JSENSE-Pro: Joint sensitivity estimation and image reconstruction in parallel imaging using pre-learned subspaces of...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Image Reconstruction
• Subspace Modeling

Background:

• Parallel imaging accelerates MRI acquisition but often requires calibration scans.
• Calibrationless methods aim to eliminate this acquisition step.
• Coil sensitivity functions are crucial for parallel imaging reconstruction.

Purpose of the Study:

• To develop a calibrationless parallel imaging method using pre-learned coil sensitivity function subspaces.
• To improve the accuracy of coil sensitivity estimation and image reconstruction without calibration data.

Main Methods:

• A subspace-based joint sensitivity estimation and image reconstruction method was developed.
• A probabilistic subspace model captured prior information of coil sensitivity functions from training data.
• The model was integrated into a regularized reconstruction framework solved via alternating minimization.

Main Results:

• The proposed method (JSENSE-Pro) achieved superior coil sensitivity estimation and image reconstruction quality compared to state-of-the-art methods at all tested acceleration factors.
• Performance was comparable or better than methods using calibration data.
• The model generalized from T2-weighted to FLAIR data reconstruction.

Conclusions:

• JSENSE-Pro offers an effective calibrationless approach for accelerated parallel imaging.
• The method enhances the practical utility of parallel imaging, particularly when calibration data acquisition is challenging.