AI-based motion artifact severity estimation in undersampled MRI allowing for selection of appropriate reconstruction...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Magnetic Resonance Imaging (MRI) acquisition is lengthy, increasing susceptibility to patient motion.
• Even minor motion (1mm) can cause severe artifacts, potentially requiring repeat scans.
• Accelerated MRI techniques use partial k-space acquisition and advanced reconstruction, but AI methods often assume no motion.

Purpose of the Study:

• To develop a method for retrospective detection and quantification of motion artifacts in undersampled MRI data.
• To enhance AI-based MRI reconstruction safety and provide data for reconstruction optimization.
• To enable real-time artifact detection for potential re-acquisition prompts.

Main Methods:

• A deep learning approach using a convolutional neural network (CNN) was developed to detect and quantify motion in undersampled brain MRI.
• The CNN was trained using synthetically motion-corrupted data, demonstrating generalization to real-world data.
• The motion artifact estimator was integrated as a selector for motion-robust or data-consistent reconstruction models.

Main Results:

• The model achieved 91-96% accuracy in distinguishing motion from motion-free scans.
• It correctly predicted manual quality labels (Good, Medium, Bad) with 76-85% accuracy.
• As a selector, the model chose the appropriate reconstruction network 93% of the time, yielding near-optimal SSIM values.

Conclusions:

• The developed method accurately quantifies motion artifact severity in undersampled MRI.
• Enables real-time motion artifact detection, crucial for improving AI-based MRI reconstruction safety and quality.