AgentMRI: A Vison Language Model-Powered AI System for Self-regulating MRI Reconstruction with Multiple Degradations
Gulfam Ahmed Sajua1, Marjan Akhib2, Yuchou Chang3
1Department of Computer and Information Science, University of Massachusetts Dartmouth, North Dartmouth, Dartmouth, 02747, MA, USA.
Journal of Imaging Informatics in Medicine
|July 22, 2025
Summary
AgentMRI, an AI system using vision language models, autonomously reconstructs magnetic resonance imaging (MRI) scans. It detects and corrects image degradations without human input, achieving high accuracy in tests.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Artificial intelligence (AI)-driven autonomous agents are revolutionizing various fields by unifying reasoning, decision-making, and task execution.
- In medical imaging, AI agents offer potential workflow improvements by minimizing human intervention and enhancing image quality.
- Current magnetic resonance imaging (MRI) reconstruction methods often require manual post-processing or rely on static correction models.
Purpose of the Study:
- To introduce AgentMRI, an AI-driven system utilizing vision language models (VLMs) for fully autonomous MRI reconstruction.
- To develop a system capable of dynamically detecting MRI corruption and selecting appropriate correction models without manual intervention.
- To establish a scalable and multimodal AI framework for autonomous MRI processing.
Main Methods:
- AgentMRI employs a multi-query VLM strategy for robust, consensus-based corruption detection and confidence-weighted inference.
- The system automatically selects appropriate deep learning models for MRI reconstruction, motion correction, and denoising.
- Evaluation was conducted using a comprehensive brain MRI dataset in both zero-shot and fine-tuned settings.
Main Results:
- AgentMRI achieved 73.6% accuracy in zero-shot settings and 95.1% accuracy in fine-tuned settings for MRI reconstruction.
- Experimental results demonstrate the system's ability to execute the reconstruction process accurately without human intervention.
- The framework successfully eliminated the need for manual intervention in MRI post-processing.
Conclusions:
- AgentMRI represents a significant advancement toward fully autonomous and intelligent MR image reconstruction systems.
- The system offers a scalable and multimodal AI framework for autonomous MRI processing, reducing reliance on human input.
- This AI-driven approach has the potential to transform medical imaging workflows by optimizing efficiency and image quality.
