Prediction of rotator cuff muscle fibre orientations using a population-averaged atlas generated with anatomical and diffusion-weighted magnetic resonance images

Prediction of rotator cuff muscle fibre orientations using a population-averaged atlas generated with anatomical and diffusion-weighted magnetic resonance images | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Prediction of rotator cuff muscle fibre orientations using a population-averaged atlas generated with anatomical and diffusion-weighted magnetic resonance images Yilan Zhang, Robert Lloyd, Robert D. Herbert, Lynne E. Bilston, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4683327/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Measurements of muscle architecture are crucial for understanding muscle function but are often difficult to obtain in human muscles in vivo . This study aimed to create population-averaged atlases of human rotator cuff muscle shape and muscle fibre orientations from anatomical magnetic resonance images (MRI) and diffusion-weighted images (DWI), and to utilize these atlases to predict muscle fibre orientations from anatomical MRI data alone. An image registration framework was applied to co-register anatomical MRI and DWI data of 11 male and 9 female subjects into sex-specific common spaces, forming the basis for the atlases. The accuracy of registration was quantified using Dice coefficients, angular correlation coefficients (ACCs), and angular differences. The same metrics were used to assess the capability of the atlases to predict fibre orientations for subjects not included in the atlas construction, via leave-one-out cross-validation. The results showed that individual male and female image data were accurately registered into their respective atlas spaces, with high Dice coefficients (0.888 ± 0.002 for males, 0.856 ± 0.021 for females) and consistent angular alignment as evidenced by the ACCs and angular differences. Predicted fibre orientations for out-of-sample subjects closely matched those derived from DWI images, exhibiting improved smoothness and coverage (ACC: 0.909 ± 0.011 for males, 0.942 ± 0.011 for females; angular difference: 13.8 ± 1.3° for males, 11.2 ± 1.2° for females). These findings demonstrate that population-averaged atlases not only enhance muscle architecture reconstructions but also enable the accurate prediction of muscle fibre orientations using only anatomical MRI scans. Rotator cuff muscles fibre orientation muscle architecture musculoskeletal imaging diffusion-weighted imaging multi-channel registration Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1 Introduction Skeletal muscle architecture, the macrostructural arrangement of muscle fibres in the muscle belly, is the primary determinant of a muscle’s capacity to generate force and change length (Lieber & Fridén, 2000 ). While the architecture of any particular muscle is broadly similar across individuals, muscle architecture can adapt to exercise (Alonso-Fernandez et al., 2018 ; Blazevich et al., 2003 ; Roig et al., 2008 ), ageing (Narici et al., 2003 ; Papenkort et al., 2021 ; Shur et al., 2021 ; Siebert et al., 2017 ) and disease (Bodine et al., 1982 ; Fahn-Lai et al., 2020 ; Kruse et al., 2018 ; J. E. Park et al., 2019 ). Quantitative measurement of muscle- and subject-specific architecture is therefore useful for the study of muscle adaptation and muscle dysfunction. The human rotator cuff, comprising the supraspinatus, subscapularis, infraspinatus, and teres minor muscles, plays a significant role in movement of the upper limb and provides dynamic stability to the glenohumeral joint. Rotator cuff injuries are common – they constitute up to half of all significant shoulder injuries seen in some clinical contexts (Gazielly et al., 1994 ; Gerber et al., 2000 ) – and can be difficult to treat. Surgical treatment of rotator cuff injuries might be improved with preoperative planning tools built around biomechanical models of the shoulder. Computational models can be used to predict the biomechanical consequences of rotator cuff tears on muscle function and joint forces (Khandare et al., 2022 ; Vidt et al., 2018 ) or guide the design of implants in shoulder arthroplasty procedures (Büchler & Farron, 2004 ). However, due to the difficulty in obtaining subject-specific measurements of rotator cuff muscle architecture, most musculoskeletal shoulder models are generic or scaled-generic models (Prinold et al., 2013 ), which, despite their utility, often fall short of capturing the anatomical variations unique to each individual. Computational models and surgical planning tools might be more useful if they could incorporate subject-specific measurements of rotator cuff muscle architecture. Diffusion-weighted imaging (DWI) enables accurate measurement of human rotator cuff muscle architecture in three dimensions, using fibre tracking algorithms which propagate fibre tracts along the principal eigenvectors of diffusion tensors throughout a muscle (Zhang et al., 2023 , 2024 ). This approach not only provides a visual representation of muscle fibres orientations, but also allows for the quantification of muscle architecture such as the fascicle length and pennation angle – parameters that are crucial in understanding muscle function but cannot be measured with commonly used anatomical MRI scans. However, DWI-based reconstructions of muscle architecture are sensitive to noise and image artifacts inherent in DWI data, which can distort the path of fibre tracts and create regions in the muscle that contain few fibres. New methods are needed to improve muscle architecture reconstructions from MRI data. Ideally such methods would use standard scanning protocols and automated image processing procedures so that the methods could be routinely implemented in clinical practice. Some of the current limitations of DWI-based muscle architecture reconstructions can be addressed with a population-based approach that characterises inter-individual variability. Statistical shape modelling (SSM) is now commonly used for this purpose. After computing a mean shape across the population and transforming each individual’s shape to the mean, principal component analysis can be used to identify the main modes of inter-individual shape variation (Hotelling, 1933 ; Pearson, 1901 ). SSMs have been used predominantly to study human skeletal anatomy (e.g., humerus (Vlachopoulos et al., 2018 ), scapula (Salhi et al., 2020 )). However, there have been relatively few attempts to apply SSMs to skeletal muscles (e.g., facial muscles (Tran et al., 2023 ), levator ani (Su-Lin Lee et al., 2009 ), soleus (Bin Ghouth et al., 2022 ) and hamstrings (Sutherland et al., 2023 )). To the best of our knowledge, no studies have applied SSM or population-based methods to shoulder muscles. Moreover, existing population-based methods usually only model the shape (outer surface) of the muscle, but do not model internal fibre orientations. One recent study included model fibre orientations in a population-averaged muscle modelling framework (Bolsterlee, 2022 ). However, that study only used surface features to find corresponding points between muscles from different individuals. SSMs might be able to more accurately represent the internal architecture of muscles if, in addition to using information about the shape of the muscle surface, they used information about muscle fibre orientations. This could be achieved by registering multiple channels of information – specifically, by combining muscle surface data derived from anatomical MRI with fibre orientation data derived from DWI. Registration of fibre orientations and development of DWI-based population-averaged atlases have been a focus in neuroimaging (Forsberg et al., 2011 ; Roura, 2015 ; Uus et al., 2020 , 2021 ), but to our knowledge those methods have not yet been used in muscle imaging. Population-averaged atlases built using multi-channel registration introduce novel possibilities for muscle architecture analysis. Firstly, atlases may enhance the representation of individual muscle architecture by aggregating information across the population. Aggregation can potentially reduce noise and image artifacts typically present in individual scans and enhance the robustness and reliability of muscle architecture reconstructions. Secondly, since previous studies on rotator cuff muscle architecture have reported limited interindividual variability (Zhang et al., 2024 ), there is potential that atlases can be used to predict individual muscle architecture from muscle shape alone, simplifying the generation of high-quality muscle architecture reconstructions for musculoskeletal modelling and assessments. The goals of this study were, therefore, to: 1) develop population-averaged atlases of human rotator cuff muscles using multi-channel registration that combines anatomical data derived from MRI with fibre orientation data derived from DWI; and 2) use the atlases to predict muscle fibre orientations from anatomical MRI scans. We hypothesised that fibre orientations can be predicted accurately without DWI by registering anatomical MRI from a new subject to the anatomical MRI channels of the population-averaged atlas and then applying the resulting transformation to the channels of the atlas that encode muscle fibre orientations. 2 Materials and Methods 2.1 Participants Study procedures were approved by the UNSW Human Research Ethics Committee (HREC approval HC200971). Each participant was informed about the study procedures and provided their written consent before participation. The study involved magnetic resonance imaging (MRI) of the right shoulder of 20 participants (11 males and 9 females; age 28 ± 6 years; height 171 ± 8 cm; weight 64 ± 11 kg, values are mean ± standard deviation). People with symptoms or recent history of shoulder pathology were excluded from participating. Some of the data used in this study have been used in previous studies on rotator cuff muscle architecture (Zhang et al., 2023 , 2024 ). 2.2 MRI acquisitions and processing A brief overview of image acquisition and processing will be provided here. A more detailed description can be found elsewhere (Zhang et al., 2023 , 2024 ). All MR images were acquired at 3T (Philips Ingenia CX, Philips Healthcare, Best, The Netherlands) using a 16-channel anterior body coil and a posterior coil integrated in the scanner bed. The protocol consisted of an mDixon scan (field of view 240 mm, voxel size 1 × 1 × 2 mm, 210 slices) and two diffusion-weighted scans (field of view 190 mm, voxel size 2.5 × 2.5 × 5 mm, 24 slices) covering the proximal and distal rotator cuff musculature, respectively. Segmentation of rotator cuff muscles and bones (humerus, scapula and clavicle) was initially performed manually on a subset of 12 mDixon scans, after which a deep learning model (nnU-net, (Isensee et al., 2021 )) was trained for automatic segmentation of the remaining scans. The predicted segmentations were visually verified and adjusted when needed. In four out of twenty scans the boundary between the infraspinatus and teres minor muscles was not clearly visible. These muscles were therefore grouped together for all participants. Previous analysis demonstrated excellent intra-rater reliability of segmentation (average intraclass correlation coefficient across muscles of 0.97; (Zhang et al., 2024 )). DWI scans were filtered with a Marchenko-Pastur principal component analysis filter (Veraart, Fieremans, et al., 2016; Veraart, Novikov, et al., 2016 ) to reduce image noise, and then corrected for eddy current- and possible motion-induced distortions using functions TOPUP and EDDY from FSL (Andersson & Sotiropoulos, 2016 ), implemented in MRtrix (MRtrix3; (Tournier et al., 2019 )). The processed DWI scans were then upsampled to match the spatial dimensions of the mDixon scan and combined into a single DWI image set covering the entire rotator cuff. Rigid registration via FLIRT tool from FSL (Jenkinson et al., 2002 ; Jenkinson & Smith, 2001 ) was performed to correct for small misalignments between the mDixon and DWI images within subjects. The accuracy of registration was confirmed visually in ITK-SNAP (Yushkevich et al., 2006 ). To increase anatomical contrast and improve subsequent image registration, the mDixon scans, stitched DWI scans, and segmented masks were linearly upsampled to an isotropic voxel size of 0.75 × 0.75 × 0.75 mm. 2.3 Construction of population-averaged atlases The construction of population-averaged atlases was a multi-step process involving estimation of fibre orientation distributions (FODs) from DWI data, concurrent multi-channel registration of individual mDixon images and the FODs into a common reference frame, and averaging across subjects to construct a cohesive atlas. Population-averaged atlases were created separately for males (n = 11) and females (n = 9). In the initial phase of this study we tried to construct a single multi-channel population-averaged atlas for both males and females. However, the large differences in muscle volumes between males and females led to difficulties in achieving accurate alignment, so therefore we created create sex-specific atlases. 2.3.1 Fibre orientation distribution estimation from DWI FODs were derived from DWI data using algorithms included in MRtrix (Tournier et al., 2004 , 2007 ). While FODs can describe multiple fibre orientations within a voxel, previous investigations have shown that fibres within rotator cuff muscles do not cross each other so that muscle fibre orientations within a voxel can be represented well by a single orientation (Zhang et al., 2024 ). We therefore limited the maximum spherical harmonics (SH) degree to two in the FOD estimation process, retaining only the principal diffusion direction within a voxel, analogous to a single-fibre diffusion tensor model. The response function (Tournier et al., 2004 ), which describes the diffusion signal intensity based on the orientation of a single fibre bundle when exposed to magnetic gradients, was used as a kernel for spherical deconvolution. The FOD estimation was then performed on DWI images with constrained spherical deconvolution using the response function as an input (Tournier et al., 2007 ). Segmented muscle masks were used to restrict FOD estimation to the rotator cuff muscles. 