And you know my goal of the company is really to think about so we saw this car guy. Because we felt first always a really important legal so our terms here doubtless is really about patient and helping physician, and when you take that extremely seriously and so we realized early on that project is probably one they were first of all this extremely strong need on helping new-born and paediatric patient that are heart defects, where today diagnosis were made using echo and sometimes we didn’t get it right quantification for the physician to make the right diagnosis.
And we thought that was a very important legal need and either a problem we need to work on, and so after that we went and enabled and meant with the products for cardiac acoustical case it can tackle the hardest problem in medical imaging we play build a platform they okay be address for everything else after that and that’s what we built. And now we really truly believe that with our conversation with position and they’re in the couple of hundred fish. We have offices today that really did there’s a potential for AI in medical imaging and healthcare team moves to the web and to the cloud like any other industries.
After is hope to play the last part in really helping the transition to happen while respecting all the needed health system has and all the constrain. They have from the medical device or regulatory standpoint.
Harry Glorikian: As a cloud-based system, how have you designed or are imagining the business model from a use case perspective, and how do you think about charging for a software product like this.
Fabien Beckers: Yes, that’s a great question, so at the moment the way we’re as any again we take you learning from other industries here, and is right here that sass and account base are being adopted by other industry successfully and we’ll be using sass a subscription. I think in healthcare you have a slight variation because in some ways they are sometimes large multi-year budgets, and we can do that and be more very color Catholics expenditures. But from other hospitals that tend to have more not OPEX. We also flexible to enable that as well, and one of the value of a web-based system is not just doing the same at their ease to do before better you also the new possible that it didn’t have in the past.
For example, like today if you have a local system to upgrade every six months every year. Now we upgrade with two weeks he was very hard to connect and share and interact with other physician, outside AI for medical imaging or outside institution. Now with the web-based platform it’s easy to do so and you have computer all the data it can be leveraged as well so he opens new avenues that could not be possible that are really critical for a health system, or for helping patient and doctors. So that’s why I think it’s a subscription fits potentially better. It’s a lower burden on the hospitals it can tie that to their P&L. While we’re using a cost increase of artillery and get a better accuracy across.
Harry Glorikian: So it’s interesting right because you know many conversations that I’ve had with hospital IT and so forth like… Cloud-based systems you know always to a certain degree make everybody very uncomfortable. So how did you how did you sort of get over that issue or bridge that gap when you were introducing this.
Fabien Beckers: And you’re very right here so if you really look at doing why if you understand structurally what has happened in healthcare. You will understand you’ll so you may ask a question like why Hospital at industry not to have really fully adapt to the cloud and the web. And I think you really like put your finger exactly where is all about patient privacy and so, when we started five six years ago when we arrived with really great solution that could only be done sunbathing very cost prohibitive fashion on the local basis. Also we’re very excited about the solution but say we can’t move to cloud.
Because we don’t want to send patient health information or PHI to the cloud and so, from the echo we realized that you really enable and bring the value to the clod through a health system. You have to solve the data privacy mine and so we spent four and a half almost five years now working on their unique at the data privacy solution. That allows hospital to send us images but it doesn’t have any PHI on it with strip out. All the PHI but in a browser product addition the office where we consult for the physician in that session. The pixel data and the pH eye at a time of the read and so what he means is that by doing that decision will have access to the PHI.
But without no rich of taking the PHI are outside of hospitals, and by doing that we really bring the value of the cloud in the computation you have on a true cloud native real web-based platform. To health care while making them reassured and secure on their private patient data.
Harry Glorikian: And so is this a real time interaction that’s happening or is it an upload analysis download.
Fabien Beckers: No, no it’s real time. So it’s like think about analogy with other industry like think about Netflix. We believe clicking car Netflix of medical imaging. Of course the patient data comes inside a browser for the physician. Whatever he or she looks at the images that the psi example in hospitals was famous we were inside the wall of the hospital, and people data come from the cloud. So like that we breaking the two problem into one working solution that interact integrate and is interoperable to their work so and health IT system.
