Incorporating AI Training into Cardiology Fellowship
Written by Arun Umesh Mahtani, MD MS and Gurleen Kaur, MD
Education within Cardiology fellowship predominately focuses on the Core Cardiovascular Training Statements (COCATS) that cover specific fields of competencies in the various subspecialties and include curricular milestones within each domain. Beyond these core COCATS requirements, Cardiology fellows of the current era also need to be equipped to understand and effectively use artificial intelligence (AI). As AI tools continue to grow and become more widespread within healthcare and science, having competencies in how to leverage AI to improve processes will be essential. Utilizing AI effectively in both clinical care and in the research setting will be an important skill. Many medical schools have already embraced AI into their curriculums. For example, the George Washington School of Medicine offers third and fourth-year students an elective on AI Applications in Healthcare, Harvard Medical School has introduced a one-month introductory course on AI in health care, and the Icahn School of Medicine at Mount Sinai has provided all medical students with access to OpenAI's ChatGPT Edu, a subscription-based version (1, 2). Also, the University of Michigan Internal Medicine Residency program has piloted a curriculum on data-augmented, technology-assisted medical decision making to teach trainees the fundamentals of AI and machine learning (ML) concepts (3). How incorporation of AI will influence the inherent aspects of medical training remains to be determined.
While currently there is no standardization or widespread curriculums for Cardiology fellows to learn the nuances, best practices, and pitfalls of AI, the above-mentioned examples from various medical training levels can certainly be used as models. Furthermore, here are some potential ways in which AI training could be incorporated within Cardiology fellowship.
1) Didactics
Most cardiology programs have some form of core didactic sessions available to supplement learning that happens on the clinical front. These conference sessions can include a series pertaining to AI that allow fellows to understand the nuts and bolts and foundational concepts of AI, ML methods, how these tools can be effectively used, and what are things to be cautious about, including model hallucination and biases, model generalizability, data privacy, and appropriate oversight. A variety of formats can be used for didactics including lectures from experts, case-based conferences, journal clubs with articles featuring the use of AI within cardiology research, hands-on workshops, etc.
2) Asynchronous Digital Education
Beyond the traditional in-person lectures that fellows participate in, there are additional ways that Cardiology fellows can self-direct their learning related to AI. This includes asynchronous podcasts, webinars, educational blogs, etc. Advantages of this mode of education include being tailored to the adult learner who is problem centered, self-directed, internally motivated, and time constrained and so can seek out resources that best fit their unique learning styles and schedule. Professional societies can play an important role in this as they have access to experts from varying institutions who could come together to host webinars, panel discussions, and other modes of content that could be disseminated for fellows to learn from.
3) Scholarly Activity
For fellows with a particular interest in incorporating AI into their career as a Cardiologist, working with mentors who use AI in their research and innovation endeavors can help them gain necessary skills. For example, The CarDS Lab at Yale and Dr. Ouyang's lab at Kaiser Permanente uses applied data science and AI to lead innovations in cardiovascular care and has several trainees working with the group on scholarly projects (4). The program for AI and Research in Cardiovascular Medicine at University of Kansas is another collaborative research program involving cardiologists, fellows, residents, medical students, and data scientists that has access to a robust digital cardiovascular database for cutting edge research (5).
4) Sponsorship/Funding or Dedicated Time
There are several national conferences related to AI in healthcare as well as coursework and certificate programs available online which could be an option for fellows interested in diving deeper during their non-clinical portion of Cardiology Fellowship. Cardiology fellowships could consider ways to sponsor and offer funding for such opportunities. Furthermore, for fellows who want to take a step further into the world of AI, there are opportunities to obtain extra degrees or dedicated training. For example, the AI Fellowship in Cardiovascular Disease at Northwestern School of Medicine offers an extra year of education in which fellows have an immersive experience in computation, programming, and statistics and can obtain a Master of Science in AI degree (6).
We propose a blueprint for a 5-part AI didactic curriculum that can be incorporated into general Cardiology fellowship with the aim of covering the core competencies of the field. In addition, this blueprint can serve as a guide for cardiology fellowship programs that are brainstorming ways to incorporate an AI based curriculum into their current training schedule. The proposed curriculum can be flexible and be tailored based on the cardiology fellowship program's structure and fellows' interest. Optionally, hands on experience using different programming languages like Python or R and/or different notebooks like Google Colaboratory or Jupyter (7,8) to code can be taught to fellows interested in learning in more detail.
1) Introduction to AI and ML
The objective of this module is to cover the past, present, and future of AI and ML
2) Understanding Datasets
The objective of this module is to cover the different types of datasets, formats, currently available open datasets in cardiology, cleaning datasets, and performing exploratory data analysis prior to designing an ML algorithm.
3) Understanding Different ML Algorithms
The objective of this module is to understand the different algorithms currently used. The functioning of each algorithm will be explained in detail along with interpretation of results and examples of real-world use cases. Following are the different algorithms that can be covered as part of an introductory curriculum:
a) Unsupervised Learning
Clustering Methods: K Means, K Medoids, Fuzzy C Means, Hierarchical, Gaussian Mixture Models, DBSCAN, Agglomerative,
Dimensionality Reduction Methods: PCA, LDA, FA, t-SNE, MDS
b) Supervised Learning
Regression: Linear, Logistic, Lasso, Ridge, ElasticNet
Bayesian Methods: Bayesian Ridge Regression
Decision Trees: Random Forest, Gradient Boost, AdaBoost, HistGradientBoost
Support Vector Machines
K Nearest Neighbor
c) Natural Language Processing
Text Preprocessing: Cleaning text-based data, performing tokenization, lemmatization, and stemming. Understanding stop words, parts of speech tagging, and regular expression
Text Representation: Embeddings
Different Tasks: Text classification, Sentiment analysis, Text Summarization
Introduction to Large Language Models
d) Introduction to Deep Learning
4) Model Evaluation
The objective of this module is to understand different metrics to evaluate model performance and tuning of parameters to extract optimum performance. Other core concepts to cover in this module includes:
a) Understanding different types of Bias
b) Overfitting and Underfitting of models
c) Internal and External Validation
5) AI Ethics and Regulation
The objective of this module is to understand concepts regarding AI ethics and governance. Topics covered in this module include:
a) Privacy
b) Transparency
c) Explainability and Fairness
d) Accountability
e) Ethical AI Design
f) Current Laws and Governing Bodies Regulating AI
g) Content Copyright and Plagiarism
References:
1) https://www.aamc.org/news/medical-schools-move-worrying-about-ai-teaching-it
2) https://www.mountsinai.org/about/newsroom/2025/icahn-school-of-medicine-at-mount-sinai-expands-ai-innovation-with-openais-chatgpt-edu-rollout
3) https://pubmed.ncbi.nlm.nih.gov/40378182/
4) https://www.cards-lab.org/
5) https://www.kumc.edu/school-of-medicine/academics/departments/cardiovascular-medicine/research/program-for-ai-and-research-in-cardiovascular-medicine.html
6)https://www.medicine.northwestern.edu/divisions/cardiology/education/fellowships/ai-fellowship-cardiovascular-disease.html
7) https://jupyter.org/
8) https://colab.research.google.com/
Written by: Arun Umesh Mahtani, MD MS, a first year Cardiovascular Disease Fellow at VCU, and Gurleen Kaur, MD, a first year Cardiovascular Disease fellow at Brigham and Women's Hospital.
Edited by: Francisco Ujueta, MD MS