2.3.2 Multi-channel registration Data from n-1 subjects (n = 10 for males, n = 8 for females) was used for the creation of each population-averaged atlas. For both male and female atlases, one subject was left out for out-of-sample evaluation (see Evaluation Section). The registration pipeline is based on an intensity-based multi-channel registration framework implemented in MRtrix. The input channels for each subject consist of all four mDixon images (water only, fat only, in-phase, and out-of-phase) and six FOD image volumes. Masks of rotator cuff muscles and the scapula were used as inputs to speed up registration and to target the registration to the primary region of interest, although they did not contribute to the computation of displacement fields necessary for aligning the images. The pipeline consists of two steps: initialisation and iterative registration (see Fig. 1 ). All registrations used symmetric normalisation (SyN) Demons (Avants et al., 2007 ) with the sum of squared difference (SSD) metric and reorientation of FOD using apodised point spread functions (Raffelt et al., 2012 ). Initialisation The initialisation step created an initial atlas that is unbiased to any individual subject in the population, serving as a reference for the subsequent registration process. First, the average coordinate space (or image grid) of all input FODs was calculated. Then, for each subject, the mDixon I and FOD J channels were rigidly registered to the averaged coordinate space, respectively. Each channel’s transformations were then averaged and applied to the mDixon and FOD data. The preliminary set of mDixon \({T}_{0}^{I}\) and FOD \({T}_{0}^{J}\) atlases was then generated by computing the median image intensity across all registered mDixon \({I}_{0}\) and FOD \({J}_{0}\) volumes, respectively. We empirically found that using the median instead of the mean of image intensities gave clearer structure boundaries. Iterative registration Unbiased population-averaged atlases were constructed through a process of iterative registration. In each iteration, the median intensity values of the registered mDixon and FOD volumes were computed separately to continually update and refine their respective atlases. This process involved iterating over 28 stages, organised into a sequence of 6 rigid, 6 affine, and 16 non-linear registration stages. The atlas generation at each stage k can be summarised in the following steps: 1) For each subject, mDixon \({I}_{k-1}\) ( \(k\in\) [1,28]) and FOD \({J}_{k-1}\) channels were registered to their corresponding updated atlases \({T}_{k-1}^{I}\) and \({T}_{k-1}^{J}\) (to preliminary atlases \({T}_{0}\) in the first stage) using rigid, affine or non-linear registration based on the stage. 2) Resulting transformations (rigid and affine registrations) or displacement fields (non-linear registration) from each channel were averaged ( \(\varnothing\) ) and applied to mDixon and FOD data. 3) The median of registered mDixon \({I}_{k}\) and FOD \({J}_{k}\) volumes were computed to update their atlases \({T}_{k}^{I}\) and \({T}_{k}^{J}\) . A multi-resolution pyramid with default scale values in MRtrix (Table 1 ) was used to perform registration at different resolution levels, starting with down-sampled images (scale < 1) and progressively working towards original high-resolution images (scale = 1). The use of this multi-resolution pyramid enhances efficient image registration by focusing on large structures first, followed by focusing on finer details (Thevenaz et al., 1998 ). Registration parameters can be found in Table 1 . Table 1 Summary of registration parameters. Stage no. Type of registration Number of iterations a SH degrees Multi-resolution pyramid b 1 to 6 rigid 1000 2 0.3, 0.4, 0.6, 0.8, 1.0, 1.0 7 to 12 affine 1000 2 0.3, 0.4, 0.6, 0.8, 1.0, 1.0 13 to 28 non-linear 500 2 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 a The number of registration iterations used within each stage before updating the atlas. b The list of scale values that represents the degree to which the original image was downsampled in each iteration. SH = spherical harmonic The displacement field obtained at each non-linear iteration was smoothed using a Gaussian filter with a standard deviation of twice the voxel size of the iteration’s resolution. After registration, the output consisted of three components: 1) A population-averaged mDixon atlas \({T}_{final}^{I}\) and a corresponding FOD atlas \({T}_{final}^{J}\) in a common coordinate space; 2) individual subject image data, specifically, mDixon images \({I}_{r}\) ( \(r\in \left[1,n-1\right]\) ), FODs \({J}_{r}\) , and masks of muscles and the scapula, each registered and aligned to the atlas space; and 3) for each subject, a displacement field \({\varnothing }_{r}\) that warps subject images to the atlas space. This included both linear (rigid and affine) and non-linear displacements. We computed the atlas mask by calculating the median of all registered subject masks in atlas space. 2.3.3 Evaluation Qualitative evaluation Qualitative evaluation included reconstruction, visual inspection, and comparison of fibre tractography of all muscles to visually assess fibre orientations. Fibre tracts were reconstructed from each subject’s FODs, as well as from the population-averaged FOD atlases. For tractography, we employed the MRtrix SD_STREAM algorithm (Tournier et al., 2012 ), a deterministic algorithm based on spherical deconvolution. The tracts were dynamically seeded using the SIFT model within MRtrix (Smith et al., 2015 ), a mechanism designed to improve the evenness of tract density distribution. In each muscle, a total of 3000 fibre tracts were generated with the following settings: integration step size = 1.0 mm; 0.1 \(\le\) fractional anisotropy \(\le\) 0.5; maximum turning angle between successive steps > 15°; 25 mm \(\le\) tract length \(\le\) 200 mm. These reconstructions were then examined in 3D Slicer (Fedorov et al., 2012 ) for 3D visualisation and visual comparison of fibre orientations. Quantitative evaluation The multi-channel registration process was validated using leave-one-out cross-validation (LOOCV), independently for male and female atlases. Thus, 11 male and 9 female atlases were created, each based on a different subset of 10 and 8 subjects, respectively. All atlases were visually inspected to ensure proper alignment between channels and to verify the anatomical fidelity of structures. We then quantitatively evaluated the performance of multi-channel registration using three metrics per muscle. The spatial alignment accuracy was measured by the Dice coefficient (Dice, 1945 ), which quantifies the degree of overlap between the atlas mask and the mask of each subject warped to the atlas space: $$\text{D}\text{i}\text{c}\text{e}=\frac{2\left|A\cap B\right|}{\left|A\right|+\left|B\right|}$$ 1 where A and B represents the region of the atlas mask and the warped mask of each subject. A Dice coefficient closer to 1 indicates a greater degree of alignment between the spatial locations of the masks. The angular alignment between estimated and subject-specific fibre orientations was quantified by calculating the angular correlation coefficient (ACC) (Anderson, 2005 ) between the FOD atlas and the FOD of each subject warped to the atlas space. The ACC was determined for each muscle on a voxel-by-voxel basis as: $$\text{A}\text{C}\text{C}=\frac{\sum _{m=-l}^{l}{a}_{l}^{m}{b}_{l}^{m}}{\left({\sum _{m=-l}^{l}{{(a}_{l}^{m})}^{2} )}^{\frac{1}{2}}\right({\sum _{m=-l}^{l}{{(b}_{l}^{m})}^{2} )}^{\frac{1}{2}}}$$ 2 where \({a}_{l}^{m}\) and \({b}_{l}^{m}\) represents the coefficient for the SH function of degree \(l\) ( \(l=2\) in this study) and order \(m\) of the FOD atlas and the warped FOD, respectively. To further quantify the alignment between predicted and subject-specific fibre orientations, we first calculated the principal diffusion direction within each voxel of the FODs using the sh2peaks function in MRtrix. Subsequently, we calculated the 3D angular difference between each of the peak direction vectors for the atlas and the corresponding vectors warped from the subject’s FOD. While the ACC provides a dimensionless measure of similarity in orientation distribution, the angular difference provides a measure of muscle fibre orientation alignment accuracy expressed in a familiar metric (degrees). The median ACCs and median angular difference were used to summarise the alignment between predicted and subject-specific fibre orientations for a muscle. Histograms illustrating the distributions can be found in the Supplementary Material. The mean Dice coefficient, ACC, and angular difference, along with their respective standard deviations, were calculated for each individual case from the LOOCV. 2.4 Prediction of fibre orientations from anatomical MRI For the second aim of this study, we used the atlas to predict fibre orientations for out-of-sample subjects, using only the subject’s anatomical mDixon scan (which does not contain information about fibre orientations). Predicted fibre orientations were compared to orientations determined from the subject’s DWI data. 2.4.1 FOD prediction The mDixon scan of the out-of-sample subject was registered to the population-averaged mDixon atlas using SyN Demons with the SSD metric, mirroring the rigid, affine, and non-linear registration stages used in atlas building. The resulting displacement field was inverted, reoriented, and then used to warp the FOD atlas into the anatomical space of the out-of-sample subject. The atlas muscle mask was also warped to the subject space for subsequent evaluations. 2.4.2 Evaluation For the qualitative evaluation of fibre orientation prediction, we applied the same fibre tractography methods as described previously. Using 3D Slicer, we visually examined and compared the fibre tract reconstructions from subject-specific DWI-derived FODs against atlas-predicted FODs for the out-of-sample subjects. To determine the accuracy of image registration between the subject’s mDixon and the population-averaged mDixon atlas, the Dice coefficient was calculated between the subject’s mask and the atlas mask warped to subject space. Similarly, to determine the accuracy of fibre orientation prediction, the ACC and angular difference between the subject’s FOD and the atlas FOD warped to subject space was calculated. 3 Results We successfully constructed 11 sets of atlases of mDixon and FOD maps for the male cohort and 9 sets for the female cohort. Data derived for each atlas, including Dice coefficients, ACCs, and angular differences for both the evaluation of multi-channel registration performance and the fibre orientation prediction, can be found in Table S1 and Table S2 in the Supplementary Material. 3.1 Multi-channel population-averaged atlas and registration performance Figure 2 . Examples of one male (top) and one female (bottom) multi-channel atlas. There are two rows for the male and two rows for the female: the upper and lower rows illustrate transverse mDixon slices approximately midway through the acromioclavicular and glenohumeral joint, respectively. Shown from left to right are: the mDixon water image, masks of rotator cuff muscles and scapula overlayed on mDixon image (blue, scapula; beige, subscapularis; pink, infraspinatus and teres minor; orange, supraspinatus), and an FOD image overlayed on mDixon. FODs are colour-coded according to orientation (red: left-right; green: anterior-posterior; blue: inferior-superior). 3.1.1 Qualitative evaluation Figure 3 . Comparison of 3D fibre tractography reconstructions of rotator cuff muscles from the right shoulder, using data from an atlas generated from 10 male subjects. For each muscle, the illustration showcases the tractographies for subjects with the (a) best, (b) median, and (c) worst angular alignment within this cohort. Each pair highlights the fibre reconstructions set against the 3D surface model of the muscle in transparent yellow; the top and bottom row displays fibre reconstructed from the subject’s original FODs and the FOD atlas, respectively. Fibre tracts are colour-coded according to orientation (red: medial-lateral; green: anterior-posterior; blue: inferior-superior). 3.1.2 Quantitative evaluation There was a consistently high degree of spatial overlap and angular alignment between the images of subjects registered to the atlases and the corresponding atlases for both male and female cohorts as evidenced by the high Dice coefficients (males: 0.888; females: 0.856), high ACCs (males: 0.949; females: 0.974), and small angular differences (males: 10.5°; females: 7.8°; Fig. 4 ). Small standard deviations observed across the atlases for the LOOCV (Table 2 ) further indicate the high degree of consistency in registration performance. Table 2 Summary of leave-one-out cross-validation results. Metric Dice coefficient ACC Angular difference (°) INF + TER SUB SUP INF + TER SUB SUP INF + TER SUB SUP Male 0.890 (0.006) 0.890 (0.003) 0.885 (0.005) 0.956 (0.001) 0.937 (0.002) 0.953 (0.002) 9.8 (0.2) 11.8 (0.2) 9.8 (0.2) Female 0.881 (0.005) 0.857 (0.006) 0.829 (0.010) 0.974 (0.001) 0.968 (0.001) 0.981 (0.001) 7.6 (0.2) 8.9 (0.2) 6.9 (0.2) Values are means (standard deviations) for each metric across all atlases within each cohort. INF = infraspinatus; TER = teres minor; SUB = subscapularis; SUP = supraspinatus. 3.2 Prediction of fibre orientations for out-of-sample subjects 3.2.1 Qualitative evaluation Fibre tracts of out-of-sample subjects reconstructed from original and predicted fibre orientations showed a high degree of visual correspondence (Fig. 5 ), demonstrating that the atlas can effectively predict complex patterns of 3D fibre orientations using anatomical image data alone. Fibres reconstructed from the predicted fibre orientations generally showed smoother tracts which covered a larger region of the muscles, particularly at the medial border of the subscapularis, infraspinatus and teres minor, compared to those from the original fibre orientations. 