Vision-Language Model-Guided Deep Unrolling Enables Personalized, Fast MRI

2026-04-08

AI Summary

Traditional accelerated MRI methods struggle to simultaneously achieve high imaging speed and clinically relevant image quality, often lacking personalization capabilities. This work proposes the PASS framework, which, for the first time, integrates high-level semantic priors from pretrained vision-language models into the entire MRI sampling and reconstruction pipeline. By generating patient-specific k-space trajectories and performing anomaly-aware reconstruction tailored to particular clinical tasks, PASS enables rapid, task-oriented imaging. The approach synergistically combines physics-model-driven deep unfolding networks with vision-language-guided semantic priors, substantially improving image quality across diverse anatomies, contrast types, pathology categories, and acceleration factors. Consequently, it enhances downstream performance in fine-grained anomaly detection, localization, and diagnosis.

Abstract

Magnetic Resonance Imaging (MRI) is a cornerstone in medicine and healthcare but suffers from long acquisition times. Traditional accelerated MRI methods optimize for generic image quality, lacking adaptability for specific clinical tasks. To address this, we introduce PASS (Personalized, Anomaly-aware Sampling and reconStruction), an intelligent MRI framework that leverages a Vision-Language Model (VLM) to guide a deep unrolling network for task-oriented, fast imaging. PASS dynamically personalizes the imaging pipeline through three core contributions: (1) a deep unrolled reconstruction network derived from a physics-based MRI model; (2) a sampling module that generates patient-specific $k$-space trajectories; and (3) an anomaly-aware prior, extracted from a pretrained VLM, which steers both sampling and reconstruction toward clinically relevant regions. By integrating the high-level clinical reasoning of a VLM with an interpretable, physics-aware network, PASS achieves superior image quality across diverse anatomies, contrasts, anomalies, and acceleration factors. This enhancement directly translates to improvements in downstream diagnostic tasks, including fine-grained anomaly detection, localization, and diagnosis.

Problem

• MRI acceleration
• personalized imaging
• task-oriented reconstruction
• clinical adaptability
• long acquisition time

Innovation

• Vision-Language Model
• Deep Unrolling
• Personalized MRI
• Anomaly-aware Prior
• k-space Sampling