Enhanced Deep-Learning-Based Magnetic Resonance Image Reconstruction by Leveraging Prior Subject-Specific Brain Imaging: Proof-of-Concept Using a Cohort of Presumed Normal Subjects

• Published in: IEEE journal of selected topics in signal processing Vol. 14; no. 6; pp. 1126 - 1136
• Main Authors: Souza, Roberto, Beauferris, Youssef, Loos, Wallace, Lebel, Robert Marc, Frayne, Richard
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.10.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Archiving; Brain; Communications systems; Compressed sensing; Deep learning; Image acquisition; Image enhancement; Image quality; Image reconstruction; longitudinal information; Magnetic resonance imaging; Magnetic resonance imaging (MRI); Medical imaging; picture archiving and communication system (PACS); Picture archiving and communication systems; Rank tests; Signal to noise ratio; Volumetric analysis
• ISSN: 1932-4553; 1941-0484
• DOI: 10.1109/JSTSP.2020.3001525

Deep learning models have shown potential for reconstructing undersampled, multi-channel magnetic resonance (MR) image acquisitions. Recently proposed methods, however, have not leveraged information from prior subject-specific MR imaging sessions. Such data are often readily available through a picture archiving and communication system (PACS). We propose a flexible three-step method to incorporate this prior information into an enhanced deep-learning-based reconstruction process. The method consists of Step 1: an initial reconstruction; Step 2: registration of the previous scan to the initial reconstruction; and Step 3: an enhancement network. Training and testing used longitudinally acquired, three-dimensional, T1-weighted brain images acquired with different acquisition parameters. We tested our networks using data from <inline-formula><tex-math notation="LaTeX">\mathbf {2808}</tex-math></inline-formula> images (obtained in 18 subjects) under four different acceleration factors (<inline-formula><tex-math notation="LaTeX">\mathbf {R=\lbrace 5,10,15,20\rbrace }</tex-math></inline-formula>). Our enhanced reconstruction (Steps 1-3) produced higher-quality images: structural similarity and peak signal-to-noise ratio increased, and normalized root mean squared error decreased on average by <inline-formula><tex-math notation="LaTeX">\mathbf {16.5\%}</tex-math></inline-formula>, <inline-formula><tex-math notation="LaTeX">\mathbf {7.0\%}</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">\mathbf {21.1\%}</tex-math></inline-formula>, respectively, compared to the non-enhanced reconstruction (Step 1 only) under the same network capacity as the enhanced reconstruction model. These differences were statistically significant (<inline-formula><tex-math notation="LaTeX">\boldsymbol{p< 0.001}</tex-math></inline-formula>, Wilcoxon signed-rank test). Further volumetric analysis performed on key brain regions (brain, white matter, gray matter and cortex) indicated that our enhanced images had better volume agreement with the fully sampled reference images compared to the non-enhanced images. Our enhanced images for <inline-formula><tex-math notation="LaTeX">\mathbf {R=20}</tex-math></inline-formula> were comparable to the non-enhanced images for <inline-formula><tex-math notation="LaTeX">\mathbf {R=10}</tex-math></inline-formula> demonstrating that our proposed method can use prior scan information to further accelerate MR examinations.