3.2.2 Quantitative evaluation The primary results of the quantitative evaluation of fibre orientation prediction accuracy are illustrated in Fig. 6 . High Dice coefficients and ACCs, coupled with relatively low angular differences were found for both male and female atlases. Variability in these metrics was noted between subjects and across different muscles but remained within a relatively narrow range. Detailed histograms illustrating the distribution of these metrics across each muscle for each atlas are available as Fig. S2 in the supplementary material. Furthermore, Fig. 7 provides a visualisation of fibre orientation prediction accuracy on transverse slices for an out-of-sample subject. Regions of relatively low ACCs and correspondingly high angular differences, indicating regions of less satisfactory angular alignment, were most obvious at the muscle boundaries and within the subscapularis. 4 Discussion This manuscript describes the construction of an atlas of human rotator cuff muscle shape and architecture, and demonstrates two important applications. First, we used the atlases to smooth fibre tracts and fill in areas where fibre tracts were sparsely reconstructed due to image noise, substantially improving the quality of subject-specific muscle reconstructions. Second, we demonstrated the use of the atlases to accurately predict fibre orientations from anatomical MRI alone, bypassing the need to conduct diffusion-weighted scans, at least for some applications. Our multi-channel registration pipeline was built on methods that have been widely applied in neuroimaging (Forsberg et al., 2011 ; Roura, 2015 ; Uus et al., 2020 , 2021 ). One example is a study that integrated T 1 , T 2 and FOD data from 20 neonatal brains using similar registration methods and reported the mean Dice coefficient of 0.735 and ACC of 0.455, averaged across brain tissues (Uus et al., 2020 ). Our study yielded much higher Dice coefficients and ACCs, probably because rotator cuff muscles have less complex fibre pathways than the neonatal brain. The high Dice coefficients and ACCs demonstrate the ability of our registration pipeline to effectively characterise inter-individual differences in both the macrostructural anatomy and internal architecture of the muscles. Despite good quantitative registration results, we noticed that boundaries between adjacent muscle groups in the mDixon atlases were sometimes blurry, and these regions exhibited lower angular alignment, indicating relatively poor registration performance. In general, the need for multi-channel registration to synthesise information about anatomical structure obtained from mDixon images with information about fibre orientation obtained from DWI presented unique challenges. For example, noise and image artifacts in DWI data could bias the registration process. Nonetheless, population-averaged atlases had noticeably less noise and fewer artifacts than the muscle reconstructions calculated from original DWIs of individual subjects, as demonstrated by both qualitative and quantitative evaluation. Thus population-averaging of muscle reconstructions has a noise-cancelling effect, as has been observed previously in brain imaging studies (Jones et al., 2002 ; H. Park, 2003 ). The fidelity of our registration outputs, atlas construction and predicted fibre orientations might be enhanced by further developing the multi-channel registration algorithm. For instance, a weighted averaging approach to combining channel-specific updates to displacement fields (Forsberg et al., 2011 ), which assigns greater weight to channels with higher certainty or image quality, may offer a more accurate registration. The efficacy of this approach in enhancing registration accuracy of the neonatal brain has been demonstrated previously (Uus et al., 2020 ). The multi-channel registration approach and the resultant population-averaged atlases have potentially broad application. The atlases could serve as a highly accurate mean shape for SSMs, with correspondences established through the advanced registration process that integrates information from both anatomical and diffusion MRIs. This integration offers an enhanced level of detail and accuracy over traditional landmark-based approach, and provides a more precise representation of the complex rotator cuff anatomy. Moreover, incorporating data on fibre orientations, which are typically not accounted for in population-based approaches to muscle modelling, could improve predictions of muscle function made by musculoskeletal models. Furthermore, the population-averaged FOD atlas presents a promising approach to mitigating the challenges posed by the intrinsic sensitivity of DWI to noise and image artifacts. By leveraging the atlas’s smoothed representations and expanded coverage of muscle fibres, it becomes possible to refine the analysis of subject-specific fibre orientations. For example, the atlas could be employed as a filtering tool to selectively combine subject fibre orientations derived from DWI with the atlas information, according to certain criteria that prioritise anatomically plausible fibre orientations. This integration enables the use of anatomically constrained fiber tractography—a methodological framework we have established in our previous work (Zhang et al., 2023 , 2024 ) – for subject-specific reconstructions and measurements of muscle architecture. By reducing or mitigating anatomically implausible fiber tracts and enhancing tract coverage, we anticipate that this refined analysis may improve measurement accuracy and deepen our understanding of muscle function. Another promising application of the approach presented here is prediction of fibre orientations from anatomical images when DWI data is not available. For example, predicted fibre orientations could be used to assign fibre orientations in subject-specific finite element models of muscles, thereby enhancing their anatomical fidelity and the accuracy of simulations. It has been shown that fibre orientations significantly affected muscle mechanics during simulated contractions (Alipour et al., 2017 ), highlighting the significance of accurate assignment of fibre orientations in finite element models. 4.1 Limitations and future directions The generalisability of the atlases presented in this study is limited by the small number and age range of subjects included in our cohort. This specificity limits the atlases’ applicability to populations that diverge from our study group, potentially reducing their utility in more diverse research and clinical settings. Future research should focus on expanding the demographic breadth of the study cohorts by including data from subject with a wider range of age and physical condition. Examination of the out-of-sample subjects for whom fibre orientation prediction was least accurate found that these subjects’ mDixon images had poorer image quality. It was also evident that partial volume effects in some mDixon images made identification of muscle boundaries difficult. These aspects of image quality are likely to be key factors affecting registration accuracy. Registration accuracy is also likely to be reduced when there is large between-subject variation in muscle volume. This challenge was particularly evident in our initial attempts to construct a single atlas including both male and female subjects. The significant differences in muscle volumes between sexes hindered our ability to achieve accurate alignment, compromising registration quality. This issue underscored the need for further improvements in image registration to accommodate large anatomical variations across individuals and cohorts. Collectively, these findings underscore the importance of obtaining high-quality anatomical MRI scans and the need for robust registration algorithms capable of compensating for anatomical variability. Variability in imaging protocols and equipment across different studies also poses a challenge. Our registration and prediction algorithms were optimised for a specific set of MRI scanner settings and protocols. As such, differences in equipment and scanning parameters that exist in broader clinical practice could affect the robustness and reliability of the atlases when applied outside the controlled conditions of our study. Specifically, our current fibre orientation prediction method hinges on accurate registration of mDixon images from new subjects to these mDixon-based atlases, where the imaging conditions were closely matched. Applying these atlases to images acquired with different protocols or modalities, such as more commonly used T 1 - or T 2 -weighted MRI sequences, would require validation of the ability of our atlases to accurately align with these new modalities. Multi-centre, multi-modal validation studies could determine the limits and capabilities of the atlas-based approach when applied to broader datasets. Additionally, the utility of atlases might be enhanced by adaptive algorithms that can fine-tune the registration parameters based on the specific characteristics of the input data. By incorporating machine learning techniques or advanced image processing algorithms, it may be possible to automate the adjustment of these parameters, thereby improving the adaptability and accuracy of our atlases across diverse imaging environments. Declarations Conflict of interest statement I hereby state that none of the authors have had any financial or personal relationships with other people or organizations that could inappropriately influence (bias) our work. Author Contribution Yilan Zhang: Conceptualisation, methodology, software, validation, formal analysis, investigation, resources, data curation, visualisation, writing – original draft, writing – review and editing, project administration.Rob Lloyd: Conceptualisation, validation, writing - review and editing, supervision.Robert D. Herbert: Conceptualisation, methodology, writing – review and editing, project administration, supervision.Lynne E. Bilston: Conceptualisation, validation, writing - review and editing, supervision.Bart Bolsterlee: Conceptualisation, methodology, software, validation, formal analysis, investigation, resources, data curation, visualisation, writing – review and editing, project administration, supervision. Acknowledgements This study was supported by the Australian Research Council through the Industrial Transformation Training Centre Program for Joint Biomechanics (IC190100020). Y. Zhang is supported by a UNSW Tuition Fee Scholarship. L. Bilston is supported by an NHMRC Investigator grant, (1172988). The authors acknowledge the facilities and scientific and technical assistance of the National Imaging Facility, a National Collaborative Research Infrastructure Strategy (NCRIS) capability, at Neuroscience Research Australia and UNSW. References Alipour, M., Mithraratne, K., & Fernandez, J. (2017). A diffusion-weighted imaging informed continuum model of the rabbit triceps surae complex. Biomechanics and Modeling in Mechanobiology , 16 (5), 1729–1741. https://doi.org/10.1007/s10237-017-0916-4 Alonso-Fernandez, D., Docampo-Blanco, P., & Martinez-Fernandez, J. (2018). Changes in muscle architecture of biceps femoris induced by eccentric strength training with nordic hamstring exercise. Scandinavian Journal of Medicine & Science in Sports , 28 (1), 88–94. https://doi.org/10.1111/sms.12877 Anderson, A. W. (2005). Measurement of fiber orientation distributions using high angular resolution diffusion imaging. Magnetic Resonance in Medicine , 54 (5), 1194–1206. https://doi.org/10.1002/mrm.20667 Andersson, J. L. R., & Sotiropoulos, S. N. (2016). An integrated approach to correction for off-resonance effects and subject movement in diffusion MR imaging. NeuroImage , 125 , 1063–1078. https://doi.org/10.1016/j.neuroimage.2015.10.019 Avants, B., Duda, J. T., Zhang, H., & Gee, J. C. (2007). Multivariate Normalization with Symmetric Diffeomorphisms for Multivariate Studies. In N. Ayache, S. Ourselin, & A. Maeder (Eds.), Medical Image Computing and Computer-Assisted Intervention – MICCAI 2007 (Vol. 4791, pp. 359–366). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-540-75757-3_44 Bin Ghouth, S. G., Williams, S. A., Reid, S. L., Besier, T. F., & Handsfield, G. G. (2022). A statistical shape model of soleus muscle morphology in spastic cerebral palsy. Scientific Reports , 12 (1), 7711. https://doi.org/10.1038/s41598-022-11611-z Blazevich, A. J., Gill, N. D., Bronks, R., & Newton, R. U. (2003). Training-Specific Muscle Architecture Adaptation after 5-wk Training in Athletes: Medicine & Science in Sports & Exercise , 35 (12), 2013–2022. https://doi.org/10.1249/01.MSS.0000099092.83611.20 Bodine, S. C., Roy, R. R., Meadows, D. A., Zernicke, R. F., Sacks, R. D., Fournier, M., & Edgerton, V. R. (1982). Architectural, histochemical, and contractile characteristics of a unique biarticular muscle: The cat semitendinosus. Journal of Neurophysiology , 48 (1), 192–201. https://doi.org/10.1152/jn.1982.48.1.192 Bolsterlee, B. (2022). A new framework for analysis of three-dimensional shape and architecture of human skeletal muscles from in vivo imaging data. Journal of Applied Physiology , 132 (3), 712–725. https://doi.org/10.1152/japplphysiol.00638.2021 Büchler, P., & Farron, A. (2004). Benefits of an anatomical reconstruction of the humeral head during shoulder arthroplasty: A finite element analysis. Clinical Biomechanics , 19 (1), 16–23. https://doi.org/10.1016/j.clinbiomech.2003.09.009 Dice, L. R. (1945). Measures of the Amount of Ecologic Association Between Species. Ecology , 26 (3), 297–302. https://doi.org/10.2307/1932409 Fahn-Lai, P., Biewener, A. A., & Pierce, S. E. (2020). Broad similarities in shoulder muscle architecture and organization across two amniotes: Implications for reconstructing non-mammalian synapsids. PeerJ , 8 , e8556. https://doi.org/10.7717/peerj.8556 Fedorov, A., Beichel, R., Kalpathy-Cramer, J., Finet, J., Fillion-Robin, J.-C., Pujol, S., Bauer, C., Jennings, D., Fennessy, F., Sonka, M., Buatti, J., Aylward, S., Miller, J. V., Pieper, S., & Kikinis, R. (2012). 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magnetic Resonance Imaging , 30 (9), 1323–1341. https://doi.org/10.1016/j.mri.2012.05.001 Forsberg, D., Rathi, Y., Bouix, S., Wassermann, D., Knutsson, H., & Westin, C.-F. (2011). Improving Registration Using Multi-channel Diffeomorphic Demons Combined with Certainty Maps. In T. Liu, D. Shen, L. Ibanez, & X. Tao (Eds.), Multimodal Brain Image Analysis (Vol. 7012, pp. 19–26). