Related Experiment Video
Updated: May 12, 2026
10:45
Brain Imaging Investigation of the Neural Correlates of Observing Virtual Social Interactions
Published on: July 6, 2011
12.2K
No Normal Brain: How Demographic Exclusion Undermines Neuroimaging AI Validity
Journal of Clinical Epidemiology
|February 22, 2026
Summary
Neuroimaging AI trained on non-diverse data risks inaccurate clinical decisions. Addressing demographic gaps in datasets is crucial for AI safety and scientific validity, recognizing neurological diversity as biological reality.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Imaging
Background:
- Neuroimaging AI systems are increasingly used in clinical decision-making.
- Demographic exclusions in training datasets may compromise AI scientific validity and clinical safety.
- Concentration in datasets can undermine the principle of neurological diversity.
Purpose of the Study:
- To examine demographic concentration in major neuroimaging repositories.
- To analyze how this concentration impacts AI systems.
- To investigate the potential for AI to pathologize neurological diversity.
Main Methods:
- Analysis of five major neuroimaging datasets (>55,000 participants).
- Collection of participant characteristics (race, ethnicity, sex, age, geography, socioeconomic status, education).
- Review of literature on AI performance disparities across demographic groups.
Main Results:
- Datasets show severe demographic concentration (e.g., 52-94% White, 85-100% North American/European).
- AI performance disparities observed, including lower seizure detection in females and higher false positives in dementia screening for Black participants.
- Bias amplification noted in COVID-19 brain aging research using restricted data.
Conclusions:
- Demographic exclusion in neuroimaging AI poses a significant threat to research validity and clinical safety.
- Diversity, equity, and inclusion initiatives are essential.
- Acknowledging neurological diversity is key to scientific rigor and AI safety.
