Globally Inclusive Brain Models Achieve Robust Accuracy Across Diverse Ethnoracial Groups

Writter: Ruiyang Ge (PhD, University of British Columbia)

---

A new study has provided important evidence that brain-imaging-based models can work reliably across ethnoracially diverse populations without requiring separate race-specific training. The findings address a long-standing concern in neuroscience and artificial intelligence about whether brain-based tools developed from large datasets can generalize across different populations.

Normative brain models are increasingly used to assess brain health, development, and aging by comparing an individual’s brain structure with expected patterns for their age and sex. These approaches are often described as the brain equivalent of pediatric growth charts and are considered a key step toward precision medicine in psychiatry and neurology.

However, many researchers have questioned whether such models might perform unevenly across populations because neuroimaging datasets have historically lacked diversity. Poor generalizability could limit the clinical usefulness of these tools and raise concerns about fairness and bias.

In the new work, researchers tested race-neutral models of brain morphometry using multiple independent datasets representing diverse ethnoracial groups. The models were trained on a large, globally inclusive dataset and did not use explicit ethnoracial stratification during model development.

The study found that the models maintained high accuracy across all tested populations, suggesting that sufficiently large and diverse training datasets may allow brain-based tools to generalize robustly without requiring separate race-specific models.

The findings support the idea that inclusive data collection, rather than ethnoracial partitioning, may be the most effective strategy for developing equitable neuroimaging tools. This work might represent an important step toward scalable and scientifically grounded brain-health technologies that could eventually support more personalized assessment of neurological and psychiatric conditions worldwide.