Harry Glorikian: So one of the big you know milestones for you guys. I you know because I remember reading it was the FDA approval for the product and now there’s it’s it looks like there’s been multiple FDA approvals. I’m sure there’s a whole series that you have in the pipeline with them. Now that you had your first one cleared, but can you talk a little bit about that FDA approval journey you know the implications for the for the company, and your experience with the agency.
Fabien Beckers: For sure and I think when it’s really important I think some people really tackle AI in medical imaging. Recently I would say last two three years in deepening came nation as a tech problem, and I’d offers we really started disagree with that vision we really think that it’s profoundly a medical device solution. Happens to have software and it happens you have some form of advanced analytics.
That can be put to bear, but at the core medical device that is supposed to help decisions interpretation for very critical conditions. And so therefore if you apply to the process of a true medical device’ or the process. Then you basically can test appearance and therefore you can get adopted by hospital. But I think the real landscape is the following one.
Harry Glorikian: So, how did you know what was your experience with the agency. Because this is very new to them as it is to all of us in it you know in the industry did you know, I want to say that you guys are first so you’re you’re teaching them while they’re sort of teaching you in a sense and it’s a back and forth you know.
Fabien Beckers: I do think that we’re trying to apply framework that builds trust, on how we approach a dough’s problem and try to yet to bring a solution that respect their need and it comes from. While showing that a path to bring safety and efficacy. so for example we go framework on how to test and by state we put some, we follow all the guidelines regarding testing and every single validation process that we have internally. So I do think now more and more interest around the FDA on how we can tackle the facts of IVs framework on how can we apply AI throughout the body. Rather than in the past I mean just one organ and one function, and so for example our encore appearance is applied to the whole body.
As long as of course we detailed in our processes after that how to test anybody they eat our garden, and so I think that’s really ways to really work with the APN soap operas has been a treat a very good experience, on how to encrypt take a solution. That is fully fun life and ready to get an ankle grade and be put to market once it comes to the right part.
Harry Glorikian: Does your system and I you know I don’t know every single module, but does your system learn when it’s interacting with the physician. When if they decide that they want to manipulate something or change something or adjust something accordingly.
Fabien Beckers: Yeah so I wouldn’t say that as you getting back to the point in this main hurry it’s like I think we under the FDA, that self-learning will add acid. You want users start doing some wrong or wrong correction or wrong suggestion that so will suddenly be skewed to that position. We have a QA process we collect all the changes of course as a web-based system. Sort of valuable health systems make sure that we collect for multiple end point. Therefore, the algorithm will always get better and smarter, but we don’t before the guys guidelines so in that regard we don’t sell word and automatic.
But we definitely do improvement every time we release past the QA process. Therefore, we comply them and then we can leverage those values and bring back to the network.
Harry Glorikian: So basically it’s at a different time point there’s what I would almost call a software update but it’s happening centrally as opposed to in a distributed manner.
Fabien Beckers: All right exactly so you can think about it like that also whenever we through the day. I saw some institution that will basically make some suggestion of change to make sure they can delete can’t go report and we collect all of that on your own basis, and then of course based on some institution or some rule that we have I won’t go into detail. Then we can basically collect that and make sure that we have more data and more inputs to really make the system be smarter and better than it was previously. And I mean we go to QA and then we release, and then like that every single target release. This is simply gets better and smaller all the time.
Harry Glorikian: I had seen a talk you gave or and it was referred that you were part of the World Economic Forum steering committee on value-based care. Are you still part of that?
Fabien Beckers: I am and no one I have to admit that with my struggle in the travel not as involved as I was at the beginning. But yeah was at a time I didn’t talk last I think was a few years ago -.
Harry Glorikian: Yeah, so I’m curious you know you run a data-driven medical company and you’ve been exposed now to that group. And so, how do you see what are the implications from data-driven medicine on value based medicine? And I’m asking this for self-serving purposes, I wrote a book on this so I’m curious to what your positioning is or what your thoughts are.
