Artificial intelligence-based quantitative computed tomography assessment of shoulder muscle and its association with outcomes after reverse total shoulder arthroplasty.

Background: Muscle atrophy and fatty infiltration of the shoulder muscles are known to influence clinical outcomes after reverse total shoulder arthroplasty (rTSA). Although quantitative evaluation of rotator cuff muscles has been increasingly reported, most previous studies have focused on intrinsic muscles and relied on two-dimensional or manual assessment methods. The purpose of this study was to evaluate shoulder muscle volume and composition using an artificial intelligence-based automated segmentation method on pre-operative computed tomography (CT) images and to investigate their associations with post-operative range of motion (ROM) after rTSA. Methods: This retrospective study included 13 patients (14 shoulders) who underwent rTSA for cuff tear arthropathy and related conditions with a minimum post-operative follow-up of 1 year. Pre-operative CT scans with 1-mm slice thickness were obtained within four weeks before surgery. Shoulder muscles, including intrinsic and extrinsic muscles, were automatically segmented using a deep learning-based model. Muscle volume and composition-functional muscle (FM), low-attenuation muscle, and adipose tissue (AT)-were quantified and normalized to scapular bone volume. Associations between muscle parameters and post-operative active ROM were analyzed using Spearman correlation coefficients. Results: The supraspinatus (SSP) exhibited a higher proportion of AT, possibly reflecting fatty infiltration, and was the only muscle showing a relatively balanced distribution of FM, low-attenuation muscle, and AT. The proportion of FM in the SSP was positively associated with active anterior elevation (AE), whereas the proportion of AT in the SSP and subscapularis (SSc) was negatively associated with active AE. In addition, total deltoid muscle volume and FM volume of the SSP were positively associated with post-operative AE. The AT volume of the SSc were negatively associated with active AE. No significant associations were identified between muscle composition or volume and post-operative external rotation. Conclusion: Artificial intelligence-based automated segmentation on pre-operative CT enabled detailed quantitative assessment of shoulder muscle volume and composition. Post-operative ROM tended to be associated with the deltoid, SSP, and SSc. Although these findings should be interpreted with caution due to limited statistical power, this exploratory feasibility study provides preliminary insights and may serve as a basis for future investigations.