Deep-learning-reconstructed high-resolution 3D cervical spine MRI for foraminal stenosis evaluation

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Standard two-dimensional (2D) MRI is the current standard for cervical spine imaging.
• Assessing cervical spine stenosis, particularly foraminal stenosis, can be limited by image quality and motion artifacts in 2D acquisitions.
• Deep learning-based reconstruction algorithms offer potential for improving MRI image quality.

Purpose of the Study:

• To compare the diagnostic performance of standard 2D cervical spine MRI with a novel 3D MRI acquisition reconstructed using a deep learning algorithm.
• To evaluate if the enhanced image quality from deep learning reconstruction improves inter-rater agreement for cervical spine stenosis assessment.
• To assess differences in motion artifact and scan time between the two techniques.

Main Methods:

• Forty-one patients underwent both conventional 2D and deep learning-reconstructed 3D cervical spine MRI.
• Three radiologists retrospectively evaluated images for motion artifact, foraminal stenosis, and central stenosis.
• Inter-rater agreement was quantified using weighted Fleiss's kappa, and comparisons were made using the Wilcoxon signed-rank test.

Main Results:

• Inter-rater agreement for foraminal stenosis was substantial for 2D (κ=0.76) and excellent for 3D (κ=0.81).
• Agreement for central stenosis was excellent for both 2D (κ=0.85) and 3D (κ=0.83).
• The 3D sequence demonstrated significantly reduced motion artifact (p≤0.001-0.036) and had a shorter mean scan time (7.3 min vs 10.8 min).

Conclusions:

• Deep learning-reconstructed 3D cervical spine MRI achieves excellent inter-observer agreement for stenosis assessment, comparable to standard 2D imaging.
• The 3D technique is less susceptible to motion artifacts and offers significant time savings.
• This advanced imaging approach holds promise for improving the accuracy and efficiency of cervical spine evaluations.