Fabien Beckers: Yeah, I do think that the making piece between the two together is having the right tool and right information and the right reporting to help driving the best decision possible. And so in that regard is amole and is a very siloed system that we’re dealing with. It’s very hard to have the data to serpent in a way to enable the positional institution to give the right or back special treatment to the best to the patient’s. And so, in that regard I think web-based in the doctor is, what we can provide is one person a layer of accuracy and portability across the system.
And so the beauty of AI in medical imaging is that, it’s three vendors the data are standardized structured and they’re usually – very slightly different but still much more emoji news than you have in the text data [INAUDIBLE]. And so you can really truly make value on that first level and then on top of it well aggregating those across a lot pool of patient or hospital. And a non-profit you know start adding now all the data on top of it, a body a model by integrating to give more substance to the analysis. And by cheap adding those layers I believe will get closer and closer you have a very exhaustive, of course it’s never been fully complete but unsecure that we can be to a full picture. No, one intended to the physician about what’s happening to that patient what’s the best treatment to enable by the value based chair and to gave all the best proximal pass for that patient.
Harry Glorikian: So, if I’m not mistaken the company idea came out of your work from Stanford. And how did that you know, how did you get that off the ground? How did you start the companies the company itself?
Fabien Beckers: For sure, and I think he came out and was very personally actually, I had myself a health situation I had to deal with. And I was surprised, I came to the US to Stanford for doing some mission type work that I think impacting to population that what motivates and looks like me. And so when I had personal health issues myself and scholar house variable was the clinical diagnosis process, and how people had very little information data to drive the decision. I really saw that was very interesting situation and therefore got me investigating. And the more I would learn and more I got into it, the more I realized how structural you were notice of one-offs is a situation.
And then the further I went and I start thinking about potential solution to be more data-driven element to one in this area, and then by adding other position. We realize together that the problem is not about – software the problem is not a structural level, health systems today are paid on the local, on premise or station approach. And therefore are kind of like in shackle in some way or by these bolts is very hard for data to flow. So, to compute capability put to bear to the physician, [INAUDIBLE] and hold that situation. And so we felt that it will be worse try to build from scratch, the right in front of church what kind of health care 2.0 we are thinking.
They will embrace a web, that will embrace cloud computations that will embrace the AI, that will embrace like collaboration and sharing and brain data silos neutralize some resources on a global scale to make the system and suggestion work better together. Healthcare is a very interesting field because in other industries, will be hard for engineers but to different to work together, or to bankers to work together, in health care system you’re willing to work together. And so therefore where need the system to enable them, and so that’s why we think more than any other industry, a web-based platform for medical mission critical diagnosis is essential.
Either key and the right framework to bring knowledge which is of course it’s a peer to peer industry. So, know it extremely important healthcare but doesn’t work with a siloed system. It’s about really having multiple AI run on every tape, which is about I mean neither confused, you want even more confusing need to start leveraging the cloud because our leveraging more data and not just AI in medical imaging. I don’t know how much data I think change or how much data about healthcare the increasingly like month or two months. I think it’s in the dozen terabyte data, isn’t it?
So, therefore need AI to bring the data to bear to get the learning out of it and help digesting that other level of the chase of the patient. And so that’s what we felt was needed and that’s not to do with AI, that to do with infrastructure, you have to do with a web, you have to do with browser, they can process medical imaging as a real-time wherever they are, have to do with interoperability and workflow integration, have to do with data privacy. A lot of poor ceases that are not sexy, but I’m so creatively important in order to really enable transition of that industry that feel.
Harry Glorikian: Well, it’s interesting right, I mean I keep hearing your AI, AI, machine learning. I hear all these you know everybody talks about it, but at some point you know the user actually doesn’t care. Because the user just wants the answer. So, it’s almost as if it should be transparent to them in the background.
