Compressive MRI quantification using convex spatiotemporal priors and deep encoder-decoder networks.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Computational Imaging

Background:

• Quantitative MRI (qMRI) enables precise tissue property measurement.
• Traditional methods struggle with artifacts from accelerated scan times.
• Dictionary-matching methods are computationally intensive and lack scalability.

Purpose of the Study:

• To introduce a dictionary-matching-free pipeline for multi-parametric quantitative MRI image computing.
• To improve the accuracy and consistency of quantitative information recovery from undersampled MRI data.
• To replace computationally expensive baseline methods with an efficient deep learning approach.

Main Methods:

• A two-stage pipeline involving compressed sensing reconstruction and deep learned quantitative inference.
• Convex reconstruction using efficient spatiotemporal regularizations and accelerated iterative shrinkage to minimize undersampling artifacts.
• A deep encoder-decoder network trained on Bloch equations simulations for quantitative inference, utilizing multi-scale piecewise affine approximations.

Main Results:

• The proposed scheme effectively minimizes aliasing artifacts caused by short scan times.
• Accurate and consistent quantitative information is recovered from novel, aggressively subsampled 2D/3D qMRI protocols.
• The deep learning approach demonstrates superior performance compared to traditional dictionary-matching baselines.

Conclusions:

• The developed pipeline offers an effective solution for artifact reduction in accelerated quantitative MRI.
• This dictionary-free, deep learning-based approach enhances the reliability of quantitative MRI analysis.
• The method shows significant potential for improving the efficiency and accuracy of qMRI acquisition and processing.