Dynamic MRI using model‐based deep learning and SToRM priors: MoDL‐SToRM

• Published in: Magnetic resonance in medicine Vol. 82; no. 1; pp. 485 - 494
• Main Authors: Biswas, Sampurna, Aggarwal, Hemant K., Jacob, Mathews
• Format: Journal Article
• Language: English
• Published: United States Wiley Subscription Services, Inc 01.07.2019
• Subjects: Algorithms; alternating minimization; Artificial neural networks; Breathing; Cardiac Imaging Techniques - methods; Conjugate gradient method; Consistency; Data recovery; Databases, Factual; Deep Learning; free breathing cardiac MR; Heart; Heart - diagnostic imaging; Humans; Image acquisition; Image Processing, Computer-Assisted - methods; Image quality; Image reconstruction; Iterative algorithms; Iterative methods; learned prior; Machine learning; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Manifolds; Measurement methods; model‐based; Neural networks; Noise reduction; non‐local prior; Optimization; Regularization; Respiration; Smoothness; subject specific prior; free breathing cardiac MR; non-local prior; learned prior; subject specific prior; alternating minimization; model-based
• ISSN: 0740-3194; 1522-2594; 1522-2594
• DOI: 10.1002/mrm.27706

Purpose To introduce a novel framework to combine deep‐learned priors along with complementary image regularization penalties to reconstruct free breathing & ungated cardiac MRI data from highly undersampled multi‐channel measurements. Methods Image recovery is formulated as an optimization problem, where the cost function is the sum of data consistency term, convolutional neural network (CNN) denoising prior, and SmooThness regularization on manifolds (SToRM) prior that exploits the manifold structure of images in the dataset. An iterative algorithm, which alternates between denoizing of the image data using CNN and SToRM, and conjugate gradients (CG) step that minimizes the data consistency cost is introduced. Unrolling the iterative algorithm yields a deep network, which is trained using exemplar data. Results The experimental results demonstrate that the proposed framework can offer fast recovery of free breathing and ungated cardiac MRI data from less than 8.2s of acquisition time per slice. The reconstructions are comparable in image quality to SToRM reconstructions from 42s of acquisition time, offering a fivefold reduction in scan time. Conclusions The results show the benefit in combining deep learned CNN priors with complementary image regularization penalties. Specifically, this work demonstrates the benefit in combining the CNN prior that exploits local and population generalizable redundancies together with SToRM, which capitalizes on patient‐specific information including cardiac and respiratory patterns. The synergistic combination is facilitated by the proposed framework.