J-MoDL

J-MoDL jointly optimizes MRI sampling patterns and deep-learning reconstruction parameters to reconstruct undersampled multichannel Fourier MRI data and improve image fidelity.

Key Features:

• Continuous Optimization Strategy: Joint continuous optimization of sampling locations and reconstruction network parameters.
• Multichannel Forward Model: Uses a multichannel forward model with a non-uniform Fourier transform defined at continuous sampling locations to enforce data consistency.
• Model-Based Deep Learning Scheme: Embeds the forward model into a deep learning architecture using convolutional neural networks (CNNs) to impose physics-based constraints during reconstruction.
• Improved Image Quality: Joint optimization of sampling and network parameters enhances deep-learning reconstruction performance on undersampled data.

Scientific Applications:

• Accelerated MRI acquisition: Enables reduction of scan times by optimizing undersampled multichannel Fourier sampling patterns while maintaining reconstructed image fidelity.
• Clinical MRI workflows: Supports faster acquisitions that can improve patient comfort and scanner throughput through optimized sampling and reconstruction.

Methodology:

Joint continuous optimization of sampling locations and convolutional neural network parameters within a model-based deep learning framework; the method embeds a multichannel forward model with a non-uniform Fourier transform at continuous sampling locations to enforce data consistency and uses compressed sensing principles to reconstruct undersampled multichannel Fourier MRI measurements.