Understanding Sources of Demographic Predictability in Brain MRI via Disentangling Anatomy and Contrast

2026-03-04

AI Summary

This work addresses the confounding influence of demographic attributes—such as age, sex, and race—on brain MRI analysis, where their predictability often conflates genuine anatomical variation with acquisition-related contrast differences, obscuring the true source of bias. To disentangle these factors, the authors propose a decoupled representation learning framework that separates MRI data into an anatomy-focused representation and a contrast-specific embedding, each employed independently for demographic prediction. Through multimodal MRI modeling and cross-dataset validation, the study provides the first quantitative assessment of the distinct contributions of these signals: the anatomical representation retains nearly all predictive performance of the original image, whereas the contrast embedding carries only weak and non-generalizable signals. These findings demonstrate that demographic information in MRI is predominantly rooted in authentic anatomical differences, offering both theoretical grounding and a practical pathway for robust debiasing.

Technology Category

• Machine Learning: Representation Learning
• Knowledge Representation and Reasoning: Preferences
• Computer Vision: Bias, Fairness & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Demographic attributes such as age, sex, and race can be predicted from medical images, raising concerns about bias in clinical AI systems. In brain MRI, this signal may arise from anatomical variation, acquisition-dependent contrast differences, or both, yet these sources remain entangled in conventional analyses. Without disentangling them, mitigation strategies risk failing to address the underlying causes. We propose a controlled framework based on disentangled representation learning, decomposing brain MRI into anatomy-focused representations that suppress acquisition influence and contrast embeddings that capture acquisition-dependent characteristics. Training predictive models for age, sex, and race on full images, anatomical representations, and contrast-only embeddings allows us to quantify the relative contributions of structure and acquisition to the demographic signal. Across three datasets and multiple MRI sequences, we find that demographic predictability is primarily rooted in anatomical variation: anatomy-focused representations largely preserve the performance of models trained on raw images. Contrast-only embeddings retain a weaker but systematic signal that is dataset-specific and does not generalise across sites. These findings suggest that effective mitigation must explicitly account for the distinct anatomical and acquisition-dependent origins of the demographic signal, ensuring that any bias reduction generalizes robustly across domains.

Problem

• demographic bias
• brain MRI
• anatomical variation
• acquisition contrast
• predictability

Innovation

• disentangled representation learning
• anatomy-contrast disentanglement
• demographic bias
• MRI acquisition effects
• generalizable bias mitigation