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-24446-9_3 Gazielly, D. F., Gleyze, P., & Montagnon, C. (1994). Functional and anatomical results after rotator cuff repair. Clinical Orthopaedics and Related Research , 304 , 43–53. Gerber, C., Fuchs, B., & Hodler, J. (2000). The results of repair of massive tears of the rotator cuff. JBJS , 82 (4), 505. Hotelling, H. (1933). Analysis of a complex of statistical variables into principal components. Journal of Educational Psychology , 24 (6), 417–441. https://doi.org/10.1037/h0071325 Isensee, F., Jaeger, P. F., Kohl, S. A. A., Petersen, J., & Maier-Hein, K. H. (2021). nnU-Net: A self-configuring method for deep learning-based biomedical image segmentation. Nature Methods , 18 (2), 203–211. https://doi.org/10.1038/s41592-020-01008-z Jenkinson, M., Bannister, P., Brady, M., & Smith, S. (2002). Improved optimization for the robust and accurate linear registration and motion correction of brain images. NeuroImage , 17 (2), 825–841. https://doi.org/10.1016/s1053-8119(02)91132-8 Jenkinson, M., & Smith, S. (2001). A global optimisation method for robust affine registration of brain images. Medical Image Analysis , 5 (2), 143–156. https://doi.org/10.1016/S1361-8415(01)00036-6 Jones, D. K., Griffin, L. D., Alexander, D. C., Catani, M., Horsfield, M. A., Howard, R., & Williams, S. C. R. (2002). Spatial Normalization and Averaging of Diffusion Tensor MRI Data Sets. NeuroImage , 17 (2), 592–617. https://doi.org/10.1006/nimg.2002.1148 Khandare, S., Arce, R. A., & Vidt, M. E. (2022). Muscle compensation strategies to maintain glenohumeral joint stability with increased rotator cuff tear severity: A simulation study. Journal of Electromyography and Kinesiology , 62 , 102335. https://doi.org/10.1016/j.jelekin.2019.07.005 Kruse, A., Schranz, C., Tilp, M., & Svehlik, M. (2018). Muscle and tendon morphology alterations in children and adolescents with mild forms of spastic cerebral palsy. BMC Pediatrics , 18 (1), 156. https://doi.org/10.1186/s12887-018-1129-4 Lieber, R. L., & Fridén, J. (2000). Functional and clinical significance of skeletal muscle architecture. Muscle & Nerve , 23 (11), 1647–1666. https://doi.org/10.1002/1097-4598(200011)23:113.0.co;2-m Narici, M. V., Maganaris, C. N., Reeves, N. D., & Capodaglio, P. (2003). Effect of aging on human muscle architecture. Journal of Applied Physiology , 95 (6), 2229–2234. https://doi.org/10.1152/japplphysiol.00433.2003 Papenkort, S., Böl, M., & Siebert, T. (2021). Architectural model for muscle growth during maturation. Biomechanics and Modeling in Mechanobiology , 20 (5), 2031–2044. https://doi.org/10.1007/s10237-021-01492-y Park, H. (2003). Spatial normalization of diffusion tensor MRI using multiple channels. NeuroImage , 20 (4), 1995–2009. https://doi.org/10.1016/j.neuroimage.2003.08.008 Park, J. E., Seong, Y.-J., Kim, E. S., Park, D., Lee, Y., Park, H., & Rha, D. (2019). Architectural Changes in the Medial Gastrocnemius on Sonography after Nerve Ablation in Healthy Adults. Yonsei Medical Journal , 60 (9), 876. https://doi.org/10.3349/ymj.2019.60.9.876 Pearson, K. (1901). On lines and planes of closest fit to systems of points in space. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science , 2 (11), 559–572. https://doi.org/10.1080/14786440109462720 Prinold, J. A., Masjedi, M., Johnson, G. R., & Bull, A. M. (2013). Musculoskeletal shoulder models: A technical review and proposals for research foci. Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine , 227 (10), 1041–1057. https://doi.org/10.1177/0954411913492303 Raffelt, D., Tournier, J., Crozier, S., Connelly, A., & Salvado, O. (2012). Reorientation of fiber orientation distributions using apodized point spread functions. Magnetic Resonance in Medicine , 67 (3), 844–855. https://doi.org/10.1002/mrm.23058 Roig, M., Shadgan, B., & Reid, W. D. (2008). Eccentric Exercise in Patients with Chronic Health Conditions: A Systematic Review. Physiotherapy Canada , 60 (2), 146–160. https://doi.org/10.3138/physio.60.2.146 Roura, E. (2015). Multi-channel registration of fractional anisotropy and T1-weighted images in the presence of atrophy: Application to multiple sclerosis. Functional Neurology . https://doi.org/10.11138/FNeur/2015.30.4.245 Salhi, A., Burdin, V., Boutillon, A., Brochard, S., Mutsvangwa, T., & Borotikar, B. (2020). Statistical Shape Modeling Approach to Predict Missing Scapular Bone. Annals of Biomedical Engineering , 48 (1), 367–379. https://doi.org/10.1007/s10439-019-02354-6 Shur, N. F., Creedon, L., Skirrow, S., Atherton, P. J., MacDonald, I. A., Lund, J., & Greenhaff, P. L. (2021). Age-related changes in muscle architecture and metabolism in humans: The likely contribution of physical inactivity to age-related functional decline. Ageing Research Reviews , 68 , 101344. https://doi.org/10.1016/j.arr.2021.101344 Siebert, T., Tomalka, A., Stutzig, N., Leichsenring, K., & Böl, M. (2017). Changes in three-dimensional muscle structure of rabbit gastrocnemius, flexor digitorum longus, and tibialis anterior during growth. Journal of the Mechanical Behavior of Biomedical Materials , 74 , 507–519. https://doi.org/10.1016/j.jmbbm.2017.07.045 Smith, R. E., Tournier, J.-D., Calamante, F., & Connelly, A. (2015). SIFT2: Enabling dense quantitative assessment of brain white matter connectivity using streamlines tractography. NeuroImage , 119 , 338–351. https://doi.org/10.1016/j.neuroimage.2015.06.092 Su-Lin Lee, Tan, E., Khullar, V., Gedroyc, W., Darzi, A., & Guang-Zhong Yang. (2009). Physical-Based Statistical Shape Modeling of the Levator Ani. IEEE Transactions on Medical Imaging , 28 (6), 926–936. https://doi.org/10.1109/TMI.2009.2012894 Sutherland, A. M. T., Lynch, J. T., Serpell, B. G., Pickering, M. R., Newman, P., Perriman, D. M., & Kenneally-Dabrowski, C. (2023). Statistical shape modelling reveals differences in hamstring morphology between professional rugby players and sprinters. Journal of Sports Sciences , 41 (2), 164–171. https://doi.org/10.1080/02640414.2023.2204269 Thevenaz, P., Ruttimann, U. E., & Unser, M. (1998). A pyramid approach to subpixel registration based on intensity. IEEE Transactions on Image Processing , 7 (1), 27–41. https://doi.org/10.1109/83.650848 Tournier, J.-D., Calamante, F., & Connelly, A. (2007). Robust determination of the fibre orientation distribution in diffusion MRI: Non-negativity constrained super-resolved spherical deconvolution. NeuroImage , 35 (4), 1459–1472. https://doi.org/10.1016/j.neuroimage.2007.02.016 Tournier, J.-D., Calamante, F., & Connelly, A. (2012). MRtrix: Diffusion tractography in crossing fiber regions. International Journal of Imaging Systems and Technology , 22 (1), 53–66. https://doi.org/10.1002/ima.22005 Tournier, J.-D., Calamante, F., Gadian, D. G., & Connelly, A. (2004). Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution. NeuroImage , 23 (3), 1176–1185. https://doi.org/10.1016/j.neuroimage.2004.07.037 Tournier, J.-D., Smith, R., Raffelt, D., Tabbara, R., Dhollander, T., Pietsch, M., Christiaens, D., Jeurissen, B., Yeh, C.-H., & Connelly, A. (2019). MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation. NeuroImage , 202 , 116137. https://doi.org/10.1016/j.neuroimage.2019.116137 Tran, V.-D., Nguyen, T.-N., Ballit, A., & Dao, T.-T. (2023). Novel Baseline Facial Muscle Database Using Statistical Shape Modeling and In Silico Trials toward Decision Support for Facial Rehabilitation. Bioengineering , 10 (6), 737. https://doi.org/10.3390/bioengineering10060737 Uus, A., Grigorescu, I., Pietsch, M., Batalle, D., Christiaens, D., Hughes, E., Hutter, J., Cordero Grande, L., Price, A. N., Tournier, J.-D., Rutherford, M. A., Counsell, S. J., Hajnal, J. V., Edwards, A. D., & Deprez, M. (2021). Multi-Channel 4D Parametrized Atlas of Macro- and Microstructural Neonatal Brain Development. Frontiers in Neuroscience , 15 , 661704. https://doi.org/10.3389/fnins.2021.661704 Uus, A., Pietsch, M., Grigorescu, I., Christiaens, D., Tournier, J.-D., Grande, L. C., Hutter, J., Edwards, D., Hajnal, J., & Deprez, M. (2020). Multi-channel Registration for Diffusion MRI: Longitudinal Analysis for the Neonatal Brain. In Ž. Špiclin, J. McClelland, J. Kybic, & O. Goksel (Eds.), Biomedical Image Registration (Vol. 12120, pp. 111–121). Springer International Publishing. https://doi.org/10.1007/978-3-030-50120-4_11 Veraart, J., Fieremans, E., & Novikov, D. S. (2016). Diffusion MRI noise mapping using random matrix theory: Diffusion MRI Noise Mapping. Magnetic Resonance in Medicine , 76 (5), 1582–1593. https://doi.org/10.1002/mrm.26059 Veraart, J., Novikov, D. S., Christiaens, D., Ades-aron, B., Sijbers, J., & Fieremans, E. (2016). Denoising of diffusion MRI using random matrix theory. NeuroImage , 142 , 394–406. https://doi.org/10.1016/j.neuroimage.2016.08.016 Vidt, M. E., Santago, A. C., Marsh, A. P., Hegedus, E. J., Tuohy, C. J., Poehling, G. G., Freehill, M. T., Miller, M. E., & Saul, K. R. (2018). Modeling a rotator cuff tear: Individualized shoulder muscle forces influence glenohumeral joint contact force predictions. Clinical Biomechanics (Bristol, Avon) , 60 , 20–29. https://doi.org/10.1016/j.clinbiomech.2018.10.004 Vlachopoulos, L., Lüthi, M., Carrillo, F., Gerber, C., Székely, G., & Fürnstahl, P. (2018). Restoration of the Patient-Specific Anatomy of the Proximal and Distal Parts of the Humerus: Statistical Shape Modeling Versus Contralateral Registration Method. Journal of Bone and Joint Surgery , 100 (8), e50. https://doi.org/10.2106/JBJS.17.00829 Yushkevich, P. A., Piven, J., Hazlett, H. C., Smith, R. G., Ho, S., Gee, J. C., & Gerig, G. (2006). User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability. NeuroImage , 31 (3), 1116–1128. https://doi.org/10.1016/j.neuroimage.2006.01.015 Zhang, Y., Herbert, R. D., Bilston, L. E., & Bolsterlee, B. (2023). Three-dimensional architecture of the human subscapularis muscle in vivo. Journal of Biomechanics , 161 , 111854. https://doi.org/10.1016/j.jbiomech.2023.111854 Zhang, Y., Herbert, R. D., Bilston, L. E., & Bolsterlee, B. (2024). Three‐dimensional architecture and moment arms of human rotator cuff muscles in vivo: Interindividual, intermuscular, and intramuscular variations. Journal of Anatomy , joa.14050. https://doi.org/10.1111/joa.14050 Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4683327","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":331415975,"identity":"20f51030-e685-4a7e-9fb3-81cf3d7f93b7","order_by":0,"name":"Yilan Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCElEQVRIie3PsUrDQBzH8QsHyXJH1n8RzCscHISCL5NDqEsrhYJ0kPRfCtfNzqK+Q4tQOiYc1CUP0DFQcKpQl04iXi0ODicZHe47/uHDjz8hPt9/LCkntfqAUUyI+LmJvwAhQLV4x3bWwuYkwtY9DjNRNCXJwxglX8G1fDFzyle5wmiyBHJrnCR4LHHLKxikVadPeWUUsvUNkLWbUKJQshCCZdEVlOuih9BNgYRuElpydiTPs92R5D1MdpZ8ugkDZd/XoObwvULtCksh0G4CrNRiX4GEzWu/fNJmpFln0FZ3V06STKfbOhvm5/HsclG/6VzGkVls9ocLJ/lVcfrOljUDPp/P53P0BQ7yVDssrq8TAAAAAElFTkSuQmCC","orcid":"","institution":"University of New South Wales","correspondingAuthor":true,"prefix":"","firstName":"Yilan","middleName":"","lastName":"Zhang","suffix":""},{"id":331415976,"identity":"43c9f794-b253-488d-89df-fa05a8d17e64","order_by":1,"name":"Robert Lloyd","email":"","orcid":"","institution":"Neuroscience Research Australia","correspondingAuthor":false,"prefix":"","firstName":"Robert","middleName":"","lastName":"Lloyd","suffix":""},{"id":331415977,"identity":"7427147c-861e-4605-8975-343b6a0fcfcf","order_by":2,"name":"Robert D. Herbert","email":"","orcid":"","institution":"Neuroscience Research Australia","correspondingAuthor":false,"prefix":"","firstName":"Robert","middleName":"D.","lastName":"Herbert","suffix":""},{"id":331415978,"identity":"8827ba04-0b2f-4f42-8172-4bc4ee9d4037","order_by":3,"name":"Lynne E. Bilston","email":"","orcid":"","institution":"Neuroscience Research Australia","correspondingAuthor":false,"prefix":"","firstName":"Lynne","middleName":"E.","lastName":"Bilston","suffix":""},{"id":331415979,"identity":"38945b6a-7a0a-40a6-ac3b-24d8a1733446","order_by":4,"name":"Bart Bolsterlee","email":"","orcid":"","institution":"Neuroscience Research Australia","correspondingAuthor":false,"prefix":"","firstName":"Bart","middleName":"","lastName":"Bolsterlee","suffix":""}],"badges":[],"createdAt":"2024-07-04 02:53:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4683327/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4683327/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":61358252,"identity":"268c66f1-faa1-4bbe-8262-9a9f37061d02","added_by":"auto","created_at":"2024-07-29 21:18:13","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":198829,"visible":true,"origin":"","legend":"\u003cp\u003eSee image above for figure legend\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/684b52cc1f7d69e01a31d0e1.png"},{"id":61359059,"identity":"c21426ca-0960-4f09-800c-285bda7b4ae8","added_by":"auto","created_at":"2024-07-29 21:26:13","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":586435,"visible":true,"origin":"","legend":"\u003cp\u003eExamples of one male (top) and one female (bottom) multi-channel atlas. There are two rows for the male and two rows for the female: the upper and lower rows illustrate transverse mDixon slices approximately midway through the acromioclavicular and glenohumeral joint, respectively. Shown from left to right are: the mDixon water image, masks of rotator cuff muscles and scapula overlayed on mDixon image (blue, scapula; beige, subscapularis; pink, infraspinatus and teres minor; orange, supraspinatus), and an FOD image overlayed on mDixon. FODs are colour-coded according to orientation (red: left-right; green: anterior-posterior; blue: inferior-superior).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/a0487c85c042e79c4be251e1.png"},{"id":61358253,"identity":"868402bd-fa86-41b6-8dd6-95f1ca87327a","added_by":"auto","created_at":"2024-07-29 21:18:13","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":714683,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of 3D fibre tractography reconstructions of rotator cuff muscles from the right shoulder, using data from an atlas generated from 10 male subjects. For each muscle, the illustration showcases the tractographies for subjects with the (a) best, (b) median, and (c) worst angular alignment within this cohort. Each pair highlights the fibre reconstructions set against the 3D surface model of the muscle in transparent yellow; the top and bottom row displays fibre reconstructed from the subject’s original FODs and the FOD atlas, respectively. Fibre tracts are colour-coded according to orientation (red: medial-lateral; green: anterior-posterior; blue: inferior-superior).