Lay summary
Technological advancements in machine learning and brain-MRI big data have led to rapid development of neuroimaging biomarkers useful for neurological/psychiatric disorders (e.g., Alzheimer’s and schizophrenia). But brain-MRI big data have predominantly been collected from non-diverse populations, mainly people of European descent. This has created an ethnicity bias: the neuroimaging biomarkers do not necessarily apply well to people not of European descent. For science to ensure equitable benefits for all, it needs to reflect the uniqueness of each of the world’s populations. We will harness the power of three novel interdisciplinary techniques to reduce the bias, including neuroimaging (i.e., combining different types of brain images), genomics (i.e., using genetic information to adjust the bias) and artificial intelligence (i.e., using deep-transfer learning to fine tune the biomarkers). Our project will radically change the way neuroimaging biomarkers are applied: as opposed to assumed universality, we can test the ethnicity bias and reduce it.