Ultrafast lumbar spine MRI protocol using deep learning-based reconstruction: diagnostic equivalence to a...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Conventional lumbar MRI protocols are time-consuming, often exceeding 12 minutes.
• Deep learning-based reconstruction offers potential to accelerate MRI acquisition times.
• Optimizing MRI protocols is crucial for patient throughput and comfort.

Purpose of the Study:

• To assess the diagnostic equivalence of an ultrafast (1 minute 53 seconds) lumbar MRI protocol.
• To compare an ultrafast protocol with deep learning reconstruction against a conventional (12 minutes 31 seconds) protocol.
• To evaluate the impact of accelerated MRI on the detection of degenerative changes and other pathologies.

Main Methods:

• 58 patients underwent lumbar MRI using both conventional and ultrafast protocols.
• Included sagittal T1, T2, short-TI inversion recovery, and axial T2 sequences.
• Deep learning reconstruction compensated for signal-to-noise ratio reduction in the ultrafast protocol.
• Neuroradiologists graded degenerative changes and identified pathologies.
• Interprotocol agreement assessed using kappa statistics and equivalence index.

Main Results:

• Interprotocol intrareader agreement for degenerative changes showed substantial to almost perfect agreement (kappa values 0.707–0.804).
• The two protocols were interchangeable for most assessments, with equivalence index confidence intervals <5%.
• Interprotocol intrareader agreement for detecting other pathologies exceeded 98% for each pathology.

Conclusions:

• The ultrafast lumbar spine MRI protocol demonstrates near-diagnostic equivalency to the conventional protocol.
• Minor discrepancies were noted for specific degenerative changes (foraminal stenosis, disc contour).
• Deep learning-based accelerated MRI is a viable alternative for lumbar spine imaging.