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/38fd397cc28811b56ea1da27.png"},{"id":61359060,"identity":"cbbaa252-b59b-47d4-b448-e7238b7f78f4","added_by":"auto","created_at":"2024-07-29 21:26:13","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":31838,"visible":true,"origin":"","legend":"\u003cp\u003eRegistration performance metrics across different muscles for the male (in blue) and female (in red) cohorts. The bars indicate the average of the median (a) Dice coefficients, (b) ACCs, and (c) angular differences across each muscle for every subject included in the atlas (males: n = 11; females: n = 9).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/ab2eceff54b1bc52be3ff9a1.png"},{"id":61358256,"identity":"7fedc763-9b1b-4d56-bd56-e44bf46d719d","added_by":"auto","created_at":"2024-07-29 21:18:13","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":454163,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative 3D fibre tractography reconstructions of rotator cuff muscles from the right shoulder of an out-of-sample subject in the male cohort. Fibres reconstructed from the subject’s (a) original and (b) predicted fibre orientations colour-coded according to orientation (red: medial-lateral; green: anterior-posterior; blue: inferior-superior), set against 3D surface models of rotator cuff muscles in transparent yellow. In panel (a), anatomically implausible fibre tracts and regions with missing data are highlighted with blue and red circles, respectively. Panel (c) overlays fibre from the subject’s original fibre orientations (in transparent red) with those from the predicted fibre orientations (in transparent blue) for direct visual comparison.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/2e4d630053619b182195a1c1.png"},{"id":61358257,"identity":"6378ec55-adf3-4e28-99de-20581c0194e9","added_by":"auto","created_at":"2024-07-29 21:18:13","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":89534,"visible":true,"origin":"","legend":"\u003cp\u003eScattergram of the metrics used to evaluate the fibre orientation prediction for out-of-sample subjects for the male (in blue) and female (in red) cohorts. Each data point represents the median (a) Dice coefficients, (b) ACCs, and (c) angular differences across each muscle for every out-of-sample subject of the atlas (males: n = 11; females: n = 9).\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/7602baeae99c97f0ac5361af.png"},{"id":61358255,"identity":"538adc73-bb60-456b-81e2-00b56dda7101","added_by":"auto","created_at":"2024-07-29 21:18:13","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":249200,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative visualisation of fibre orientation prediction accuracy for an out-of-sample subject on transverse slices. a and c illustrate the ACCs for the supraspinatus (a) and the combined infraspinatus, teres minor and subscapularis (c). b and d present the angular differences for the same muscle groups. The metrics were calculated between the original and predicted fibre orientations for the out-of-sample subject.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/bef81ed16f56b7f1a8fa8d66.png"},{"id":65210918,"identity":"dc3ffc5d-fccb-41bf-9b8a-afb5ba31b861","added_by":"auto","created_at":"2024-09-24 20:08:35","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3130267,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/d6bf590b-a157-4c27-a6b9-7b764744936a.pdf"},{"id":61358258,"identity":"588aac2c-58ae-4ff0-b6b1-92342a31feee","added_by":"auto","created_at":"2024-07-29 21:18:13","extension":"docx","order_by":11,"title":"","display":"","copyAsset":false,"role":"supplement","size":3126724,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-4683327/v1/1bc9ed9e20ca69c3c30dacc2.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Prediction of rotator cuff muscle fibre orientations using a population-averaged atlas generated with anatomical and diffusion-weighted magnetic resonance images","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eSkeletal muscle architecture, the macrostructural arrangement of muscle fibres in the muscle belly, is the primary determinant of a muscle\u0026rsquo;s capacity to generate force and change length (Lieber \u0026amp; Frid\u0026eacute;n, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2000\u003c/span\u003e). While the architecture of any particular muscle is broadly similar across individuals, muscle architecture can adapt to exercise (Alonso-Fernandez et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Blazevich et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Roig et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2008\u003c/span\u003e), ageing (Narici et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Papenkort et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Shur et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Siebert et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) and disease (Bodine et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e1982\u003c/span\u003e; Fahn-Lai et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Kruse et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; J. E. Park et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Quantitative measurement of muscle- and subject-specific architecture is therefore useful for the study of muscle adaptation and muscle dysfunction.\u003c/p\u003e \u003cp\u003eThe human rotator cuff, comprising the supraspinatus, subscapularis, infraspinatus, and teres minor muscles, plays a significant role in movement of the upper limb and provides dynamic stability to the glenohumeral joint. Rotator cuff injuries are common \u0026ndash; they constitute up to half of all significant shoulder injuries seen in some clinical contexts (Gazielly et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e1994\u003c/span\u003e; Gerber et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2000\u003c/span\u003e) \u0026ndash; and can be difficult to treat. Surgical treatment of rotator cuff injuries might be improved with preoperative planning tools built around biomechanical models of the shoulder. Computational models can be used to predict the biomechanical consequences of rotator cuff tears on muscle function and joint forces (Khandare et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Vidt et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) or guide the design of implants in shoulder arthroplasty procedures (B\u0026uuml;chler \u0026amp; Farron, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). However, due to the difficulty in obtaining subject-specific measurements of rotator cuff muscle architecture, most musculoskeletal shoulder models are generic or scaled-generic models (Prinold et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2013\u003c/span\u003e), which, despite their utility, often fall short of capturing the anatomical variations unique to each individual. Computational models and surgical planning tools might be more useful if they could incorporate subject-specific measurements of rotator cuff muscle architecture.\u003c/p\u003e \u003cp\u003eDiffusion-weighted imaging (DWI) enables accurate measurement of human rotator cuff muscle architecture in three dimensions, using fibre tracking algorithms which propagate fibre tracts along the principal eigenvectors of diffusion tensors throughout a muscle (Zhang et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This approach not only provides a visual representation of muscle fibres orientations, but also allows for the quantification of muscle architecture such as the fascicle length and pennation angle \u0026ndash; parameters that are crucial in understanding muscle function but cannot be measured with commonly used anatomical MRI scans. However, DWI-based reconstructions of muscle architecture are sensitive to noise and image artifacts inherent in DWI data, which can distort the path of fibre tracts and create regions in the muscle that contain few fibres. New methods are needed to improve muscle architecture reconstructions from MRI data. Ideally such methods would use standard scanning protocols and automated image processing procedures so that the methods could be routinely implemented in clinical practice.\u003c/p\u003e \u003cp\u003eSome of the current limitations of DWI-based muscle architecture reconstructions can be addressed with a population-based approach that characterises inter-individual variability. Statistical shape modelling (SSM) is now commonly used for this purpose. After computing a mean shape across the population and transforming each individual\u0026rsquo;s shape to the mean, principal component analysis can be used to identify the main modes of inter-individual shape variation (Hotelling, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e1933\u003c/span\u003e; Pearson, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e1901\u003c/span\u003e). SSMs have been used predominantly to study human skeletal anatomy (e.g., humerus (Vlachopoulos et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), scapula (Salhi et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020\u003c/span\u003e)). However, there have been relatively few attempts to apply SSMs to skeletal muscles (e.g., facial muscles (Tran et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), levator ani (Su-Lin Lee et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2009\u003c/span\u003e), soleus (Bin Ghouth et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) and hamstrings (Sutherland et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)). To the best of our knowledge, no studies have applied SSM or population-based methods to shoulder muscles. Moreover, existing population-based methods usually only model the shape (outer surface) of the muscle, but do not model internal fibre orientations. One recent study included model fibre orientations in a population-averaged muscle modelling framework (Bolsterlee, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). However, that study only used surface features to find corresponding points between muscles from different individuals. SSMs might be able to more accurately represent the internal architecture of muscles if, in addition to using information about the shape of the muscle surface, they used information about muscle fibre orientations. This could be achieved by registering multiple channels of information \u0026ndash; specifically, by combining muscle surface data derived from anatomical MRI with fibre orientation data derived from DWI. Registration of fibre orientations and development of DWI-based population-averaged atlases have been a focus in neuroimaging (Forsberg et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Roura, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Uus et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2020\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), but to our knowledge those methods have not yet been used in muscle imaging.\u003c/p\u003e \u003cp\u003ePopulation-averaged atlases built using multi-channel registration introduce novel possibilities for muscle architecture analysis. Firstly, atlases may enhance the representation of individual muscle architecture by aggregating information across the population. Aggregation can potentially reduce noise and image artifacts typically present in individual scans and enhance the robustness and reliability of muscle architecture reconstructions. Secondly, since previous studies on rotator cuff muscle architecture have reported limited interindividual variability (Zhang et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), there is potential that atlases can be used to predict individual muscle architecture from muscle shape alone, simplifying the generation of high-quality muscle architecture reconstructions for musculoskeletal modelling and assessments.\u003c/p\u003e \u003cp\u003eThe goals of this study were, therefore, to: 1) develop population-averaged atlases of human rotator cuff muscles using multi-channel registration that combines anatomical data derived from MRI with fibre orientation data derived from DWI; and 2) use the atlases to predict muscle fibre orientations from anatomical MRI scans. We hypothesised that fibre orientations can be predicted accurately without DWI by registering anatomical MRI from a new subject to the anatomical MRI channels of the population-averaged atlas and then applying the resulting transformation to the channels of the atlas that encode muscle fibre orientations.\u003c/p\u003e"},{"header":"2 Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003e2.1 Participants\u003c/h2\u003e\n \u003cp\u003eStudy procedures were approved by the UNSW Human Research Ethics Committee (HREC approval HC200971). Each participant was informed about the study procedures and provided their written consent before participation. The study involved magnetic resonance imaging (MRI) of the right shoulder of 20 participants (11 males and 9 females; age 28\u0026thinsp;\u0026plusmn;\u0026thinsp;6 years; height 171\u0026thinsp;\u0026plusmn;\u0026thinsp;8 cm; weight 64\u0026thinsp;\u0026plusmn;\u0026thinsp;11 kg, values are mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation). People with symptoms or recent history of shoulder pathology were excluded from participating.