Related Experiment Video
Updated: Jun 28, 2026
06:35
Basics of Multivariate Analysis in Neuroimaging Data
Published on: July 24, 2010
Advancing fair and explainable machine learning for neuroimaging dementia pattern classification in multi-racial and
Ngoc-Huynh Ho1, Sokratis Charisis1, Nicolas Honnorat1
1Glenn Biggs Institute for Neurodegenerative Disorders, Neuroimage Analytics Laboratory and Biggs Institute Neuroimaging Core, University of Texas Health Science Center at San Antonio, San Antonio, TX, USA.
Nature Communications
|June 26, 2026
Summary
This study addresses bias in dementia classification models across racial and ethnic groups. A new method, RegAlign, significantly reduces performance gaps, improving fairness in AI-driven medical diagnoses.
Area of Science:
- Neuroimaging and Artificial Intelligence
- Medical Diagnostics and Health Equity
Background:
- Dementia diagnosis requires precision, but current methods show bias across diverse populations.
- Inconsistent performance in AI models raises concerns about fairness and clinical reliability for underrepresented groups.
Purpose of the Study:
- To investigate performance discrepancies in dementia classification across Non-Hispanic White, Non-Hispanic African American, and Hispanic White populations.
- To evaluate RegAlign, a novel few-shot domain adaptation technique, for improving dementia classification in underrepresented groups.
Main Methods:
- Analysis of dementia classification performance on datasets from 6584 Non-Hispanic White, 1263 Non-Hispanic African American, and 713 Hispanic White individuals.
- Application of RegAlign, incorporating source-side focal learning, target-side class-weighted supervision, and class-conditional alignment.
- Evaluation of RegAlign's effectiveness in reducing cross-group bias and improving adaptation to underrepresented populations.
Main Results:
- Significant cross-group bias was observed in dementia classification models, particularly when trained on one population and tested on another.
- RegAlign substantially reduced inter-group performance gaps, showing notable improvement between Non-Hispanic White and Hispanic populations.
- The fairness-aware learning strategy demonstrated effectiveness in mitigating bias for underrepresented groups.
Conclusions:
- Fairness-aware learning strategies and diverse training data are crucial for enhancing the accuracy and equity of MRI-based dementia classification.
- RegAlign offers a promising approach to address bias in AI diagnostic tools, promoting more equitable healthcare outcomes.
- Further research into diverse datasets and adaptive learning is essential for reliable and fair AI in clinical settings.
