Accelerated 3D MR neurography of the brachial plexus using deep learning-constrained compressed sensing

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• 3D MRI of the brachial plexus is crucial for diagnosing nerve injuries.
• Improving image quality and reducing scan time are key challenges in brachial plexus MRI.

Purpose of the Study:

• To evaluate the efficacy of deep learning-constrained compressed sensing (DLCS) in enhancing 3D brachial plexus MRI.
• To compare DLCS with conventional SENSE and CS techniques in terms of image quality and acquisition speed.

Main Methods:

• Fifty-four contrast-enhanced and forty-one unenhanced 3D MRI scans of the brachial plexus were analyzed.
• Acquisitions included SENSE with 4x acceleration (SENSE4x), CS with 4x acceleration (CS4x), and DLCS with 4x (DLCS4x) and 8x (DLCS8x) acceleration.
• Image quality was assessed using apparent signal-to-noise ratio (aSNR), apparent contrast-to-noise ratio (aCNR), and qualitative scores.

Main Results:

• DLCS4x demonstrated superior aSNR and aCNR compared to SENSE4x, CS4x, and DLCS8x (p < 0.05).
• DLCS4x achieved higher qualitative scores for brachial plexus trunks, divisions, cords, and branches than SENSE4x and CS4x (p < 0.05).
• DLCS8x maintained image quality comparable to SENSE4x and CS4x but with approximately half the acquisition time.

Conclusions:

• DLCS significantly improves image quality and/or accelerates acquisition for 3D brachial plexus MRI.
• DLCS offers a promising tool for enhanced clinical evaluation of the brachial plexus and its intricate network of nerves.
• The findings suggest DLCS is a valuable advancement for clinical practice in brachial plexus imaging.