It seems we’ve been hearing a lot about artificial intelligence (AI) lately: It’s how Netflix knows what to recommend you watch next. From ChatGPT to Netflix, AI is everywhere — and it’s changing health care in all its aspects. The promise, the peril, and the prophecy of AI in health care are what we are going to examine in our next three podcast episodes. We’ll be talking with pediatric experts in one specific area — cardiology — to get their take on AI and how it’s changed and is changing the work they do with patients and families.
This episode was taped live during the 8th World Congress of Pediatric Cardiology and Cardiac Surgery in Washington, D.C.
Carol Vassar, producer and host
Featured Guests:
Rima Arnaout, MD, Assistant Professor in Medicine, University of California San Francisco
Devyani Chowdhury, MD, Cardiologist, Nemours Children’s Hospital, Delaware
Mark Friedberg, MD, Pediatric Cardiologist, Labatt Family Heart Centre, Division of Cardiology, Hospital for Sick Children Toronto, Canada
Babar Hasan, MD, MBBS, Professor of Cardiology Chair, Division of Cardio-thoracic Sciences
Sindh Institute of Urology and Transplantation (SIUT)
Charitha D. Reddy, MD, Clinical Assistant Professor, Pediatrics – Cardiology, Stanford Medicine Children’s Health
Shubhika Srivastava, MD, Co-Director, Nemours Children’s Cardiac Center, Delaware Valley, and Division Chief, Cardiology, Nemours Children’s Hospital, Delaware
Additional Episodes In Our AI Series:
Artificial Intelligence in Pediatric Healthcare: A Primer
Artificial Intelligence in Pediatric Healthcare: Dr. Mark Friedberg, Hospital for Sick Children, Toronto
Artificial Intelligence in Pediatric Healthcare: Dr. Charitha Reddy, Stanford Medicine
Artificial Intelligence in Pediatric Healthcare: Dr. Pei-ne Jone, the Ann & Robert H. Lurie Children’s Hospital of Chicago
Artificial Intelligence in Pediatric Healthcare with Dr. Babar Hasan, Aga Khan University, Pakistan
Artificial Intelligence in Pediatric Healthcare with Dr. Rima Arnaout, University of California San Francisco
Episode 35 Transcript
Carol Vassar, podcast host/producer:
Welcome to Well Beyond Medicine, the Nemours Children’s Health Podcast. Each week we’ll explore anything and everything related to the 80% of child health impacts that occur outside the doctor’s office. I’m your host, Carol Vassar. And now that you are here, let’s go.
MUSIC:
(Singing.)
Dr. Shubhika Srivastava, Nemours Children’s Health:
Artificial intelligence is actually a transformative technology, and it has a potential to revolutionize patient care, diagnosis, and management. And it also has a potential to change the course of healthcare utilization, healthcare access, and patient outcomes. It’s been a revolution in the making for a long time.
Carol Vassar, podcast host/producer:
That’s Dr. Shubhika Srivastava, Co-director of the Nemours Children’s Health Cardiac Center and Chief of Cardiology for Nemours Children’s Health, Delaware, with a definition of artificial intelligence and healthcare. It seems we’ve been hearing a lot about artificial intelligence lately. It’s how Netflix knows what to recommend you watch next. It’s what tells Amazon who’s buying what, where, how, at what price, and who to target to sell more widgets and gadgets. It’s chatGPT, and it’s changing healthcare in all its aspects. Some call it transformational, others say revolutionary. The promise, the peril, and the prophecy of AI and healthcare are what we are going to talk about in our next three podcast episodes. We’ll be speaking with pediatric experts in one specific area, cardiology, to get their take on AI: how it’s changed and is changing the work they do with patients and families.
The National Academy of Medicine defines AI as “The ability of computer algorithms to perform tasks that typically require human intelligence.” In the healthcare setting, that includes faster and more accurate medical diagnoses, better management of healthcare data, and drug development on the clinical side, not to mention more efficient and accurate resource allocation, claims processing, and supply chain management on the business side.
It’s also big business. According to the National Institutes of Health, an estimated $150 billion in healthcare costs can be cut annually by 2026 by providers and institutions using AI technologies. The marriage of tech and healthcare innovation is all around us, too, as evidenced by the $21.4 billion poured into AI healthcare apps, programs, tools, add-ons, and technologies since 2020, according to the healthcare venture capital group Rock Health Advisory, with continued growth likely for the foreseeable future. AI in healthcare is far from new. According to Nemours’ cardiologist, Dr. Devyani Chowdhury.
