Deep learning to assess cardiovascular disease risk from chest imaging - PROJECT SUMMARY Vineet Raghu, PhD is a computer scientist whose career goal is to improve early detection and prevention of chronic disease by applying artificial intelligence to large datasets of medical imaging, genomics, risk factors, and outcomes to better estimate risk and derive insight into etiology. Dr. Raghu is an Instructor of Radiology at Harvard Medical School and research faculty at the Massachusetts General Hospital’s Cardiovascular Imaging Research Center. In this project, he aims to develop and validate deep learning-based cardiovascular disease risk scores from chest imaging. He will build upon his prior training in applied artificial intelligence and genomics to gain depth of understanding in the epidemiology, pathophysiology, and preventive care of cardiovascular disease. His specific training goals are to: 1) Gain understanding of how physicians and patients make decisions about preventive interventions in older adults, 2) Learn to apply statistical genetics techniques to single nucleotide polymorphism and epigenetic data to investigate molecular pathways of cardiovascular disease, and 3) Gain deeper understanding of the pathophysiology of cardiovascular disease to improve collaboration with physician-scientists, and 4) Develop grant writing skills and training in the responsible conduct of research. Dr. Raghu’s training plan includes didactic training from Harvard Medical School and Harvard School of Public Health and hands-on research training to gain the proposed skillset in preparation for research independence. His mentorship committee has diverse expertise in imaging, deep learning, genetics, geriatrics, and cardiovascular epidemiology to support his development. He and his mentorship team will have individual meetings to discuss specific research aims and advisory meetings to discuss broader career development goals. Dr. Raghu will supplement didactic training by completing the following innovative research aims. In Aim 1, he will develop deep learning models to estimate cardiovascular disease risk using chest CT. In Aim 2, he will identify genetic loci and biologic aging indices associated with CT-based risk estimates. Finally, in Aim 3, he will test whether CT-based risk estimates can identify persons at risk for incident cardiovascular events beyond current risk scores and screening criteria. These aims will be carried out in existing large epidemiologic cohorts comprising over 70,000 individuals in total. Completion of these aims will provide preliminary data for a potential R01 application in Years 4 and 5 to investigate whether CT-based models improve uptake and efficacy of preventive care and reduce cardiovascular events, to expand this approach to other imaging modalities, or to use imaging and multi-omics to identify molecular pathways of disease. This proposal will give Dr. Raghu excellent training to achieve his goal of being an independent investigator who develops deep learning-based risk scores and 1) collaborates with physicians to implement such models to improve preventive care and 2) uses such models to better understand disease etiology.
