Deep learning of personalized priors from past MRI scans enables fast, quality-enhanced point-of-care MRI with low-cost systems

2025-05-05

AI Summary

Low-field MRI suffers from prolonged scan times and poor image quality—characterized by low signal-to-noise ratio (SNR) and weak tissue contrast—posing a critical clinical bottleneck. Method: This work proposes a personalized prior-driven accelerated reconstruction framework that leverages anatomical and pathology-specific priors extracted from a patient’s prior high-field MRI scans to enable rapid, high-fidelity follow-up imaging on cost-effective low- or ultra-low-field systems. Contribution/Results: We introduce ViT-Fuser, a novel vision transformer-based feature fusion architecture, which—uniquely—enables single-shot generalization of priors across vendors, field strengths, and acquisition sequences, establishing a “one-prior-for-life” paradigm. Evaluated on glioblastoma data and real-world 50 mT and 6.5 mT low-field scanners, our method achieves ≥4× higher acceleration than state-of-the-art methods while significantly improving SNR and tissue contrast, with strong robustness to out-of-distribution data.

Technology Category

• Computer Vision: Multi-modal Vision
• Search and Optimization: Learning to Search
• Machine Learning: Multi-instance/Multi-view Learning

Application Category

• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Security and Privacy: Privacy-enhancing technologies

Abstract

Magnetic resonance imaging (MRI) offers superb-quality images, but its accessibility is limited by high costs, posing challenges for patients requiring longitudinal care. Low-field MRI provides affordable imaging with low-cost devices but is hindered by long scans and degraded image quality, including low signal-to-noise ratio (SNR) and tissue contrast. We propose a novel healthcare paradigm: using deep learning to extract personalized features from past standard high-field MRI scans and harnessing them to enable accelerated, enhanced-quality follow-up scans with low-cost systems. To overcome the SNR and contrast differences, we introduce ViT-Fuser, a feature-fusion vision transformer that learns features from past scans, e.g. those stored in standard DICOM CDs. We show that extit{a single prior scan is sufficient}, and this scan can come from various MRI vendors, field strengths, and pulse sequences. Experiments with four datasets, including glioblastoma data, low-field ($50mT$), and ultra-low-field ($6.5mT$) data, demonstrate that ViT-Fuser outperforms state-of-the-art methods, providing enhanced-quality images from accelerated low-field scans, with robustness to out-of-distribution data. Our freely available framework thus enables rapid, diagnostic-quality, low-cost imaging for wide healthcare applications.

Problem

• Enhancing low-field MRI image quality using deep learning
• Reducing scan time and cost for longitudinal MRI care
• Overcoming SNR and contrast limitations in affordable MRI systems

Innovation

• Deep learning extracts personalized MRI features
• ViT-Fuser fuses features for enhanced low-field scans
• Single prior scan from any MRI system suffices