Dr. Devyani Chowdhury, Nemours Children’s Health:
AI has been around for a very long time. It’s not anything new, but I think we’ve advanced a lot in our technology, that AI is coming home to healthcare. Healthcare is usually the last industry to be impacted whenever anything new happens. So, I think AI is impacting us in day-to-day life, which all of us know. But in our day-to-day healthcare, I think it has a major impact. And specific to congenital heart disease, we don’t even realize actually AI is kicking in, and AI is actually happening, and we don’t even recognize that. For example, when we are reading an echocardiogram, a lot of the measurements are now on auto measurement. So, where does the auto measurement come from? That’s an AI-based tool that’s coming up.
Carol Vassar, podcast host/producer:
The echocardiogram, which uses sound waves to create pictures of the heart to check its structure and function, has become a first-line diagnostic tool for pediatric cardiologists, according to Dr. Charitha Reddy, Clinical Assistant Professor of Pediatric Cardiology at Stanford Medicine in Palo Alto, California.
Dr. Charitha Reddy, Stanford Medicine:
We use it as a treatment management system. We use it as a way to evaluate whether things have changed, things like that. So it’s part of our regular workflow all the time. And so you can imagine that when we do that, we’re taking pictures of the heart. We have to do measurements. We have to make sure that the pictures are perfect. We have to do the same measurements each time to really look at the function of the heart and things like that.
Carol Vassar, podcast host/producer:
And it’s an area within cardiology that lends itself easily to AI, according to Dr. Mark Friedberg, pediatric cardiologist at the Labat Family Heart Center in the division of Cardiology at the Hospital for Sick Children in Toronto.
Dr. Mark Friedberg, Hospital for Sick Kids, Toronto:
There’s specific goals and objectives, but there’s a complexity of factors and a wealth of information that just for the human brain is difficult to incorporate those together. And even if we knew how to incorporate those together, which we don’t, it would be difficult to understand how they relate to outcomes. Really, we’re using the data to tell us what the risk or what’s happening with the patients. And that’s something we’ve always tried to do. So take a non-biased approach of certain factors and parameters, not that we are biased towards thinking or the literature tells us that these are what are important or not important, but really let the data tell us. And in a much more comprehensive and complex way in that it’s not one data point telling us. It’s a wealth of data telling us.
AI can recognize what view a certain echo picture shows just in the same way that face recognition recognizes a face. It can improve image quality, which I think is a huge area that is untapped and I think will be built into the echo machines. So that operator and it’s not only operator dependent, it’s subject dependent because some people have better ultrasound windows than others. So you sometimes get poor pictures or good pictures from the same operator, depending on who the patient is. And so, AI can be used to enhance those images to fill in missing gaps when there are artifacts or other aberrations in the image or noise in the image. AI can clean that up much as being used in regular photographs to enhance a photograph, to sharpen the photo, to get rid of blurry pictures. It’s being used in that same way. So that’s one aspect.
And then once an AI algorithm can recognize a picture and recognize the borders of the heart, it can also quantify the parameters that we now do manually or semi-manually. That’s going to reduce a huge amount of variability because even if the AI algorithm is right or wrong, it doesn’t really matter as long as it’s consistently wrong by a certain margin of error that we know and we can track, and from time-to-time, we can tell what has changed. And so once you can quantify that, and all of this already exists, some of it is in routine practice already, but I would say it’s the beginnings. So once you can identify a chamber and you can identify its borders, you can then track that chamber over time as the heart squeezes and relaxes. And then, you can quantify many of the parameters that we use on a daily basis in every echocardiogram to quantify function.
So you can quantify the size of the heart and how much it’s ejecting, how well it’s ejecting. And then once you can track, and this has existed for a long time already, but we take certain parameters like Doppler, so just like the same technology that the cops use to trap speeding cars. And we have an envelope. I’m sure everyone has seen those, when we make a sound on the ultrasound and get a tracing, well, you can automatically detect that tracing, quantify it, and now you can turn that into numbers and into data that can then be incorporated into an algorithm, into a prognostic algorithm.