Fabien Beckers: But it’s not like if I tell you today, that they can get a lot of transistor for your phone like you don’t care, just on a phone, nice phone or whatever brand.
Harry Glorikian: Yes, you want to pick up and make a phone call.
Fabien Beckers: Yes, you want to make phone call, you don’t want like [INAUDIBLE] like Motorola, XYZ, the transistor you want to make a solution. And I can tell you that from now I know AI is above and it’s true that AI can really help drive for giving accurate anyone. Like deep running really truly can reproduce path that I don’t buy really accurately and very well, but the hard part is not the AI the hard part of the platform. It took us five, six years of the platform you can take us weeks or months to build an algorithm is not many more.
So, it’s really not the hard part, – working in hospital to work integration, the data privacy that’s are the real Himalayas climb, that the AI every single banker and trade on Wall Street and download test affluent and did download from NIH 4000 dataset and build a model and become a medical invention, AI can be. But what they’re going to do when they are PCI right there I went to get it to the physician at the end oh they seem quite patient and that’s mile between the AI and the patient and a position is caught that thing, what we believe, much harder problem to solve.
Harry Glorikian: So, Fabian what worse if you had to juxtapose. What were some of the challenges in the early days of the company and what are some of the you know business and technical challenges you face now right, you know year one versus year five or six, I’m guessing?
Fabien Beckers: I think the really interesting question. So, I think you’re one, it was really like okay so we had a vision on how to bring that data-driven medicine how to build that infrastructure, I think we’re far as there is to be the hard it is to build it. So, we thought that we can within I don’t know month or years we’ll be done it was for you ready, and we’re secure then and we feel like it still more to do and which every day we’ve done more.
So, I think that’s your one and so and then also it’s covering issues about, if you come from the tech world it seems so easy that yeah of course there’s a web solution, everyone is the web where which is the local system. But you have to understand all the complex cities and intricacies of building a web solution or healthcare environment for hospital. And so we had to learn a lot of those – not about something that can do overnight a lot of iteration work but you just that’s one bottleneck, and you have another bottle neck another, one by one in chief addressing to have over years.
And so I think that was the first few years, now we have a different problem about scale. So, now we can in health system that I’ve seen and I think the mentality has shifted to now CIO, CEO, CMO or radiologist see the value of whatever web-based got them to bring them as long as it’s big data privacy and is regulated. And so now we have straight what we’re talking to Health System to transition celebrity offers. But of course how many, I mean tackle how many can we get a connector to the platform.
Those are the different situations; we’re live now in China so we have community in the world to manage. But we are in nearly 95 or 100 countries where we regulated today. So, we have a question of scale which it’s very different now that a question of building from scratch, which was the case five, six years ago.
Harry Glorikian: So, you know we talked about the growth of the company and everything but you’re not just hiring anybody. You need a particular skill set of individuals and you’re in the Bay Area that is not a, those people are highly sought after individuals. So, how are you identifying the staff, how are you hiring and retaining the right people to do this?
Fabien Beckers: Yeah, I think [INAUDIBLE] whatever because part that we have of course the skilling team. And I think one of the, I would say the number one asset of their company today, it is people. Like we spend so much time trying – to hire the best team and we have, I think this is an amazing team. Now your right is hard to attract talent is your invaluable is very competitive, but I would say that a lot of the real top talent that we have -, are coming is because it all our passionate about want to change health care, any plaque patient lives. And I really think that connects us as a group and therefore you have a lot of amazing engineers, and if we work on the financial app or they can work on thinking up or whatever is possible to do core value.
But I think a lot of them find a sense of purpose and trying to make sure that you know they can help efficient that have lung cancer or collectively the liver cancer or someone else stroke I mean, imagine how many imagine the power that engineers that work for maybe like six, nine months. And if can have a million life after that where his worth each line of code with darkly in fact patient care and I think that’s makes a very big difference from other companies that are in the second dying for sure.
