1. Artificial intelligence in cardiology is more than using deep learning for cardiovascular imaging interpretation and can be extended into wearable technology (edge AI), administrative tasks (robotic process automation), and decision making (recurrent neural network for time series data).
2. In working with EHR, more data is not always better but data quality and selection impact greatly on the performance of the model.
3. Good data and full understanding of cardiovascular data are key for coupling with machine learning leading to good clinical insights (Good AI, like good cooking, starts with good ingredients).
4. An individual can have normal or elevated blood pressure throughout the day so we need to reconcile this issue of variation in physiological parameters.
5. Collaborative artificial intelligence is combining clinician with artificial intelligence rather than these stakeholders being separate forces (guidelines can be embedded into the EHR).
6. An application programming interface (API) would be very useful for data and guidelines to be intertwined and lessen clinician burden.
7. Machine learning can be very helpful for RV volume determination with MRI for its automation and accuracy as well as improved understanding of RV mechanics.
8. Rather than memorizing EKG criteria for diseases like hypertrophic cardiomyopathy, it would be useful to have a bottom up approach using machine learning and EKG data from disease positive patients (let the data speak for itself).
9. Using machine learning for atrial fibrillation detection to make this diagnosis expediently to decrease the chance of high morbidity.
10. The imbalanced data set problem is one of the most common issues in biomedicine but the prevalence of heart disease populations is high enough that this does not need to be an issue.
11. The more difficult aspects of using analytics and artificial intelligence in cardiology is the steps that involve data (access, curation, storage).
12. Natural language processing, with emergence of GPT-3 and combined with machine learning, can be a powerful tool in clinical insights from unstructured data.
13. Future direction will be using NLP for multimodal data fusion towards contextual artificial intelligence that will be much more helpful for clinicians.
14. Using mobile technology and digital medicine to help screen and follow up cardiac patients will need machine learning to accommodate the data.
15. Artificial intelligence is a good tool for risk stratification scores and adding additional features such as retinal images of sensor data.
16. Reinforcement learning and unsupervised learning as well as semi-supervised learning will be additional methodologies for disease management in cardiac patients.
17. Machine learning can free up time from repetitive low-level work to higher caliber of interpretation and care.
18. We may consider being fair to the artificial intelligence tools as we as human clinicians are often far from perfect.
19. We need to have a sense of urgency to explore all the possibilities of utilizing artificial intelligence tools and concomitantly have the patience and perseverance needed for long term success.
20. While the ethics and bias issues are essential in the deployment of artificial intelligence tools in cardiology, we should also consider the ethics of not involving AI tools that can improve outcome.
Thank you faculty and attendees for your knowledge and expertise as we all learned a great deal today at AIMed Cardiology!
Anthony Chang, MD, MBA, MPH, MS
Chief Intelligence and Innovation Officer
Medical Director, The Sharon Disney Lund
Medical Intelligence and Innovation Institute (mi3)
Children’s Hospital of Orange County