This prospective study was approved by the local institutional review board (Cantonal Ethics Committee, Zurich, Switzerland). After providing written informed consent, ten healthy volunteers with no history of muscle injury, muscle disease, or previous shoulder surgery were enrolled, and demographic data was collected. Research was conducted in accordance with the ethical standards established by the institutional and/or national ethics committee and the 1964 Declaration of Helsinki and its subsequent amendments.

MRI data were acquired were performed on a 3-T system (MAGNETOM Prisma, Siemens Healthineers AG, Erlangen, Germany) using a dedicated 16-channel shoulder coil. MRI scans were performed between May and August 2023, and comprised a total of three scans of the same shoulder for each volunteer, with an interval of seven days between the scan sessions, respectively. Patient positioning was performed for each examination according to the local clinical standard. For the planning of diffusion-weighted imaging, the maximum visible length of the intramuscular tendon of the anterior SSP muscle bundle was identified on the water-only images of a coronal oblique proton-density weighted turbo-spin-echo DIXON sequence. Figure 1 illustrates the anatomy of the SSP muscle, which consists of an anterior fusiform portion with a bipennate fiber arrangement that contains the dominant tendon and a posterior portion, which has a more parallel, “strap-like” arrangement [19]. The diffusion-weighted imaging parallel to the SSP tendon orientation was performed along 64 directions [20, 21] with three b = 0 s/mm2 images as reference using a monopolar stimulated echo acquisition mode research sequence (mixing time: 200 ms) with a single-shot echo-planar imaging readout. The sequence parameters are listed in Table 1.

Schematic drawing depicting the anatomy of the SSP muscle of the right side at the level of the supraspinous fossa in an axial plane. The fusiform anterior portion of the supraspinatus (SSPant) has a bipennate architecture with a long intramuscular tubular tendon that inserts on the anterior aspect of the superior facet of the greater tuberosity and lies mostly superficial to the posterior portion of the muscle (SSPpost). The SSPpost has a “strap-like” appearance with fibers that are arranged in a more parallel pattern and inserts broadly on the posterior aspect of the superior facet of the greater tuberosity. C, clavicula; H, humeral head; Sc, scapula

Geometric inconsistencies due to eddy-current-induced distortions or physiological motion were corrected by non-rigid image registration [22]. For this purpose, all other images were registered to the first b = 0 s/mm2 image. Images were then processed in an in-house developed Matlab script (R2020b, Mathworks, Natick, MA, USA) for diffusion-tensor fitting. Next, the slice in which the maximum visible intramuscular length of the SSP tendon was visible was located in the b = 0 s/mm2 images. PAs were derived by projecting the first diffusion tensor Eigenvector onto the transverse imaging plane and computing the angle relative to the SSP tendon, which was manually drawn on the b = 0 s/mm2 image by a fellowship-trained radiologist (A.A.M.) (Fig. 2). Images were then analyzed by three fellowship-trained radiologists (A.A.M., G.W.K., and S.S.G.) who were blinded to all participant data and read images in random order on a dedicated workstation. Two freehand regions of interest (ROIs) were drawn by each reader on the relevant slice of maximum tendon visualization by the readers in the computed PA maps using ITK-SNAP (v4.0.2) [23]. The ROIs encircled the anterior and posterior SSP muscle bundle respectively (hereafter, ROIant and ROIpost), while surrounding structures and the SSP tendon itself were carefully excluded. PA measurements of one reader (A.A.M.) were repeated after a three-week interval to assess intra-reader agreement.

Following acquisition and registration, images were processed using an in-house developed Matlab script: first, the SSP tendon (arrowheads, A; white line, B) was manually drawn into the b = 0 s/mm2 images, followed by tensor fitting. Color-coded directions of the first Eigenvector in each voxel are shown, with green color indicating anteroposterior orientation, red color indicating mediolateral orientation, and blue color indicating superoinferior orientation (B). The mathematical operation to obtain the 2D PA in relation to the SSP tendon (T) is visualized in C. The first Eigenvector (v) is projected in the transverse plane (vp), followed by a computation of the angle relative to the SSP tendon. Freehand ROIs (dashed lines) were then drawn in the anterior SSP portion (ROIant) and posterior SSP portion (ROIpost) in the PA maps (D). Images of the right shoulder of a 31-year-old male subject in the oblique axial plane are shown

DTI tractography was performed solely for visualization purposes and was not a primary endpoint of this study. The motion-corrected STEAM dataset of an exemplary subject at a single time point was processed in MRtrix3 [24]. After estimating a mask, a deterministic tensor algorithm was applied for tractography. At each streamline step, the diffusion tensor was fitted to the local (trilinear-interpolated) diffusion data, and the trajectory was determined by the principal eigenvector of that tensor. Seed points were set randomly, and a total of 100.000 streamlines were generated, whereby a minimum length of 50 mm was enforced, and the angular cutoff was set to 20°.

SPSS (v29, IBM Corp., Armonk, USA) was used for all statistical analyses. The Shapiro–Wilk test was used to test for the normal distribution of continuous data. Data are presented as mean ± standard deviation if normally distributed and as median with range in parentheses in cases of non-normal distribution. PAs were averaged across examination time points and readers for further analysis. To assess the test-retest reliability, inter-reader agreement, and intra-reader agreement of PAs measured with DTI, intraclass correlation coefficients (ICCs) for absolute agreement were calculated using a two-way mixed model. Results were categorized according to Koo and Li [25]: ICC > 0.90: excellent; 0.75–0.90 = good; 0.50–0.75 = moderate; < 0.50 = poor. p-values below an alpha level of 0.05 were considered significant.