Investigating underrepresentation in AI datasets for cardiovascular health asssessment

Dunn, Craig

Student Work

Investigating underrepresentation in AI datasets for cardiovascular health asssessment

Public Deposited

Cardiovascular diseases (CVDs) are the single most common global cause of death, causing around 30% of all deaths globally. With recent advancements in computer technology, Machine Learning (ML) and artificial intelligence (AI) are now common approaches for CVD research. However, while CVDs manifest differently across race/ethnicity and gender groups, it is unclear whether CVD datasets utilized in research include adequate representation of all races/ethnicities and gender, which could lead to inaccurate results. The research characterizes the representation of various racial, ethnic and gender groups in CVD datasets utilized in CVD ML and AI studies. After analyzing 11 CVD datasets, it was found three datasets which included information on race/ethnicity and gender, all of which were demographically consistent with the US Census. However, the remaining 8 datasets reported on neither the race/ethnicity nor gender of study participants. Additional investigation is necessary to quantify the existence and impact of misrepresentation across demographic groups in CVD research. In addition to that, a website describing this work was developed, as an easy and accessible way of communicating the content produced during this research. The website went over usability evaluation to ensure that the correct message was being communicated. Participant feedback received was generally encouraging.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator