Accelerated 7T 3D DESS MR Neurography of the Cervical Spine Using CS-CAIPIRINHA and Deep Learning Reconstruction

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• High-resolution 7T MRI of cervical nerves and rootlets is crucial for diagnosing neurological conditions.
• Conventional undersampling techniques like GRAPPA can limit acceleration factors, impacting scan times.
• Deep learning (DL) reconstruction shows promise for improving MRI quality and speed.

Purpose of the Study:

• To evaluate the efficacy of combined coherent and incoherent undersampling with DL reconstruction for accelerated 7T cervical spine MRI.
• To assess image quality, artifacts, and sharpness at various acceleration factors.
• To determine the potential for significant scan time reduction without compromising diagnostic information.

Main Methods:

• Prospective study involving asymptomatic volunteers undergoing 7T cervical spine MRI.
• Comparison of GRAPPA undersampling (reference) with DL-based reconstructions using CAIPIRINHA (coherent) and compressed sensing (CS) (incoherent) undersampling.
• Evaluation of image quality, noise, artifacts, and sharpness by musculoskeletal radiologists using Likert scales.
• Statistical analysis including Friedman and Wilcoxon signed-rank tests.

Main Results:

• DL-based reconstructions significantly improved image quality and reduced noise compared to GRAPPA (P<0.001).
• Combined CAIPIRINHA and CS undersampling with DL allowed for acceleration factors up to 16 with preserved image quality and fewer artifacts.
• Sharpness of neural structures was highest with 4-fold DL-CAIPIRINHA.
• Scan times were reduced by over 6-fold with DL reconstructions.

Conclusions:

• Combined coherent and incoherent undersampling with DL reconstruction offers a robust framework for high-resolution 7T cervical spine MRI.
• This approach enables substantial acceleration (beyond R=8) with preserved image quality and reduced scan times.
• The technique facilitates efficient, high-resolution imaging of cervical nerves and rootlets, exceeding 6-fold scan time reductions.