So, that’s just some advantage of how AI can help with imaging. But I think it’s far more than that because there was a very nice example here at the World Congress of people giving patients who live in remote areas who don’t have easy access to care handheld ultrasound machines. The patient can have minimal or little training, so you can train in pediatrics so the parents can get trained minimally on just where to put the echo probe. And the AI software will guide the patient by that view recognition on the correct image and then quantify that image. So you can imagine a patient that needs serial or repeated echocardiograms every week or every two weeks, or if there’s a change or they’ve started a therapy, how this can be very powerful and those images can then be sent in remotely, even quantified already, for the physician.
Carol Vassar, podcast host/producer:
AI-enhanced treatment is already happening across the US and the world to bridge the distance gap between patients and practitioners, according to Dr. Chowdhury.
Dr. Devyani Chowdhury, Nemours Children’s Health:
So, one place I know in Nemours also, we are using AI quite a bit. Our AI-based methodologies for rhythm management. This patient had ablation of the heart rhythm done by Dr. Temple at Nemours Cardiac Center. Dr. Temple came to me and said, I think this kid needs blah, blah, blah. I need to check an EKG in a month. I said, “Geez, he’s going to be in the Amazon. I talked to this family. They’re going to be living up the river.” When they come to the base camp, they have access to the satellite phone. In that satellite phone, so we got them this device, they carried that with them. So they come to the base, they take that picture through that satellite phone. They’re texting that to me, which I text to Dr. Temple and say, “Hey, the kid’s doing okay.” So this is what we are doing.
We have another patient in Burma like this mom. We sent the kid home. We couldn’t capture the rhythm while the child was here. The family had to go on a mission trip, talked to Dr. Temple said, “Yes, let’s give them a prescription of Atenolol. Don’t have them take it. When the child goes into a rhythm, tell the mom to record.” She did, sent it to Dr. Temple. He says, “Yeah, you’re okay to put her on atenolol.” Put that kid on Atenolol. Kid’s fine.
When they come to the US, we will take care of them. This is AI-driven algorithms that are allowing us to take care of these patients. And then once we start these kids on meds, I tell them, send me another tracing. They do.
Carol Vassar, podcast host/producer:
AI is also empowering community health workers in under-resourced areas such as Pakistan, according to Dr. Babar Hasan, associate professor and a consultant pediatric cardiologist at the Department of Pediatric and Child Health at Aga Khan University in Pakistan.
Dr. Babar Hasan, Aga Khan University, Pakistan:
Majority of the population, their very first contact with the healthcare system is actually a community health worker or a lady health worker, who may not have all the years of healthcare education that a physician will have or somebody. And that’s a fact that’s not going anywhere. It’s going to stay. The population is growing. It’s growing in the rural and peri-urban or urban slums, and the very first contact is going to be a community health worker. And so, how can we now use AI to really augment their ability to screen a child who’s at a higher risk of having an event and then, from there, refer them to a center that can take care of it? Just to give you a kind of an example around this, look at our pulse ox work. For example, our pulse oximeter AI it gives you a tool, or the tool at the community health level person, to pick a child who is at risk of having an event over the next one month of life.
So if they have the ability of picking that child and they actually have a clear kind of SOP or standard operating procedure around how to go about treating or managing that child, imagine the impact that can happen. So as soon as they see that there is something wrong, because a very specific tool was giving an abnormal number like pulse oximeter, and they refer, the problem that I’m facing upfront in a tertiary care center, is that I’m receiving kids who are presenting late to me, who are presenting with comorbidities. And immediately, you solve that problem, right? Because now the community health worker, they’re using a tool which is AI-powered, is able to pick that child very early and then send them over. And so when they come to me in my center, they’re not as sick as what I would be getting them if they came a week later or ten days later because nobody was able to pick them. So I think there’s a huge role in that manner.
Carol Vassar, podcast host/producer:
Another area of healthcare affected positively by the AI boom is cardiac research, where applications are wide-ranging, according to Dr. Chowdhury.
