Noise behavior of MR brain reconstructions using compressed sensing

Applicability of deep learning-based reconstruction trained by brain and knee 3T MRI to lumbar 1.5T MRI

Background

Several deep learning-based methods have been proposed for addressing the long scanning time of magnetic resonance imaging. Most are trained using brain 3T magnetic resonance images, but is unclear whether performance is affected when applying these methods to different anatomical sites and at different field strengths.

Purpose

To validate the denoising performance of deep learning-based reconstruction method trained by brain and knee 3T magnetic resonance images when applied to lumbar 1.5T magnetic resonance images.

Material and Methods

Using a 1.5T scanner, we obtained lumber T2-weighted sequences in 10 volunteers using three different scanning times: 228 s (standard), 119 s (double-fast), and 68 s (triple-fast). We compared the images obtained by the standard sequence with those obtained by the deep learning-based reconstruction-applied faster sequences.

Results

Signal-to-noise ratio values were significantly higher for deep learning-based reconstruction-double-fast than for standard and did not differ significantly between deep learning-based reconstruction-triple-fast and standard. Contrast-to-noise ratio values also did not differ significantly between deep learning-based reconstruction-triple-fast and standard. Qualitative scores for perceived signal-to-noise ratio and overall image quality were significantly higher for deep learning-based reconstruction-double fast and deep learning-based reconstruction-triple-fast than for standard. Average scores for sharpness, contrast, and structure visibility were equal to or higher for deep learning-based reconstruction-double-fast and deep learning-based reconstruction-triple-fast than for standard, but the differences were not statistically significant. The average scores for artifact were lower for deep learning-based reconstruction-double-fast and deep learning-based reconstruction-triple-fast than for standard, but the differences were not statistically significant.

Conclusion

The deep learning-based reconstruction method trained by 3T brain and knee images may reduce the scanning time of 1.5T lumbar magnetic resonance images by one-third without sacrificing image quality.

Knee imaging: Rapid three-dimensional fast spin-echo using compressed sensing

PURPOSE

To investigate the feasibility of using compressed sensing (CS) to accelerate three-dimensional fast spin-echo (3D-FSE) imaging of the knee.

MATERIALS AND METHODS

A 3D-FSE sequence was performed at 3T with CS (CUBE-CS with 3:16-minute scan time) and without CS (CUBE with 4:44-minute scan time) twice on the knees of 10 healthy volunteers to assess signal-to-noise ratio (SNR) using the addition-subtraction method and once on the knees of 50 symptomatic patients to assess diagnostic performance. SNR of cartilage, muscle, synovial fluid, and bone marrow on CUBE and CUBE-CS images were measured in the 10 healthy volunteers. The CUBE and CUBE-CS sequences of all 50 symptomatic patients were independently reviewed twice by two musculoskeletal radiologists. The radiologists used CUBE and CUBE-CS during each individual review to determine the presence or absence of knee joint pathology. Student's t-tests were used to compare SNR values between sequences, while the kappa statistic was used to determine agreement between sequences for detecting knee joint pathology. Sensitivity and specificity of CUBE and CUBE-CS for detecting knee joint pathology was also calculated in the 18 symptomatic patients who underwent subsequent arthroscopic knee surgery.

RESULTS

CUBE and CUBE-CS had similar SNR (P = 0.15-0.67) of cartilage, muscle, synovial fluid, and bone marrow. There was near-perfect to perfect agreement between CUBE and CUBE-CS for both radiologists for detecting cartilage and bone marrow edema lesions, medial and lateral meniscus tears, anterior cruciate ligament tears, effusions, and intra-articular bodies. CUBE and CUBE-CS had similar sensitivity (75.0-100%) and specificity (87.5-100%) for detecting 60 cartilage lesions, 20 meniscus tears, four anterior cruciate ligament tears, and four intra-articular bodies confirmed at surgery.

CONCLUSION

CS provided a 30% reduction in scan time for 3D-FSE imaging of the knee without a corresponding decrease in SNR or diagnostic performance.

LEVEL OF EVIDENCE

1 J. MAGN. RESON. IMAGING 2017;45:1712-1722.