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

Abstract

Background

Glenoid morphology and measurement techniques for bone loss in anterior shoulder instability have been the subject of much debate and multiple studies. We performed a retrospective comparison of computerised tomography (CT) and magnetic resonance imaging (MRI) scans, in order to assess if MRI can replace CT as the preferred imaging modality in these patients.

Methods

A custom deep learning algorithm was trained and validated in automatically segmenting T1 fat-suppressed (39 patients) and Volumetric Interpolated Breath-hold Examination MRI arthrogram images (25 patients). The MRI segmentations were compared to manually derived CT segmentations of the same shoulder, using a Dice Similarity Coefficient (DICE) score. Other important glenoid parameters were also measured and compared.

Results

The DICE mean was more than 0.95 for all image comparisons, showing near-perfect accuracy of the automatically segmented MRI images. The Spearman correlation coefficient for all measured variables was more than 0.84.

Discussion

Despite the fact that CT is still considered by most authors to be a superior imaging modality when compared to MRI in glenoid anatomy, we have demonstrated that our automated MRI processing platform provides nearly identical anatomical definition when compared to CT, with the additional benefits of soft tissue visualisation and avoidance of ionising radiation.