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

2026-04-08 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Medical Devices & Health Technology · Depth: Expert, medium

Summary

A new intelligent MRI framework, PASS (Personalized, Anomaly-aware Sampling and reconStruction), addresses the long acquisition times and lack of clinical task adaptability in traditional MRI. Introduced on April 8, 2026, PASS integrates a Vision-Language Model (VLM) to guide a deep unrolling network, enabling personalized, fast imaging. The framework features a deep unrolled reconstruction network based on a physics-based MRI model, a sampling module for patient-specific k-space trajectories, and an anomaly-aware prior derived from a pretrained VLM. This VLM prior directs both sampling and reconstruction towards clinically relevant regions. PASS achieves superior image quality across diverse anatomies, contrasts, anomalies, and acceleration factors, directly enhancing downstream diagnostic tasks like fine-grained anomaly detection, localization, and diagnosis.

Key takeaway

For Machine Learning Engineers developing medical imaging solutions, PASS offers a compelling approach to overcome generic MRI limitations. You should explore integrating Vision-Language Models with physics-aware deep unrolling networks to create personalized, anomaly-aware imaging pipelines. This method can significantly improve image quality and diagnostic utility for specific clinical tasks, reducing acquisition times and enhancing downstream analysis.

Key insights

PASS uses a VLM to guide personalized MRI sampling and reconstruction for faster, anomaly-aware imaging.

Principles

• Integrate VLM for clinical reasoning
• Derive reconstruction from physics models
• Personalize k-space trajectories

Method

PASS employs a deep unrolled reconstruction network, a patient-specific k-space sampling module, and an anomaly-aware prior from a VLM to guide both sampling and reconstruction for task-oriented fast MRI.

In practice

• Improve anomaly detection in MRI
• Accelerate MRI acquisition times
• Enhance diagnostic accuracy

Topics

• Personalized MRI
• Vision-Language Models
• Deep Unrolling Networks
• k-space Sampling
• Anomaly Detection

Code references

• mkhateri/Awesome-MRI-Super-Resolution
• deepseek-ai/DeepSeek-VL2

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Attention-Gated Convolutional Networks for Scanner-Agnostic Quality Assessment — A hybrid CNN-Attention model effectively assesses MRI quality, generalizing across diverse scanners without retraining.
• Scalable Training of Spatially Grounded 2D Vision-Language Models for Radiology — Spatially grounded radiology VLMs can be trained at scale without manual annotations, maintaining language quality.
• Physics-Driven Zero-Shot MRI Reconstruction with Non-local Image Priors — Combining physics-driven constraints with non-local image priors enhances zero-shot MRI reconstruction stability and performance.
• Looking Where the Eye Cannot See: Improving MS Diagnosis with Deep Learning — DeepMS improves MS diagnosis by analyzing subtle microstructural changes in normal-appearing white matter using deep learning.
• Dynamic breast MRI with Flexible Temporal Resolution Aided by Deep Learning — Combining locally low-rank subspace modeling with deep learning enables flexible temporal resolution in DCE-MRI.
• Unified Multimodal Model for Brain MRI Imputation and Understanding — UniBrain unifies MRI imputation and understanding, integrating multimodal data and self-alignment for robust medical reasoning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.