CoRRECT: A Deep Unfolding Framework for Motion-Corrected Quantitative R2* Mapping

Area of Science:

• Medical Imaging
• Biophysics
• Artificial Intelligence

Background:

• Quantitative MRI (qMRI) quantifies biological tissue parameters but faces challenges with artifacts.
• Traditional qMRI methods address artifacts like motion and magnetic field inhomogeneities separately, limiting performance.
• Accelerated data acquisition in qMRI exacerbates artifact issues, necessitating advanced solutions.

Purpose of the Study:

• To present CoRRECT, a unified deep unfolding framework for artifact reduction in qMRI.
• To develop a model-based neural network that integrates motion and field inhomogeneity correction.
• To enable high-quality qMRI with accelerated acquisition without pre-computed correction parameters.

Main Methods:

• Developed a unified deep unfolding (DU) framework named CoRRECT.
• Implemented a model-based, end-to-end neural network trained with self-supervised learning.
• The network learns to correct for motion and field inhomogeneities directly from k-space data.

Main Results:

• CoRRECT successfully recovers artifact-free R2* maps from accelerated multi-gradient recalled echo (mGRE) MRI data.
• The framework accounts for motion and field inhomogeneities without requiring pre-computed correction parameters.
• Demonstrated robust performance in highly accelerated acquisition settings.

Conclusions:

• CoRRECT offers a unified approach to artifact correction in qMRI, improving image quality.
• Deep unfolding methods can integrate physical, biophysical, and learned models for advanced qMRI.
• This work paves the way for more efficient and accurate quantitative MRI techniques.