MRI augmented with novel artificial intelligence system is equivalent to CT in glenoid imaging

Area of Science:

• Orthopedic surgery
• Radiology
• Artificial intelligence in medicine

Background:

• Glenoid morphology and bone loss measurement are critical in anterior shoulder instability.
• Computed tomography (CT) is often considered the gold standard for evaluating glenoid anatomy.
• Debate exists regarding the optimal imaging modality for assessing glenoid bone loss.

Purpose of the Study:

• To compare the accuracy of magnetic resonance imaging (MRI) with a deep learning algorithm against CT for glenoid bone loss assessment.
• To determine if MRI can replace CT as the preferred imaging modality in patients with anterior shoulder instability.
• To evaluate the efficacy of automated MRI segmentation for glenoid anatomy.

Main Methods:

• Retrospective comparison of CT and MRI scans.
• Development and validation of a custom deep learning algorithm for automated MRI segmentation (T1 fat-suppressed and VIBE MRI arthrograms).
• Comparison of MRI segmentations with manually derived CT segmentations using Dice Similarity Coefficient (DICE) and Spearman correlation.

Main Results:

• The deep learning algorithm achieved a mean DICE score > 0.95, indicating near-perfect accuracy in MRI segmentation.
• Spearman correlation coefficients for all measured glenoid parameters exceeded 0.84.
• Automated MRI processing demonstrated high accuracy in anatomical definition.

Conclusions:

• Automated MRI processing platforms can provide glenoid anatomical definition comparable to CT.
• MRI offers additional benefits of soft tissue visualization and avoidance of ionizing radiation.
• This automated MRI approach shows potential to replace CT for glenoid assessment in shoulder instability.