\u003c/p\u003e\n \u003cp\u003eSome of the data used in this study have been used in previous studies on rotator cuff muscle architecture (Zhang et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n \u003ch2\u003e2.2 MRI acquisitions and processing\u003c/h2\u003e\n \u003cp\u003eA brief overview of image acquisition and processing will be provided here. A more detailed description can be found elsewhere (Zhang et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eAll MR images were acquired at 3T (Philips Ingenia CX, Philips Healthcare, Best, The Netherlands) using a 16-channel anterior body coil and a posterior coil integrated in the scanner bed. The protocol consisted of an mDixon scan (field of view 240 mm, voxel size 1 \u0026times; 1 \u0026times; 2 mm, 210 slices) and two diffusion-weighted scans (field of view 190 mm, voxel size 2.5 \u0026times; 2.5 \u0026times; 5 mm, 24 slices) covering the proximal and distal rotator cuff musculature, respectively. Segmentation of rotator cuff muscles and bones (humerus, scapula and clavicle) was initially performed manually on a subset of 12 mDixon scans, after which a deep learning model (nnU-net, (Isensee et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e)) was trained for automatic segmentation of the remaining scans. The predicted segmentations were visually verified and adjusted when needed. In four out of twenty scans the boundary between the infraspinatus and teres minor muscles was not clearly visible. These muscles were therefore grouped together for all participants. Previous analysis demonstrated excellent intra-rater reliability of segmentation (average intraclass correlation coefficient across muscles of 0.97; (Zhang et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e)).\u003c/p\u003e\n \u003cp\u003eDWI scans were filtered with a Marchenko-Pastur principal component analysis filter (Veraart, Fieremans, et al., 2016; Veraart, Novikov, et al., \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e) to reduce image noise, and then corrected for eddy current- and possible motion-induced distortions using functions TOPUP and EDDY from FSL (Andersson \u0026amp; Sotiropoulos, \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e), implemented in MRtrix (MRtrix3; (Tournier et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e)). The processed DWI scans were then upsampled to match the spatial dimensions of the mDixon scan and combined into a single DWI image set covering the entire rotator cuff. Rigid registration via FLIRT tool from FSL (Jenkinson et al., \u003cspan class=\"CitationRef\"\u003e2002\u003c/span\u003e; Jenkinson \u0026amp; Smith, \u003cspan class=\"CitationRef\"\u003e2001\u003c/span\u003e) was performed to correct for small misalignments between the mDixon and DWI images within subjects. The accuracy of registration was confirmed visually in ITK-SNAP (Yushkevich et al., \u003cspan class=\"CitationRef\"\u003e2006\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eTo increase anatomical contrast and improve subsequent image registration, the mDixon scans, stitched DWI scans, and segmented masks were linearly upsampled to an isotropic voxel size of 0.75 \u0026times; 0.75 \u0026times; 0.75 mm.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\n \u003ch2\u003e2.3 Construction of population-averaged atlases\u003c/h2\u003e\n \u003cp\u003eThe construction of population-averaged atlases was a multi-step process involving estimation of fibre orientation distributions (FODs) from DWI data, concurrent multi-channel registration of individual mDixon images and the FODs into a common reference frame, and averaging across subjects to construct a cohesive atlas. Population-averaged atlases were created separately for males (n\u0026thinsp;=\u0026thinsp;11) and females (n\u0026thinsp;=\u0026thinsp;9). In the initial phase of this study we tried to construct a single multi-channel population-averaged atlas for both males and females. However, the large differences in muscle volumes between males and females led to difficulties in achieving accurate alignment, so therefore we created create sex-specific atlases.\u003c/p\u003e\n \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\n \u003ch2\u003e2.3.1 Fibre orientation distribution estimation from DWI\u003c/h2\u003e\n \u003cp\u003eFODs were derived from DWI data using algorithms included in MRtrix (Tournier et al., \u003cspan class=\"CitationRef\"\u003e2004\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e). While FODs can describe multiple fibre orientations within a voxel, previous investigations have shown that fibres within rotator cuff muscles do not cross each other so that muscle fibre orientations within a voxel can be represented well by a single orientation (Zhang et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). We therefore limited the maximum spherical harmonics (SH) degree to two in the FOD estimation process, retaining only the principal diffusion direction within a voxel, analogous to a single-fibre diffusion tensor model. The response function (Tournier et al., \u003cspan class=\"CitationRef\"\u003e2004\u003c/span\u003e), which describes the diffusion signal intensity based on the orientation of a single fibre bundle when exposed to magnetic gradients, was used as a kernel for spherical deconvolution. The FOD estimation was then performed on DWI images with constrained spherical deconvolution using the response function as an input (Tournier et al., \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e). Segmented muscle masks were used to restrict FOD estimation to the rotator cuff muscles.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\n \u003ch2\u003e2.3.2 Multi-channel registration\u003c/h2\u003e\n \u003cp\u003eData from n-1 subjects (n\u0026thinsp;=\u0026thinsp;10 for males, n\u0026thinsp;=\u0026thinsp;8 for females) was used for the creation of each population-averaged atlas. For both male and female atlases, one subject was left out for out-of-sample evaluation (see \u003cspan class=\"InternalRef\"\u003eEvaluation\u003c/span\u003e Section).\u003c/p\u003e\n \u003cp\u003eThe registration pipeline is based on an intensity-based multi-channel registration framework implemented in MRtrix. The input channels for each subject consist of all four mDixon images (water only, fat only, in-phase, and out-of-phase) and six FOD image volumes. Masks of rotator cuff muscles and the scapula were used as inputs to speed up registration and to target the registration to the primary region of interest, although they did not contribute to the computation of displacement fields necessary for aligning the images. The pipeline consists of two steps: initialisation and iterative registration (see Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). All registrations used symmetric normalisation (SyN) Demons (Avants et al., \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e) with the sum of squared difference (SSD) metric and reorientation of FOD using apodised point spread functions (Raffelt et al., \u003cspan class=\"CitationRef\"\u003e2012\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eInitialisation\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eThe initialisation step created an initial atlas that is unbiased to any individual subject in the population, serving as a reference for the subsequent registration process. First, the average coordinate space (or image grid) of all input FODs was calculated. Then, for each subject, the mDixon \u003cem\u003eI\u003c/em\u003e and FOD \u003cem\u003eJ\u003c/em\u003e channels were rigidly registered to the averaged coordinate space, respectively. Each channel\u0026rsquo;s transformations were then averaged and applied to the mDixon and FOD data. The preliminary set of mDixon \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{0}^{I}\\)\u003c/span\u003e\u003c/span\u003e and FOD \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{0}^{J}\\)\u003c/span\u003e\u003c/span\u003e atlases was then generated by computing the median image intensity across all registered mDixon \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({I}_{0}\\)\u003c/span\u003e\u003c/span\u003e and FOD \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({J}_{0}\\)\u003c/span\u003e\u003c/span\u003e volumes, respectively. We empirically found that using the median instead of the mean of image intensities gave clearer structure boundaries.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eIterative registration\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eUnbiased population-averaged atlases were constructed through a process of iterative registration. In each iteration, the median intensity values of the registered mDixon and FOD volumes were computed separately to continually update and refine their respective atlases. This process involved iterating over 28 stages, organised into a sequence of 6 rigid, 6 affine, and 16 non-linear registration stages. The atlas generation at each stage \u003cem\u003ek\u003c/em\u003e can be summarised in the following steps:\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e1) For each subject, mDixon \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({I}_{k-1}\\)\u003c/span\u003e\u003c/span\u003e (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\\in\\)\u003c/span\u003e\u003c/span\u003e [1,28]) and FOD \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({J}_{k-1}\\)\u003c/span\u003e\u003c/span\u003e channels were registered to their corresponding updated atlases \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{k-1}^{I}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{k-1}^{J}\\)\u003c/span\u003e\u003c/span\u003e (to preliminary atlases \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{0}\\)\u003c/span\u003e\u003c/span\u003e in the first stage) using rigid, affine or non-linear registration based on the stage.\u003c/p\u003e\n \u003c/span\u003e \u003cspan\u003e\n \u003cp\u003e2) Resulting transformations (rigid and affine registrations) or displacement fields (non-linear registration) from each channel were averaged (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\varnothing\\)\u003c/span\u003e\u003c/span\u003e) and applied to mDixon and FOD data.\u003c/p\u003e\n \u003c/span\u003e \u003cspan\u003e\n \u003cp\u003e3) The median of registered mDixon \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({I}_{k}\\)\u003c/span\u003e\u003c/span\u003e and FOD \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({J}_{k}\\)\u003c/span\u003e\u003c/span\u003e volumes were computed to update their atlases \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{k}^{I}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{k}^{J}\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\n \u003c/span\u003e\n \u003cp\u003eA multi-resolution pyramid with default scale values in MRtrix (Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e) was used to perform registration at different resolution levels, starting with down-sampled images (scale\u0026thinsp;\u0026lt;\u0026thinsp;1) and progressively working towards original high-resolution images (scale\u0026thinsp;=\u0026thinsp;1). The use of this multi-resolution pyramid enhances efficient image registration by focusing on large structures first, followed by focusing on finer details (Thevenaz et al., \u003cspan class=\"CitationRef\"\u003e1998\u003c/span\u003e). Registration parameters can be found in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eSummary of registration parameters.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"5\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eStage no.\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eType of registration\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNumber of iterations \u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSH degrees\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMulti-resolution pyramid \u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1 to 6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003erigid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.3, 0.4, 0.6, 0.8, 1.0, 1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7 to 12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eaffine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.3, 0.4, 0.6, 0.8, 1.0, 1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13 to 28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enon-linear\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003csup\u003ea\u003c/sup\u003eThe number of registration iterations used within each stage before updating the atlas.\u003c/p\u003e\n \u003cp\u003e\u003csup\u003eb\u003c/sup\u003eThe list of scale values that represents the degree to which the original image was downsampled in each iteration.\u003c/p\u003e\n \u003cp\u003eSH\u0026thinsp;=\u0026thinsp;spherical harmonic\u003c/p\u003e\n \u003cp\u003eThe displacement field obtained at each non-linear iteration was smoothed using a Gaussian filter with a standard deviation of twice the voxel size of the iteration\u0026rsquo;s resolution.\u003c/p\u003e\n \u003cp\u003eAfter registration, the output consisted of three components:\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e1) A population-averaged mDixon atlas \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{final}^{I}\\)\u003c/span\u003e\u003c/span\u003e and a corresponding FOD atlas \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({T}_{final}^{J}\\)\u003c/span\u003e\u003c/span\u003e in a common coordinate space;\u003c/p\u003e\n \u003c/span\u003e \u003cspan\u003e\n \u003cp\u003e2) individual subject image data, specifically, mDixon images \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({I}_{r}\\)\u003c/span\u003e\u003c/span\u003e (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(r\\in \\left[1,n-1\\right]\\)\u003c/span\u003e\u003c/span\u003e), FODs \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({J}_{r}\\)\u003c/span\u003e\u003c/span\u003e, and masks of muscles and the scapula, each registered and aligned to the atlas space; and\u003c/p\u003e\n \u003c/span\u003e \u003cspan\u003e\n \u003cp\u003e3) for each subject, a displacement field \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\varnothing }_{r}\\)\u003c/span\u003e\u003c/span\u003e that warps subject images to the atlas space. This included both linear (rigid and affine) and non-linear displacements.\u003c/p\u003e\n \u003c/span\u003e\n \u003cp\u003eWe computed the atlas mask by calculating the median of all registered subject masks in atlas space.