Accelerated Dual-contrast Three-dimensional Knee Magnetic Resonance Imaging Using Super-resolution Reconstructed Deep Learning-enhanced Two-dimensional Dixon Turbo Spin-echo Imaging

Purpose or Learning Objective: Recent work on three-dimensional (3D) super-resolution reconstruction (SRR) of conventional two-dimensional (2D) turbo spin-echo (TSE) knee magnetic resonance imaging (MRI) shows that high-resolution isotropic knee MRI is technically feasible. Yet the use of acceleration techniques and contrast optimization are needed for clinical validation of this 3D method. With the advent of deep learning (DL) image reconstruction techniques, high acceleration of SRR input data is now achievable through extended use of simultaneous multislice and parallel imaging methods. Moreover, the combination of accelerated TSE with the Dixon method allows us to acquire fat-suppressed and non-fat-suppressed data within a single acquisition that further accelerates the dual-contrast SRR protocol.

This study evaluated the technical feasibility of 3D SRR MRI based on DL-enhanced highly accelerated 2D Dixon TSE MRI. It also compared image quality and diagnostic confidence of this dual-contrast 3D technique to (accelerated) conventional 2D TSE knee MRI. We hypothesized that the SRR method provides the required contrasts needed for comprehensive knee joint evaluation in a competitive acquisition time.

Methods or Background: Five patient-volunteers (three men; mean age: 35.6 years; mean body mass index: 27.8 kg/m²) who were referred for clinical knee MRI were scanned on a 3-T MR scanner (MAGNETOM Vida XT, Siemens Healthcare) with an 18-channel knee coil after submitting written informed consent.

Each patient was scanned with (1) a standard clinical 2D TSE protocol (i.e., axial and coronal proton-density (PD)-weighted fat suppressed (FS), coronal T1-weighted, and sagittal PD-weighted and T2-weighted FS; total acquisition time: 11 min, 23 s), (2) a DL-enhanced accelerated 2D TSE protocol (same contrasts as for [1]; total acquisition time: 6 min, 24 s), and (3) an SRR protocol based on DL-enhanced accelerated 2D Dixon TSE MRI (i.e., 6 PD-weighted repetitions rotated around the phase-encoding axis, uniformly distributed over 180 degrees; total acquisition time: 6 min, 24 s).

Dixon data sets with acquired in-phase and reconstructed water-only contrasts were reconstructed to an isotropic voxel grid of 0.35 × 0.35 × 0.35 m³ using a scalar SRR method with joint rigid motion estimation. Two certified radiologists evaluated all data sets regarding image quality and diagnostic confidence for the assessment of various knee structures.

Results or Findings: No significant differences in image quality or diagnostic confidence were found when comparing the conventional and highly accelerated 2D TSE data sets. The SRRs of in-phase and water-only Dixon data sets provided good to adequate image quality and the necessary contrasts for acceptable diagnostic confidence.

Conclusion: Dual-contrast 3D SRR knee MRI based on DL-enhanced accelerated 2D Dixon TSE imaging is technically feasible and provides the necessary image quality and diagnostic confidence while requiring nearly 50% less scan time compared with conventional 2D knee MRI.

Publication History

Article published online:
26 May 2023