In this study Levin et al. aimed to develop a deep learning model to automatically segment rotator cuff muscles based on CT imaging to produce a standardized classification for volumetric muscle atrophy. A data set of 952 manually labeled shoulder CT scans were used to train and test the model. A novel classification scale called T-scores was developed to classify the degree of muscle atrophy, while 3D fat infiltration (3DFI%) and anterior-posterior (AP) muscle balance assessments were also evaluated by the model. The accuracy of the model to automatically segment rotator cuff muscles was assessed using a 5-fold cross validation which yielded Dice coefficients around 0.92. Upon comparison of muscle atrophy assessment across three groups– a healthy control, anatomic total shoulder arthroplasties (aTSA), and reverse total shoulder arthroplasties (rTSA) – it was found that the most severe atrophy occurred in the rTSA group while distinct atrophy patterns exist between rTSAs & aTSA. Results from fat infiltration experiments reported the highest 3DFI% in the rTSA group and only moderate 3DFI% in the aTSA group. AP balance investigations found that the aTSA cohort had a more posterior-dominant balance whereas the rTSA cohort had a more anterior-dominant balance. Integrating all data generated from the model, the authors concluded that rotator cuff atrophy was significant in aTSA (despite the notion that this approach may protect the muscle) and rTSA showed the most amount of muscle degeneration. Understanding differences in atrophy phenotypes and predictions of rotator cuff muscle atrophy allows surgeons to better predict surgical outcomes and complication risks.