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e\n \u003ch2\u003e2.3.3 Evaluation\u003c/h2\u003e\u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003eQualitative evaluation\u003c/strong\u003e\u003c/p\u003e\n \u003c/span\u003e\n \u003cp\u003eQualitative evaluation included reconstruction, visual inspection, and comparison of fibre tractography of all muscles to visually assess fibre orientations. Fibre tracts were reconstructed from each subject\u0026rsquo;s FODs, as well as from the population-averaged FOD atlases. For tractography, we employed the MRtrix SD_STREAM algorithm (Tournier et al., \u003cspan class=\"CitationRef\"\u003e2012\u003c/span\u003e), a deterministic algorithm based on spherical deconvolution. The tracts were dynamically seeded using the SIFT model within MRtrix (Smith et al., \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e), a mechanism designed to improve the evenness of tract density distribution. In each muscle, a total of 3000 fibre tracts were generated with the following settings: integration step size\u0026thinsp;=\u0026thinsp;1.0 mm; 0.1 \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\le\\)\u003c/span\u003e\u003c/span\u003e fractional anisotropy \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\le\\)\u003c/span\u003e\u003c/span\u003e 0.5; maximum turning angle between successive steps \u0026gt; 15\u0026deg;; 25 mm \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\le\\)\u003c/span\u003e\u003c/span\u003e tract length \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\le\\)\u003c/span\u003e\u003c/span\u003e 200 mm. These reconstructions were then examined in 3D Slicer (Fedorov et al., \u003cspan class=\"CitationRef\"\u003e2012\u003c/span\u003e) for 3D visualisation and visual comparison of fibre orientations.\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003eQuantitative evaluation\u003c/strong\u003e\u003c/p\u003e\n \u003c/span\u003e\n \u003cp\u003eThe multi-channel registration process was validated using leave-one-out cross-validation (LOOCV), independently for male and female atlases. Thus, 11 male and 9 female atlases were created, each based on a different subset of 10 and 8 subjects, respectively.\u003c/p\u003e\n \u003cp\u003eAll atlases were visually inspected to ensure proper alignment between channels and to verify the anatomical fidelity of structures. We then quantitatively evaluated the performance of multi-channel registration using three metrics per muscle. The spatial alignment accuracy was measured by the Dice coefficient (Dice, \u003cspan class=\"CitationRef\"\u003e1945\u003c/span\u003e), which quantifies the degree of overlap between the atlas mask and the mask of each subject warped to the atlas space:\u003c/p\u003e\n \u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e$$\\text{D}\\text{i}\\text{c}\\text{e}=\\frac{2\\left|A\\cap B\\right|}{\\left|A\\right|+\\left|B\\right|}$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003ewhere A and B represents the region of the atlas mask and the warped mask of each subject. A Dice coefficient closer to 1 indicates a greater degree of alignment between the spatial locations of the masks.\u003c/p\u003e\n \u003cp\u003eThe angular alignment between estimated and subject-specific fibre orientations was quantified by calculating the angular correlation coefficient (ACC) (Anderson, \u003cspan class=\"CitationRef\"\u003e2005\u003c/span\u003e) between the FOD atlas and the FOD of each subject warped to the atlas space. The ACC was determined for each muscle on a voxel-by-voxel basis as:\u003c/p\u003e\n \u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e$$\\text{A}\\text{C}\\text{C}=\\frac{\\sum _{m=-l}^{l}{a}_{l}^{m}{b}_{l}^{m}}{\\left({\\sum _{m=-l}^{l}{{(a}_{l}^{m})}^{2} )}^{\\frac{1}{2}}\\right({\\sum _{m=-l}^{l}{{(b}_{l}^{m})}^{2} )}^{\\frac{1}{2}}}$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({a}_{l}^{m}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({b}_{l}^{m}\\)\u003c/span\u003e\u003c/span\u003e represents the coefficient for the SH function of degree \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(l\\)\u003c/span\u003e\u003c/span\u003e (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(l=2\\)\u003c/span\u003e\u003c/span\u003e in this study) and order \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(m\\)\u003c/span\u003e\u003c/span\u003e of the FOD atlas and the warped FOD, respectively.\u003c/p\u003e\n \u003cp\u003eTo further quantify the alignment between predicted and subject-specific fibre orientations, we first calculated the principal diffusion direction within each voxel of the FODs using the sh2peaks function in MRtrix. Subsequently, we calculated the 3D angular difference between each of the peak direction vectors for the atlas and the corresponding vectors warped from the subject\u0026rsquo;s FOD. While the ACC provides a dimensionless measure of similarity in orientation distribution, the angular difference provides a measure of muscle fibre orientation alignment accuracy expressed in a familiar metric (degrees). The median ACCs and median angular difference were used to summarise the alignment between predicted and subject-specific fibre orientations for a muscle. Histograms illustrating the distributions can be found in the Supplementary Material.\u003c/p\u003e\n \u003cp\u003eThe mean Dice coefficient, ACC, and angular difference, along with their respective standard deviations, were calculated for each individual case from the LOOCV.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\n \u003ch2\u003e2.4 Prediction of fibre orientations from anatomical MRI\u003c/h2\u003e\n \u003cp\u003eFor the second aim of this study, we used the atlas to predict fibre orientations for out-of-sample subjects, using only the subject\u0026rsquo;s anatomical mDixon scan (which does not contain information about fibre orientations). Predicted fibre orientations were compared to orientations determined from the subject\u0026rsquo;s DWI data.\u003c/p\u003e\n \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\n \u003ch2\u003e2.4.1 FOD prediction\u003c/h2\u003e\n \u003cp\u003eThe mDixon scan of the out-of-sample subject was registered to the population-averaged mDixon atlas using SyN Demons with the SSD metric, mirroring the rigid, affine, and non-linear registration stages used in atlas building. The resulting displacement field was inverted, reoriented, and then used to warp the FOD atlas into the anatomical space of the out-of-sample subject. The atlas muscle mask was also warped to the subject space for subsequent evaluations.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\n \u003ch2\u003e2.4.2 Evaluation\u003c/h2\u003e\n \u003cp\u003eFor the qualitative evaluation of fibre orientation prediction, we applied the same fibre tractography methods as described previously. Using 3D Slicer, we visually examined and compared the fibre tract reconstructions from subject-specific DWI-derived FODs against atlas-predicted FODs for the out-of-sample subjects.\u003c/p\u003e\n \u003cp\u003eTo determine the accuracy of image registration between the subject\u0026rsquo;s mDixon and the population-averaged mDixon atlas, the Dice coefficient was calculated between the subject\u0026rsquo;s mask and the atlas mask warped to subject space. Similarly, to determine the accuracy of fibre orientation prediction, the ACC and angular difference between the subject\u0026rsquo;s FOD and the atlas FOD warped to subject space was calculated.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e"},{"header":"3 Results","content":"\u003cp\u003eWe successfully constructed 11 sets of atlases of mDixon and FOD maps for the male cohort and 9 sets for the female cohort. Data derived for each atlas, including Dice coefficients, ACCs, and angular differences for both the evaluation of multi-channel registration performance and the fibre orientation prediction, can be found in Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e and Table S2 in the Supplementary Material.\u003c/p\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Multi-channel population-averaged atlas and registration performance\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Examples of one male (top) and one female (bottom) multi-channel atlas. There are two rows for the male and two rows for the female: the upper and lower rows illustrate transverse mDixon slices approximately midway through the acromioclavicular and glenohumeral joint, respectively. Shown from left to right are: the mDixon water image, masks of rotator cuff muscles and scapula overlayed on mDixon image (blue, scapula; beige, subscapularis; pink, infraspinatus and teres minor; orange, supraspinatus), and an FOD image overlayed on mDixon. FODs are colour-coded according to orientation (red: left-right; green: anterior-posterior; blue: inferior-superior).\u003c/p\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e3.1.1 Qualitative evaluation\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Comparison of 3D fibre tractography reconstructions of rotator cuff muscles from the right shoulder, using data from an atlas generated from 10 male subjects. For each muscle, the illustration showcases the tractographies for subjects with the (a) best, (b) median, and (c) worst angular alignment within this cohort. Each pair highlights the fibre reconstructions set against the 3D surface model of the muscle in transparent yellow; the top and bottom row displays fibre reconstructed from the subject\u0026rsquo;s original FODs and the FOD atlas, respectively. Fibre tracts are colour-coded according to orientation (red: medial-lateral; green: anterior-posterior; blue: inferior-superior).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e3.1.2 Quantitative evaluation\u003c/h2\u003e \u003cp\u003eThere was a consistently high degree of spatial overlap and angular alignment between the images of subjects registered to the atlases and the corresponding atlases for both male and female cohorts as evidenced by the high Dice coefficients (males: 0.888; females: 0.856), high ACCs (males: 0.949; females: 0.974), and small angular differences (males: 10.5\u0026deg;; females: 7.8\u0026deg;; Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Small standard deviations observed across the atlases for the LOOCV (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) further indicate the high degree of consistency in registration performance.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of leave-one-out cross-validation results.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"10\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eDice coefficient\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003eACC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c10\" namest=\"c8\"\u003e \u003cp\u003eAngular difference (\u0026deg;)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eINF\u0026thinsp;+\u0026thinsp;TER\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSUB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSUP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eINF\u0026thinsp;+\u0026thinsp;TER\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSUB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eSUP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eINF\u0026thinsp;+\u0026thinsp;TER\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSUB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003eSUP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.890\u003c/p\u003e \u003cp\u003e(0.006)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.890\u003c/p\u003e \u003cp\u003e(0.003)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.885\u003c/p\u003e \u003cp\u003e(0.005)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.956\u003c/p\u003e \u003cp\u003e(0.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.937\u003c/p\u003e \u003cp\u003e(0.002)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.953\u003c/p\u003e \u003cp\u003e(0.002)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e9.8\u003c/p\u003e \u003cp\u003e(0.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e11.8\u003c/p\u003e \u003cp\u003e(0.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e9.8\u003c/p\u003e \u003cp\u003e(0.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.881\u003c/p\u003e \u003cp\u003e(0.005)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.857\u003c/p\u003e \u003cp\u003e(0.006)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.829\u003c/p\u003e \u003cp\u003e(0.010)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.974\u003c/p\u003e \u003cp\u003e(0.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.968\u003c/p\u003e \u003cp\u003e(0.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.981\u003c/p\u003e \u003cp\u003e(0.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e7.6\u003c/p\u003e \u003cp\u003e(0.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e8.9\u003c/p\u003e \u003cp\u003e(0.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e6.9\u003c/p\u003e \u003cp\u003e(0.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eValues are means (standard deviations) for each metric across all atlases within each cohort.\u003c/p\u003e \u003cp\u003eINF\u0026thinsp;=\u0026thinsp;infraspinatus; TER\u0026thinsp;=\u0026thinsp;teres minor; SUB\u0026thinsp;=\u0026thinsp;subscapularis; SUP\u0026thinsp;=\u0026thinsp;supraspinatus.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Prediction of fibre orientations for out-of-sample subjects\u003c/h2\u003e \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Qualitative evaluation\u003c/h2\u003e \u003cp\u003eFibre tracts of out-of-sample subjects reconstructed from original and predicted fibre orientations showed a high degree of visual correspondence (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e), demonstrating that the atlas can effectively predict complex patterns of 3D fibre orientations using anatomical image data alone. Fibres reconstructed from the predicted fibre orientations generally showed smoother tracts which covered a larger region of the muscles, particularly at the medial border of the subscapularis, infraspinatus and teres minor, compared to those from the original fibre orientations.