Dr. Devyani Chowdhury, Nemours Children’s Health:
AI research is actually happening in every aspect of cardiology. It’s happening in the community level space, happening at outpatient space, inpatient space, echocardiography, ICU. So everywhere there is AI, and the question is, at what level are you looking at AI? Do you call AI and machine learning when you’re developing predictive algorithms, right? So predictive algorithms are coming out everywhere. Who is the sicker patient in the ICU who needs more care, which echo doesn’t look normal, which fetal echo is abnormal? So these are all predictive analytics that are coming out from our existing data. So, that is a lot of AI. Then the other question becomes is how do we implement some of this AI to make things better? First, we have to learn that is AI as good as the humans. So that’s the first thing we are trying to figure out.
Does AI add anything to this? Does it make us better in any way? Does it make us more efficient? Does it bridge a gap of workforce training that we don’t have? We have an acute shortage of nurses. We put an AI algorithm in the ICU. Will that help a new nurse recognize a patient’s acuity better rather than having a ten-year experience? So that’s one space we are trying to see. But a lot of the AI is still in the research space. A lot of the algorithms haven’t been validated, but I personally feel like, because of my area where I work, I think at a community level, AI is going to be most meaningful to people who don’t even use technology. People without technology will benefit the most from this advanced technology.
Carol Vassar, podcast host/producer:
One person working on algorithmic validation in her research is Dr. Reddy, who is focused on using AI to automate processes associated with echocardiograms in both adults and children.
Dr. Charitha Reddy, Stanford Medicine:
Right now, we have an expert clinician, either a sonographer or a physician, who’s taking those measurements and actually manually tracing the borders of the heart to assess whether it’s squeezing correctly or not. And so, in the adult world, they often tend to be ahead of us because the resources are bigger. They have higher patient populations, and so they’re able to kind of develop some of this research a lot faster than we are. And so, on the adult side, they had developed something where they took the video clip of the heart and automated the process of assessing the function of the left ventricle.
So I teamed up with the same group and said, hey, I want to recreate this in pediatrics. But there’s a few key differences. One of the differences is that we use a couple of different views in order to assess the function that are different than the adults. So, can we use our pediatric-specific views? And the other is we have a huge range of sizes and ages that’s different than the adults. We have kind of going anywhere from neonates, zero to one month, newborns all the way to 18 to 25-year-olds.
And so I wanted to really make sure that we assess the full gamut of ages and sizes because the pictures on a little baby are very different than the pictures on a fully grown 18-year-old. And so I wanted to make sure that what we developed was functional for all of that. And so, when we took our dataset, we made sure to look at all the age ranges and sizes and also the image qualities that we would get in a typical clinical workflow. We didn’t pick the most perfect images. We picked everything. And then we did take anatomically or structurally normal hearts, so we didn’t use congenital heart disease. But we applied the algorithm just at baseline, the adult algorithm, to pediatrics, and it didn’t perform that well.
And then, we trained the model specifically on pediatric data only, and the model performed much, much better. And then, we did a combination of the adult model and the pediatric model to take how much data because the adult had so much more data, and that performed really well. And so we were able to then automate the assessment of left ventricular function, which is one of the big markers of heart health in some of our pediatric patients. We were able to automate that.
Carol Vassar, podcast host/producer:
About 35 miles down the road from Dr. Reddy’s base in Palo Alto. Dr. Rima Arnaout works as an assistant professor in medicine and a member of the Baker Computational Health Sciences Institute at the University of California, San Francisco. In 2021. She was the lead author of a paper published in the journal Nature, which found that a form of AI machine learning known as ensemble learning, the combination of insights obtained from multiple learning models, could significantly improve fetal congenital heart disease detection. But it requires data and lots of it. More than one human or even one group of humans could provide with the scale and accuracy required. Enter the machines.
Dr. Rima Arnaout, UCSF:
I’m an echocardiographer. I know there’s only a certain number of studies I can read in a day before my brain is just too tired, and I got to stop because I know I’m not paying attention like I should. Machines don’t have that problem. You can start to do analysis at a scale that a human could never, ever do. And that can change how we look at quality control. It can change how we look at precision medicine analysis. My postdoctoral work was actually in genetics, and we’re interested in precision medicine. These very large cohorts people are developing, where we have gene sequencing on everyone. What they want to do is they want to correlate the genotype, the gene sequence, with aspects of the phenotype, the physical characteristics. That might be height or weight, or whether you have a rash or whether you have diabetes, or what your imaging looks like.
