Evaluating Undersampling Schemes and Deep Learning Reconstructions for High-Resolution 3D Double Echo Steady State...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Radiology
• Biomedical Engineering

Background:

• 3D DESS MRI offers high isotropic resolution for knee cartilage imaging, especially at high field strengths.
• Accelerated acquisition techniques are crucial for clinical feasibility, but optimal undersampling schemes remain undefined.

Purpose of the Study:

• To evaluate the optimal undersampling scheme and its limits for 3D DESS MRI of knee cartilage.
• To compare incoherent (compressed sensing) and regular (GRAPPA, CAIPIRINHA) undersampling patterns with deep learning reconstruction.

Main Methods:

• Acquired 7 T 3D DESS images of 40 knees with varying undersampling factors (R=4-30) and schemes (GRAPPA, CAIPIRINHA, CS).
• Reconstructed images using deep learning (DL) algorithms.
• Assessed image quality and interreader agreement for cartilage lesions.

Main Results:

• 16-fold accelerated CS images were comparable in quality to 4-fold GRAPPA and 8-fold CAIPIRINHA but acquired significantly faster.
• Interreader agreement for cartilage lesions was almost perfect across tested acceleration factors and schemes.

Conclusions:

• Incoherent undersampling (CS) combined with DL reconstruction offers significant advantages for accelerated 3D DESS cartilage imaging.
• This approach enables fast, high-resolution acquisitions without compromising image quality, suitable for musculoskeletal assessment at 7 T.