The Muscle Cross-sectional Area on MRI of the Shoulder Can Predict Muscle Volume: An MRI Study in Cadavers

Heath B. Henninger, Garrett V. Christensen, Carolyn E. Taylor, Jun Kawakami, Bradley S. Hillyard, Robert Z. Tashjian, Peter N. Chalmers

Research output: Contribution to journalArticlepeer-review

12 Citations (Scopus)


BackgroundMuscle volume is important in shoulder function. It can be used to estimate shoulder muscle balance in health, pathology, and repair and is indicative of strength based on muscle size. Although prior studies have shown that muscle area on two-dimensional (2-D) images correlates with three-dimensional (3-D) muscle volume, they have not provided equations to predict muscle volume from imaging nor validation of the measurements.Questions/purposesWe wished to create an algorithm that quickly, accurately, and reliably estimates the volume of the shoulder muscles using cross-sectional area on MR images with low error. Specifically, we wished to (1) determine which MR imaging planes provide the highest correlation between shoulder muscle cross-sectional area and volume; (2) derive equations to predict muscle volume from cross-sectional area and validate their predictive capability; and (3) quantify the reliability of muscle cross-sectional area measurement.MethodsThree-dimensional MRI was performed on 10 cadaver shoulders, with sample size chosen for comparison to prior studies of shoulder muscle volume and in consideration of the cost of comprehensive analysis, followed by dissection for muscle volume measurement via water displacement. From each MR series, 3-D models of the rotator cuff and deltoid muscles were generated, and 2-D slices of these muscle models were selected at defined anatomic landmarks. Linear regression equations were generated to predict muscle volume at the plane(s) with the highest correlation between volume and area and for planes identified in prior studies of muscle volume and area. Volume predictions from MR scans of six different cadaver shoulders were also made, after which they were dissected to quantify muscle volume. This validation population allowed the calculation of the predictive error compared with actual muscle volume. Finally, reliability of measuring muscle areas on MR images was calculated using intraclass correlation coefficients for inter-rater reliability, as measured between two observers at a single time point.ResultsThe rotator cuff planes with the highest correlation between volume and area were the sum of the glenoid face and the midpoint of the scapula, and for the deltoid, it was the transverse plane at the top of the greater tuberosity. Water and digital muscle volumes were highly correlated (r ≥ 0.993, error < 4%), and muscle areas correlated highly with volumes (r ≥ 0.992, error < 2%). All correlations had p < 0.001. Muscle volume was predicted with low mean error (< 10%). All intraclass correlation coefficients were > 0.925, suggesting high inter-rater reliability in determining muscle areas from MR images.ConclusionDeltoid and rotator cuff muscle cross-sectional areas can be reliably measured on MRI and predict muscle volumes with low error.Clinical RelevanceUsing simple linear equations, 2-D muscle area measurements from common clinical image analysis software can be used to estimate 3-D muscle volumes from MR image data. Future studies should determine if these muscle volume estimations can be used in the evaluation of patient function, changes in shoulder health, and in populations with muscle atrophy. Additionally, these muscle volume estimation techniques can be used as inputs to musculoskeletal models examining kinetics and kinematics of humans that rely on subject-specific muscle architecture.

Original languageEnglish
Pages (from-to)871-883
Number of pages13
JournalClinical Orthopaedics and Related Research
Issue number4
Publication statusPublished - 2020 Apr 1


Dive into the research topics of 'The Muscle Cross-sectional Area on MRI of the Shoulder Can Predict Muscle Volume: An MRI Study in Cadavers'. Together they form a unique fingerprint.

Cite this