AI's Ability to Determine Racial Categories from Heart Scans and Its Implications

In recent advancements in medical artificial intelligence, researchers have developed models capable of estimating an individual's racial background solely from heart scans, even when the AI has no explicit instruction to do so. This breakthrough, documented in a study published in the European Heart Journal - Digital Health, reveals that an AI system could accurately classify whether a patient is Black or white with up to 96% accuracy based on heart imaging.

This finding raises critical questions about the objectivity and fairness of AI tools in healthcare. It underscores that AI algorithms, which are trained on real-world data, tend to absorb and reflect societal biases and stereotypes embedded within that data. Understanding that race is a social construct — a classification system devised by societies based on perceived physical traits rather than biological truths — is essential. Genetics show significant variation within racial groups and little distinction between them, highlighting that race is not a biological categorization.

Despite this, many AI systems inadvertently learn to associate social and racial categories because the data they are trained on is shaped by societal inequalities. These systems can pick up on indirect signals, such as differences in subcutaneous fat or image artifacts like motion blur, which correlate with racial differences due to social and biological factors like body composition and socioeconomic status.

The repercussions are profound because these biases can lead AI to reinforce disparities in healthcare. For example, if an AI learns to identify race rather than disease-specific characteristics, it may lead to misdiagnoses or unequal treatment, perpetuating health inequities.

Addressing these issues requires robust solutions, including diversifying training datasets to better represent all populations, implementing transparency measures such as explainable AI, and handling racial data with care to avoid reinforcing harmful stereotypes. As AI continues to transform healthcare—reading images faster, analyzing complex data, and optimizing diagnosis—integrity and fairness must remain central to its development.

This research emphasizes that AI systems mirror the world they are built upon and that unexamined biases can have serious safety implications. As such, the AI community must prioritize ethical standards to prevent AI from amplifying societal inequalities while harnessing its potential to improve health outcomes for all.