Conversations With Cardiologists: G. Hamilton Baker, MD

G. Hamilton Baker, MD, associate professor of pediatrics at the Medical University of South Carolina (MUSC) and co-director of the MUSC Artificial Intelligence Hub, shares his advice with ACC FITs.

What is the mission of the MUSC Artificial Intelligence Hub and what are some accomplishments that have come of it?

The mission is to expand and promote the use, understanding and research of artificial intelligence (AI) and related technologies at MUSC. We hope to weave AI into the three-part mission of MUSC to optimize human health, provide medical education and perform medical research. It was created by me in collaboration with an informaticist with the goal to bring everyone together as a central "Hub." We created a speaker seminar that started a community that has grown, and our membership quickly increased past 150 people. The pandemic helped us grow as we were able to record the talks on Microsoft Teams and Zoom and people could watch them on their own time. There is now student involvement and an AI consultation service to help integrate AI and machine learning (ML) into proposed research projects. It is nice as they do not require a background in ML or data science. We try to connect the researchers with faculty at MUSC or the AI Hub partners at Clemson University. The goal is to have investigators learn how to integrate AI in their projects and realize that it differs from traditional research. The Hub has continued to grow, and we now have obtained grant funding to continue to support new projects.

How did you find yourself as a founding member of this novel and future-driven organization and what were some of the biggest hurdles to clear in getting there?

This was one of those circumstances where you have to stay as positive and optimistic as possible. There weren't a lot of people buying into the idea of AI and ML at the time. We had to remain convinced that we needed this. It was important that we needed to construct this infrastructure, and fortunately, that ended up being the case 3-4 years down the line. The main obstacle was getting everyone in the same room as frequently as possible, and that is still a challenge. At many institutions, there is a separation of clinical and research personnel. Initially, we had to determine "where did this belong?" Ultimately, this ended up being on the research side. I needed to find the right people to help me get started and who were interested in getting things done.

For aspiring cardiology fellows, what are some ways that they could get started for a career of using and/or studying AI?

One is to realize that AI and ML have very wide applications. Understanding AI can be a very useful skill to build. It is also important to decide what roles you want to play on teams that are using AI/ML. Figure out what part of the team you want to be. Probably, the most realistic part of the team for a clinician would be to learn enough about ML and data science so you can interact with specialists in a productive way as part of the team. You get to be the person who comes up with the ideas because you understand the clinical and biomedical applications of AI. You can then pitch this idea to the data science team and work together with them.

At the MUSC AI Hub, we have partnered with Clemson University to get access to faculty and students who are willing and interested in providing data scientists and machine learning expertise. We then grew that into an AI/ML consult service that anyone on campus can utilize, a sort of AI as a service model. It is important to consider any intellectual property from the onset of a project, so we work closely with the MUSC Foundation for Research and Development who manage university IP. This way we can continue to work with many other institutions synergistically.

I have seen data science teams work in isolation without a clinician, and they don't work as efficiently. That synergy between the two is fantastic, and that is essential to a successful program at any institution. That is the most rewarding and effective role we can play as clinicians.

What skills are needed to pursue AI? How important are statistics or coding for the clinician?

You need to have enough of a grasp of it so you can have productive discussions with machine learning experts about your project and speak the same language. However, you don't need to have an in-depth biostatistics background, and you don't have to be able to code an entire deep learning neural network. You just need to understand how they work, have a basic understanding of statistics, and understand what you don't know so you can ask the questions to fill in gaps and to be a productive member of the team. A basic understanding of coding can be helpful but isn't 100% necessary.

How do you predict AI will change the way we do things in cardiology?

In the short term, we have overestimated the effects AI will have on clinical pathways. I think the way AI and ML are going to affect medicine clinically will be in the background. You won't even know that it is ML. There will be algorithms that can clear out your schedule when needed and insert a double booking when you have someone predicted to "no-show." There will also be basic and robotic automation that will take over simple, but time-consuming processes.

In the long-term, I think some of the bigger projects will become more widespread, such as with diabetic retinopathy. This has already been implemented and is really helping people. There are also projects being done at the Mayo Clinic where they trained a neural network to determine the ejection fraction being above or below 35%, just based on the EKG. That's valuable! That can decrease the number of echocardiograms performed and get value-based care going. I can see this being applied to multiple areas of medicine in 5-10 years or more.

How is AI playing a role with your background as an interventional cardiologist with a focus on image modeling and image guidance?

I think an area you will see AI will be reducing clinical variation in patient selection and procedural approach. Many studies have shown that outcomes are more related to picking the right patients for the given procedure, particularly on the larger-population scale. AI will be able to help guide clinicians and increase consistency. I think in pediatric and congenital cardiology this will be a little more difficult due to the heterogeneity of the patient population. This has been consistent with most technologies. I also think you will see AI being applied to visualization and procedure guidance in the future.

What are some of the pitfalls of AI and some factors we need to be cautious of?

I could put it as one word: development. The biggest obstacle is how are we going about developing these tools. This is currently being done by a lot of people putting together these data sets. We need to step back and realize how complex and error-prone the process is. The need to account for bias has been identified very early in the process of using AI and ML in medicine. This is slowing the process down and people are finding that a lot of energy is needed before and after an algorithm is developed. There is so much work that is done on the front end to make sure your data is clean, fair and bias-free. Overall, the biggest obstacle is just to be aware that these biases exist. Currently, the NIH has launched a program "Bridge to AI" to try to generate a large amount of data quickly that are free from bias and ML ready.

What general advice do you have that all fellows could benefit from?

Do something that you like and something you are passionate about. If you read about AI and ML and find it interesting and you want to learn more – that is a sign you should get in touch with somebody who can serve as a mentor and get involved. We need as many people as possible to make the implementation successful. Now is the time to dive in.

This article was co-authored by David Leone, MD.

This content was developed independently from the content developed for This content was not reviewed by the American College of Cardiology (ACC) for medical accuracy and the content is provided on an "as is" basis. Inclusion on does not constitute a guarantee or endorsement by the ACC and ACC makes no warranty that the content is accurate, complete or error-free. The content is not a substitute for personalized medical advice and is not intended to be used as the sole basis for making individualized medical or health-related decisions. Statements or opinions expressed in this content reflect the views of the authors and do not reflect the official policy of ACC.