In this brief presentation, pediatric cardiologists Kami Gill, MD, and Aneela Reddy, MD, describe what UCSF’s program has to offer patients as early as 11 weeks into a pregnancy, including how specialists have harnessed artificial intelligence (AI) for better detection of congenital heart disease and how they collaborate across multiple disciplines to inform and support families from the moment a diagnosis is made. Learn why access to early fetal echocardiography is priceless and glimpse a future where providers in less resourced areas could deliver enhanced maternal-fetal care thanks to an advanced method of screening for heart abnormalities.
Hello, my name is Anila Reddy, and I'm one of the pediatric and fetal cardiologists here at UCSF. Today, Doctor Cammy Gill and I would like to highlight some of the incredible advances in fetal cardiology here at UCSF and how this impacts both our patients and our referral base. Today we will touch on who we are as a program, some of the exciting work we are doing, and how to get a patient referred to us. Both Doctor Gill and I are part of a large team shown here all within the fetal cardiovascular program here at UCSF uh, all the cardiologists are listed here as well as our nurses. We work in collaboration with the pediatric Heart Center and the fetal Treatment Center, serving as a world leader in diagnosing and addressing birth defects before delivery. We are also incredibly dedicated to a holistic experience for both the mother and baby affected with heart disease. This means that the fetal heart care, maternal care, delivery center, and ultimately where the heart surgery will take place are all in one place. We're also part of the neonatal cardiovascular Center of Excellence where we have a dedicated specialized intensive care unit for babies with congenital heart disease. This is proven to lead to better outcomes for both the mother and the baby in the near and long term. The introduction of fetal cardiology has transformed the prenatal experience for families as well as significantly improved perinatal outcomes for patients affected with congenital heart disease. When families are referred to us for their initial fetal echo, we have a dedicated counseling session that focuses on preparing families on what it means to have a child with heart disease and ensuring they feel adequately supported every step of the way. The cardiac screen, including the 4 chamber view outflow track views, and 3 vessel trachea view have been part of the routine obstetric care for over a decade. There are 5 typical screening views that are completed in the OB screening ultrasound to accurately screen for heart disease. Typically when a patient is referred for a fetal echocardiogram, this is done around 18 to 22 weeks. Sometimes this can be done a little bit earlier at the 13 to 14 week mark. The scan itself takes about 45 to 60 minutes, followed by a detailed counseling session with the family and the cardiologist. 60% of congenital heart disease should be identifiable on the basis of an abnormal four chamber view. Up to 40% will have an abnormal outflow track view or a three vessel trachea view. So, although we should be able to screen for heart disease with all these diagnostic screening views, with the 4 chamber outflow tract and 3 vessel view, in practice, our fetal diagnosis rate is as low as 30% nationally. As a world leader in fetal cardiovascular research here at UCSF today, we would like to highlight two aspects of our program that are working to improve prenatal detection rate here at our institution as well as nationally. First, Dr. Gail will touch on early fetal echocardiography, and then I will talk about artificial intelligence within the fetal cardiology space. Hello, my name's Cammy Gill. I'm also one of the pediatric and fetal cardiologists here at UCSF. Today I will be sharing with you our work in the early fetal cardiology space and its availability to our patients. An early fetal echocardiogram can be completed as early as 11 weeks of gestation, allowing us to identify complex congenital heart disease in the late first trimester. Advancements in technology along with refinement of our scanning techniques has allowed us to provide accurate and reliable cardiac diagnosis for an emerging population of high risk patients. Taking a look at early embryology, cardiac development begins in the 3rd week and is largely completed by about 9 weeks gestation. Initially, the heart begins as a single tube. This tube undergoes a series of complex folding and septation to form a 4 chamber heart with two great vessels. At 9 weeks, the heart is still too small to be imaged well, and it's less than 6 millimeters in size. At about 11 weeks, the fetal heart is large enough to obtain diagnostic resolution. The 11 week mark also corresponds in timing with the assessment of nucle translucency. The measurement of nucle translucency is a screening tool that's offered to all pregnancies and it's now the standard of care. It's typically completed at about 11 to 14 weeks gestation and involves measuring the fluid filled space behind the back of a fetus's neck. It was initially introduced when it was noted that an increased NT is associated with genetic abnormalities, which in turn are associated with various forms of congenital heart disease. By the mid 90s, it became apparent that even in the absence of genetic abnormalities, an increased nucle translucency was directly associated with congenital heart disease. As seen in this graph, there's a direct correlation between increasing NT and the risk of congenital heart disease. As mentioned, the nucle translucency is measured as early as 11 weeks gestation, and a standard fetal echocardiogram is completed at 18 weeks. Without the availability of early fetal cardiac imaging, this may leave women in the high risk population waiting nearly 2 months for a fetal cardiac assessment. An early fetal echocardiogram provides mothers with access to a prompt evaluation. In addition to increased nucle translucency, there are indications for an early fetal echocardiogram which include a family history of congenital heart disease, abnormal genetic screening results, exposure to any tratogens during your pregnancy, abnormalities in other organ systems, or the presence of certain maternal medical conditions. This sweep captures all 5, beginning at the abdomen, all the way up to the 3 vessel tracheal view in a fetus that is 13 weeks gestation. The addition of color Doppler allows us to assess valve function and blood flow through the heart and its great vessels. Advanced technology available in our ecolab allows us to better assess intracardiac structures and small vessels with high resolution and in great detail. Scanning in the sagittal plane with color Doppler optimization allows us to assess the fetal arches. Here we can see the ductal arch. And here we can see nice laminar flow through the fetal aortic arch. The impact of early fetal cardiac diagnostics in pregnancies where an abnormality is detected is palpable at the bedside. It allows time for parents to process information regarding the nature of the fetal condition, potential treatment options