Globally Inclusive Brain Models Achieve Robust Accuracy Across Diverse Ethnoracial Groups

Writer: Dr. Ruiyang Ge @ University of British Columbia

---

A new study has published in PNAS found that models based on brain scans can work well for people from many different ethnoracial backgrounds, without needing to build separate models for different racial or ethnic groups. The findings help answer an important question about whether brain-imaging tools developed using large datasets can be used fairly and accurately across diverse populations.

These brain models compare a person's brain scan with what is typical for someone of the same age and sex. They work much like children's growth charts, helping researchers and doctors identify whether a person's brain is developing or aging differently than expected. Scientists hope these tools will one day help diagnose and monitor brain disorders more accurately.

One concern has been that many brain-imaging datasets have not included enough diversity. If a model is trained mostly on one group of people, it may not work as well for others, raising concerns about fairness and reliability.

In this new work study, researchers tested their brain models using several independent datasets that included people from a wide range of ethnoracial backgrounds. The models were trained on one large, globally diverse dataset and were not designed separately for different populations.

The researchers found that the models performed consistently well across all of the groups they tested. This suggests that using large and diverse training datasets may be more effective than creating separate models for different ethnoracial groups.

The study highlights the importance of including people from many backgrounds when building medical AI tools. Doing so could help ensure that future brain-imaging technologies are accurate, fair, and useful for people around the world, ultimately supporting more personalized care for brain health conditions.