So they are building these cohorts of million-plus people with all the DNAs already done. Who’s going to phenotype that? Who’s going to read a million echocardiograms for those precision medicine studies? Nobody. But a machine could. So that’s how it can power new precision medicine analysis, even for quality control, to return back to the fetal example. One thing that we found, even in a subset of our own UCSF data, we found that there are five screening views that the guidelines require to be in that fetal study. They were not always all present. Normal studies were more likely to have all five views than abnormal studies. One would want to look at that more. Let me analyze every single study we’ve had in the past five years and just see how complete they are for the guidelines, recommended views, and see whether that affects diagnostic accuracy. Nobody’s going to detect views in five years worth of fetal studies as a scientific study.
It’s not humanly possible. But the algorithm can do it, and you can imagine putting that algorithm at a point of care. You’re in the clinic with your patient, you’ve done the ultrasound, and before they leave, it’d be nice to get a little readout, “Hey, you missed something.” They’re still there. You can get it. So, these are the opportunities for scalable and automatable quality control.
Carol Vassar, podcast host/producer:
For patients, the use of AI technology is seamlessly integrated into their treatment and care. So much so, that most don’t even know that AI is being used to help their clinicians treat them. Dr. Chowdhury …
Dr. Devyani Chowdhury, Nemours Children’s Health:
We are using AI in Pakistan to detect which baby is going to die within the first month of life. These women who are delivering babies, we are using AI to predict which woman’s going to have a baby that is not going to be alive, who’s going to have a stillbirth. Which woman’s going to have a baby that is going to be small for gestational age because these small for gestational age babies are high risk. So the mother who comes in for an ultrasound, or when we go to the house and do a pulse oximeter, they have no idea that we are using AI to make their lives better. But in reality, we are trying to improve access to care for that population. Because just think about it: there are millions and millions of babies born with congenital heart disease. 90% of that population lives outside the United States.
Even in the United States, we have a huge workforce shortage where it is everybody claims we have too many people taking care of smaller number of patients. So, for us to catch up with that workforce isn’t going to happen that easily. We are not going to be able to train 50 million doctors in one day to take care of these children. So, the population is what it is. Their education is what it is. So our job is to make that community health worker who’s not even eighth-grade educated, to give them the same expertise that a pediatric cardiologist sitting in Wilmington, Delaware, can say, “This baby is not going to do well.” How do you do that? How do you give that baby that same chance of survival that you give a baby in America? Like I say, the future of a child should not be decided by geographical lines that are created by humans on a globe. So AI, I think, has that potential.
Dr. Shubhika Srivastava, Nemours Children’s Health:
I would like to take a step back and talk about the utilization of AI in access and medical decision-making. And that is integral to not only patient diagnosis but outcomes and really ensuring lifelong commitment to the patient’s care and quality of life. This requires, really, as Dr. Chowdhury mentioned, the application of data that you get from resources all around you to see how that data can be applied to a particular patient. Let me give you an example. We all went through COVID, where access to care was crucial. And as you all know, Nemours covers almost eight states and areas where patients do not have direct access to care. And it’s really important to use the information that we’ve learned from COVID, where access to care for certain patients was so limited, to see how we could apply that going forward to make sure we can get the best access to each and every patient.
MUSIC:
(Singing.)
Carol Vassar, podcast host/producer:
Thanks for listening to the first of a three-part series on AI in healthcare with me, Carol Vassar, and our guests, recorded at the World Congress of Cardiology and Cardiac Surgery in Washington, DC. Dr. Shubhika Srivastava, Dr. Devyani Chowdhury, Dr. Charitha Reddy, Dr. Mark Friedberg, Dr Babar Hasan, and Dr. Rima Arnaout.
What are your thoughts on the promise of artificial intelligence and healthcare? Leave a voicemail on our website, nemourswellbeyond.org. That’s nemourswellbeyond.org, where you can also listen to each and every one of our previous episodes and subscribe to the podcast. Thanks to our production team this week, Che Parker, Cheryl Munn, Susan Masucci, and Allison [inaudible 00:23:49]. You know, for all the hope that AI brings to healthcare, there are perils too. Not enough data, too much data, patient privacy concerns, legal and ethical considerations, the fear of biased algorithms, and public trust.
We’ll talk with the experts about all of that in our next episode, AI and Healthcare: The Perils. Until then, remember, we can change children’s health for good well beyond medicine.
MUSIC:
(Singing.)