available, anticipated outcomes and opportunities for genetic testing. This time is invaluable for parents to absorb the news, make decisions that are best for their family in pregnancies where the early fetal echocardiogram is normal, it provides a degree of reassurance and relief in the setting of a high risk pregnancy. Here at UCSF we've been part of international efforts to better understand and push the frontier of early fetal cardiology with the involvement both in bedside research all the way up to national guidelines which set a standard for fetal care in the field. We see early fetal patients on both sides of the bay, primarily at the fetal treatment center in Mission Bay and also at UCSF Children's Hospital in Oakland. Doctor Reddy will now speak on our work in the artificial intelligence space. Thank you, Doctor Gi for touching on the important work on early fetal echo here at UCSF. I know firsthand that this has made a huge impact on our families affected with heart disease and planning their pregnancies. I will now switch gears to talk about how here at UCSF we are leveraging artificial intelligence to one improve prenatal detection rate of heart disease, and 2, creating a scalable screening tool that can hopefully be used by any institution nationally or internationally to achieve the same quality of care regardless of where you live, addressing one of our missions here at UCSF to promote health equity. There are now lab here at UCSF, which I have had the pleasure of working with, published in Nature in 2021, are exciting data that shows that a deep learning model can generate expert level prenatal detection of complex congenital heart disease. Neural networks or deep learning is basically just a type of machine learning algorithm that is specifically meant to analyze image data. This model was trained to distinguish normal versus an abnormal heart, um, between 18 to 24 weeks gestation. And was tested on over 4000 studies, equating to about 4 million images. The model was then found to accurately detect 95% of congenital heart disease in comparison to the national fetal cardiac diagnosis rate of 30 to 50%. This preliminary data has incredibly profound clinical implications for the standard of prenatal detection. Here's an overview of how this model works. There is a view classifier that identifies 5 axial views all listed here. Then the views get compiled into a composite rules-based classifier to determine if the heart is normal or abnormal. Another component to this model is the segmentation model where the axial 4 chamber view shown here gets plugged into a segmentation model which is trained to extract fetal cardiac biometry such as cardiac axis, cardiothoracic ratio, and fractional area change for each cardiac chamber. Just for some context, here is how cardiac axis is measured. The normal heart is on the left and the right image is with extreme left axis deviation highly associated with conotrunkal defects. Cariothoracic ratio is measured as the ratio of heart circumference to thoracic circumference. Areas for each cardiac chamber are derived from drawing borders of each chamber in both diastole and systole, and then subsequently measuring the area change or area ratio. Our initial data showed that the biometrics extracted from this model correlated with literature norms for each of these indices. Since our preliminary data on segmentation was very promising, we decided to take it one step further to see if segmentation alone can help to determine normal versus abnormal hearts. We thought this would have the most relevance to the screening ultrasound because the axial four chamber view is the most reliably obtained cardiac view in the OB screening ultrasound. So we really wanted to leverage this view to optimize our screening capabilities. Here you can see two examples of four chamber views, the left being a normal heart and the right being severe heart disease with a univentricular heart, whereas here there can be much more subtleties present in this view. Here on the left being a normal heart, and on the right tetralogy of holo with moderate pulmonary stenosis, eventually requiring neonatal surgical repair. So our focus of this next study was to figure out how to leverage all these biometrics to pick up much more subtle findings and improve the prenatal screening utility of the A4C view to distinguish normal versus abnormal. This data was published in JAS in 2024, where we focused on creating a composite tool of all six of the biometrics listed here to distinguish normal versus abnormal hearts. And the reason why artificial intelligence is so critical here is it's extremely time consuming and tedious for a human to perform each of these biometric measurements at the time of the screening ultrasound. But computer aided automation via deep learning can make this much more accessible and possibly even more accurate than human measurements. We found that using all six of these biometrics in combination had a sensitivity of 87% for detecting heart disease in comparison to the current clinical sensitivity of the axial 4 chamber view, which is closer to 50-60%. So by adding the biometric clustering, we are indeed improving the screening capability of the axial 4 chamber view. This has incredible implications for increasing the screening rates and a scalable tool that could hopefully be available to obstetric providers anywhere. There's still a lot more work that needs to be done to retrain the model with obstetric ultrasounds instead of fetal echoes and to better automate this process, but here's just a glimpse into the future where deep learning models like this can serve as an effective tool that obstetric providers can use anywhere. Lastly, I want to shift to our patient experience here at UCSF. From the moment of diagnosis, which can be incredibly overwhelming through the critical newborn period and beyond, we offer a holistic multidisciplinary approach. We work closely with the fetal treatment center so that when a fetus is diagnosed with a heart problem. They are meeting with the MFM doctor to talk about their obstetric care, a genetic counselor, and a social worker, typically all within the same day, so they leave with a pragmatic sense of what it means to have a child with heart disease. Families are also given the opportunity to tour the hospital or to meet with a heart surgeon prenatally so they can feel adequately prepared for the delivery and the postnatal period. Logistically we have two sites for our fetal treatment center, one in San Francisco and the other one in Oakland, but we have 14 clinic locations all over the Bay Area, all with the capability to do a fetal echocardiogram. So although some of the more detailed delivery planning will occur at the fetal treatment center in San Francisco or Oakland, the initial and follow up echoes throughout the pregnancy can often happen at a location much more convenient to the family. If you'd like to get in touch with our fetal cardiac program, please scan the QR code seen here or use the information on the slide. Doctor Reddy and I would be more than happy to answer any questions about your fetal patients. Please don't hesitate to reach out.