Harry Glorikian: Well, Fabian I don’t want to take any more of your time. You’ve out there really making a difference in the industry as a you know thought leader and Trailblazer in the area. Especially from a regulatory approval perspective, getting a product out there that’s useful. I want to thank you so much for your time and you know everything you’ve shared with us, and thank you for joining us on the show.
Fabien Beckers: Thank you Harry so much and it was a real pleasure, have a great day.
Harry Glorikian: Thank you, bye-bye. And that’s it for this episode. Join me for the next episode, where I speak to Sharon Terry the president and CEO of genetic alliance. Where we discuss how drug discovery diagnostics and the treatment of patients is changing. If you enjoyed Money ball medicine, please head over to iTunes to subscribe, rate and leave a review. It is greatly appreciated, hope you join us next time, until then farewell.
FAQS about artificial intelligence in medical imaging
What is the role of AI in medical imaging?
AI has a significant role in medical imaging, which involves the use of various imaging modalities to capture images of the human body for diagnosis, treatment planning, and monitoring of medical conditions. Medical imaging encompasses a wide range of technologies, including X-rays, computed tomography (CT), magnetic resonance imaging (MRI), ultrasound, and positron emission tomography (PET).
Here are some of the ways AI is being used in medical imaging:
- Image analysis: AI can analyze medical images to detect abnormalities or diagnose medical conditions. AI algorithms can help identify patterns in medical images that may be difficult for human doctors to spot. For example, AI can help detect tumors in X-rays, CT scans, or MRI images.
- Diagnosis: AI can assist doctors in diagnosing medical conditions by interpreting medical images. For instance, AI can help detect breast cancer or lung cancer, assist in the detection of pneumonia or COVID-19 on chest X-rays and CT scans.
- Treatment planning: AI can assist doctors in developing treatment plans for patients by analyzing medical images. For example, AI can help plan radiation therapy for cancer patients by identifying the exact location of the tumor.
- Workflow optimization: AI can also improve the workflow of medical imaging procedures by automating certain tasks, such as image segmentation or measurement. This can help reduce the time and effort required to interpret medical images, and increase the efficiency of the overall medical imaging process.
In summary, AI can help improve the accuracy, speed, and efficiency of medical imaging procedures, enabling doctors to make more accurate diagnoses and develop more effective treatment plans for their patients.
How does ai imaging work in medicine?
AI imaging works in medicine by using deep learning algorithms that are trained on large datasets of medical images to detect patterns and features in the images that are indicative of specific medical conditions or abnormalities. These algorithms can be applied to a wide range of medical imaging modalities, including X-rays, CT scans, MRI scans, and ultrasound.
Here is a general overview of how AI imaging works in medicine:
- Data collection: Medical images are collected from various imaging modalities and labeled with relevant clinical information, such as the patient’s age, sex, and medical history, as well as the presence or absence of specific medical conditions.
- Data preprocessing: The medical images are preprocessed to enhance their quality and remove any noise or artifacts that may interfere with the analysis. The images are then converted into a digital format that can be processed by AI algorithms.
- Training the AI algorithm: The AI algorithm is trained on a large dataset of labeled medical images using a deep learning approach called convolutional neural networks (CNNs). The CNNs learn to identify patterns and features in the images that are indicative of specific medical conditions or abnormalities.
- Validation: The trained AI algorithm is validated on a separate dataset of medical images to ensure that it can accurately detect the relevant features and make accurate diagnoses.
- Clinical application: Once the AI algorithm has been validated, it can be applied in a clinical setting to help doctors interpret medical images and diagnose medical conditions. The AI algorithm can analyze medical images and provide a diagnosis or recommend further testing, which can help improve patient outcomes and reduce the time and cost of medical imaging procedures.
In summary, AI imaging works by using deep learning algorithms to analyze medical images and detect patterns and features that are indicative of specific medical conditions or abnormalities. These algorithms can help improve the accuracy and efficiency of medical imaging procedures and assist doctors in making more accurate diagnoses and developing more effective treatment plans for their patients.