A physics-informed foundation model for quantitative diffusion MRI

Researchers have developed PIGMENT, a physics-informed AI foundation model that dramatically improves diffusion MRI brain imaging by learning universal tissue patterns and adapting them to individual scans. The model enables reliable quantitative brain mapping from sparse, heterogeneous data across multiple imaging systems, extending capabilities to low-field and clinical settings previously unsuitable for detailed analysis.

PIGMENT represents a significant advancement in medical imaging by applying foundation model principles to a domain where data heterogeneity and acquisition constraints have historically limited clinical utility. The model was trained on 11,375 scans across multiple institutions and imaging parameters, allowing it to extract meaningful tissue microstructure information from acquisitions that conventional methods cannot reliably analyze. This approach bridges the gap between research-grade imaging protocols and real-world clinical settings where standardized acquisition conditions are impractical.

The development addresses a fundamental challenge in neuroimaging: diffusion MRI provides unique insights into brain tissue architecture, but extracting quantitative measurements has required dense sampling and optimized protocols accessible primarily in specialized research environments. By learning a generalizable prior of human brain microstructure, PIGMENT enables zero-shot adaptation to individual participant data while maintaining biological validity. Validation across five independent external centers demonstrates the model's robustness across different equipment vendors and field strengths.

The practical implications extend across multiple medical applications. PIGMENT's ability to extract reliable biomarkers from ultra-fast clinical protocols could accelerate tumor assessment and monitoring. For pediatric neuroscience, the model preserves developmental trajectories even from 10-fold accelerated scans, enabling longitudinal studies with reduced acquisition burden. The extension to low-field MRI systems has significant global health implications, as these cost-efficient devices are increasingly deployed in resource-limited settings. The framework establishes a template for physics-informed foundation models in medical imaging, suggesting broader applicability to other quantitative imaging modalities that face similar heterogeneity and protocol constraints.

• →Physics-informed foundation models successfully extend quantitative brain MRI to sparse and heterogeneous clinical data unsuitable for conventional analysis
• →PIGMENT validates across five independent medical centers, demonstrating robust generalization across equipment vendors and field strengths
• →The model enables reliable brain mapping on cost-efficient low-field MRI systems, expanding neuroimaging access to resource-limited settings
• →Ultra-fast acquisition protocols become viable for tumor biomarker extraction and developmental tracking without sacrificing measurement reliability
• →The architecture provides a replicable template for applying foundation models to physics-based medical imaging problems