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Quantitative evaluation\u003c/h2\u003e \u003cp\u003eThe primary results of the quantitative evaluation of fibre orientation prediction accuracy are illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. High Dice coefficients and ACCs, coupled with relatively low angular differences were found for both male and female atlases. Variability in these metrics was noted between subjects and across different muscles but remained within a relatively narrow range. Detailed histograms illustrating the distribution of these metrics across each muscle for each atlas are available as Fig. S2 in the supplementary material. Furthermore, Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e provides a visualisation of fibre orientation prediction accuracy on transverse slices for an out-of-sample subject. Regions of relatively low ACCs and correspondingly high angular differences, indicating regions of less satisfactory angular alignment, were most obvious at the muscle boundaries and within the subscapularis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eThis manuscript describes the construction of an atlas of human rotator cuff muscle shape and architecture, and demonstrates two important applications. First, we used the atlases to smooth fibre tracts and fill in areas where fibre tracts were sparsely reconstructed due to image noise, substantially improving the quality of subject-specific muscle reconstructions. Second, we demonstrated the use of the atlases to accurately predict fibre orientations from anatomical MRI alone, bypassing the need to conduct diffusion-weighted scans, at least for some applications.\u003c/p\u003e \u003cp\u003eOur multi-channel registration pipeline was built on methods that have been widely applied in neuroimaging (Forsberg et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Roura, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Uus et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2020\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). One example is a study that integrated T\u003csub\u003e1\u003c/sub\u003e, T\u003csub\u003e2\u003c/sub\u003e and FOD data from 20 neonatal brains using similar registration methods and reported the mean Dice coefficient of 0.735 and ACC of 0.455, averaged across brain tissues (Uus et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Our study yielded much higher Dice coefficients and ACCs, probably because rotator cuff muscles have less complex fibre pathways than the neonatal brain. The high Dice coefficients and ACCs demonstrate the ability of our registration pipeline to effectively characterise inter-individual differences in both the macrostructural anatomy and internal architecture of the muscles.\u003c/p\u003e \u003cp\u003eDespite good quantitative registration results, we noticed that boundaries between adjacent muscle groups in the mDixon atlases were sometimes blurry, and these regions exhibited lower angular alignment, indicating relatively poor registration performance. In general, the need for multi-channel registration to synthesise information about anatomical structure obtained from mDixon images with information about fibre orientation obtained from DWI presented unique challenges. For example, noise and image artifacts in DWI data could bias the registration process. Nonetheless, population-averaged atlases had noticeably less noise and fewer artifacts than the muscle reconstructions calculated from original DWIs of individual subjects, as demonstrated by both qualitative and quantitative evaluation. Thus population-averaging of muscle reconstructions has a noise-cancelling effect, as has been observed previously in brain imaging studies (Jones et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; H. Park, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). The fidelity of our registration outputs, atlas construction and predicted fibre orientations might be enhanced by further developing the multi-channel registration algorithm. For instance, a weighted averaging approach to combining channel-specific updates to displacement fields (Forsberg et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), which assigns greater weight to channels with higher certainty or image quality, may offer a more accurate registration. The efficacy of this approach in enhancing registration accuracy of the neonatal brain has been demonstrated previously (Uus et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe multi-channel registration approach and the resultant population-averaged atlases have potentially broad application. The atlases could serve as a highly accurate mean shape for SSMs, with correspondences established through the advanced registration process that integrates information from both anatomical and diffusion MRIs. This integration offers an enhanced level of detail and accuracy over traditional landmark-based approach, and provides a more precise representation of the complex rotator cuff anatomy. Moreover, incorporating data on fibre orientations, which are typically not accounted for in population-based approaches to muscle modelling, could improve predictions of muscle function made by musculoskeletal models. Furthermore, the population-averaged FOD atlas presents a promising approach to mitigating the challenges posed by the intrinsic sensitivity of DWI to noise and image artifacts. By leveraging the atlas\u0026rsquo;s smoothed representations and expanded coverage of muscle fibres, it becomes possible to refine the analysis of subject-specific fibre orientations. For example, the atlas could be employed as a filtering tool to selectively combine subject fibre orientations derived from DWI with the atlas information, according to certain criteria that prioritise anatomically plausible fibre orientations. This integration enables the use of anatomically constrained fiber tractography\u0026mdash;a methodological framework we have established in our previous work (Zhang et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) \u0026ndash; for subject-specific reconstructions and measurements of muscle architecture. By reducing or mitigating anatomically implausible fiber tracts and enhancing tract coverage, we anticipate that this refined analysis may improve measurement accuracy and deepen our understanding of muscle function.\u003c/p\u003e \u003cp\u003eAnother promising application of the approach presented here is prediction of fibre orientations from anatomical images when DWI data is not available. For example, predicted fibre orientations could be used to assign fibre orientations in subject-specific finite element models of muscles, thereby enhancing their anatomical fidelity and the accuracy of simulations. It has been shown that fibre orientations significantly affected muscle mechanics during simulated contractions (Alipour et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), highlighting the significance of accurate assignment of fibre orientations in finite element models.\u003c/p\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Limitations and future directions\u003c/h2\u003e \u003cp\u003eThe generalisability of the atlases presented in this study is limited by the small number and age range of subjects included in our cohort. This specificity limits the atlases\u0026rsquo; applicability to populations that diverge from our study group, potentially reducing their utility in more diverse research and clinical settings. Future research should focus on expanding the demographic breadth of the study cohorts by including data from subject with a wider range of age and physical condition. Examination of the out-of-sample subjects for whom fibre orientation prediction was least accurate found that these subjects\u0026rsquo; mDixon images had poorer image quality. It was also evident that partial volume effects in some mDixon images made identification of muscle boundaries difficult. These aspects of image quality are likely to be key factors affecting registration accuracy. Registration accuracy is also likely to be reduced when there is large between-subject variation in muscle volume. This challenge was particularly evident in our initial attempts to construct a single atlas including both male and female subjects. The significant differences in muscle volumes between sexes hindered our ability to achieve accurate alignment, compromising registration quality. This issue underscored the need for further improvements in image registration to accommodate large anatomical variations across individuals and cohorts. Collectively, these findings underscore the importance of obtaining high-quality anatomical MRI scans and the need for robust registration algorithms capable of compensating for anatomical variability.\u003c/p\u003e \u003cp\u003eVariability in imaging protocols and equipment across different studies also poses a challenge. Our registration and prediction algorithms were optimised for a specific set of MRI scanner settings and protocols. As such, differences in equipment and scanning parameters that exist in broader clinical practice could affect the robustness and reliability of the atlases when applied outside the controlled conditions of our study. Specifically, our current fibre orientation prediction method hinges on accurate registration of mDixon images from new subjects to these mDixon-based atlases, where the imaging conditions were closely matched. Applying these atlases to images acquired with different protocols or modalities, such as more commonly used T\u003csub\u003e1\u003c/sub\u003e- or T\u003csub\u003e2\u003c/sub\u003e-weighted MRI sequences, would require validation of the ability of our atlases to accurately align with these new modalities. Multi-centre, multi-modal validation studies could determine the limits and capabilities of the atlas-based approach when applied to broader datasets. Additionally, the utility of atlases might be enhanced by adaptive algorithms that can fine-tune the registration parameters based on the specific characteristics of the input data. By incorporating machine learning techniques or advanced image processing algorithms, it may be possible to automate the adjustment of these parameters, thereby improving the adaptability and accuracy of our atlases across diverse imaging environments.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eConflict of interest statement\u003c/h2\u003e \u003cp\u003eI hereby state that none of the authors have had any financial or personal relationships with other people or organizations that could inappropriately influence (bias) our work.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eYilan Zhang: Conceptualisation, methodology, software, validation, formal analysis, investigation, resources, data curation, visualisation, writing \u0026ndash; original draft, writing \u0026ndash; review and editing, project administration.Rob Lloyd: Conceptualisation, validation, writing - review and editing, supervision.Robert D. Herbert: Conceptualisation, methodology, writing \u0026ndash; review and editing, project administration, supervision.Lynne E. Bilston: Conceptualisation, validation, writing - review and editing, supervision.Bart Bolsterlee: Conceptualisation, methodology, software, validation, formal analysis, investigation, resources, data curation, visualisation, writing \u0026ndash; review and editing, project administration, supervision.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eThis study was supported by the Australian Research Council through the Industrial Transformation Training Centre Program for Joint Biomechanics (IC190100020). Y. Zhang is supported by a UNSW Tuition Fee Scholarship. L. Bilston is supported by an NHMRC Investigator grant, (1172988). The authors acknowledge the facilities and scientific and technical assistance of the National Imaging Facility, a National Collaborative Research Infrastructure Strategy (NCRIS) capability, at Neuroscience Research Australia and UNSW.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAlipour, M., Mithraratne, K., \u0026amp; Fernandez, J. (2017). A diffusion-weighted imaging informed continuum model of the rabbit triceps surae complex. \u003cem\u003eBiomechanics and Modeling in Mechanobiology\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e(5), 1729\u0026ndash;1741. https://doi.org/10.1007/s10237-017-0916-4\u003c/li\u003e\n\u003cli\u003eAlonso-Fernandez, D., Docampo-Blanco, P., \u0026amp; Martinez-Fernandez, J. (2018). Changes in muscle architecture of biceps femoris induced by eccentric strength training with nordic hamstring exercise. \u003cem\u003eScandinavian Journal of Medicine \u0026amp; Science in Sports\u003c/em\u003e, \u003cem\u003e28\u003c/em\u003e(1), 88\u0026ndash;94. https://doi.org/10.1111/sms.12877\u003c/li\u003e\n\u003cli\u003eAnderson, A. W. (2005). Measurement of fiber orientation distributions using high angular resolution diffusion imaging. \u003cem\u003eMagnetic Resonance in Medicine\u003c/em\u003e, \u003cem\u003e54\u003c/em\u003e(5), 1194\u0026ndash;1206. https://doi.org/10.1002/mrm.20667\u003c/li\u003e\n\u003cli\u003eAndersson, J. L. R., \u0026amp; Sotiropoulos, S. N. (2016). An integrated approach to correction for off-resonance effects and subject movement in diffusion MR imaging. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e125\u003c/em\u003e, 1063\u0026ndash;1078. https://doi.org/10.1016/j.neuroimage.2015.10.019\u003c/li\u003e\n\u003cli\u003eAvants, B., Duda, J. T., Zhang, H., \u0026amp; Gee, J. C. (2007). Multivariate Normalization with Symmetric Diffeomorphisms for Multivariate Studies. In N. Ayache, S. Ourselin, \u0026amp; A. Maeder (Eds.), \u003cem\u003eMedical Image Computing and Computer-Assisted Intervention \u0026ndash; MICCAI 2007\u003c/em\u003e (Vol. 4791, pp. 359\u0026ndash;366). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-540-75757-3_44\u003c/li\u003e\n\u003cli\u003eBin Ghouth, S. G., Williams, S. A., Reid, S. L., Besier, T. F., \u0026amp; Handsfield, G. G. (2022). A statistical shape model of soleus muscle morphology in spastic cerebral palsy. \u003cem\u003eScientific Reports\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(1), 7711. https://doi.org/10.1038/s41598-022-11611-z\u003c/li\u003e\n\u003cli\u003eBlazevich, A. J., Gill, N. D., Bronks, R., \u0026amp; Newton, R. U. (2003). Training-Specific Muscle Architecture Adaptation after 5-wk Training in Athletes: \u003cem\u003eMedicine \u0026amp; Science in Sports \u0026amp; Exercise\u003c/em\u003e, \u003cem\u003e35\u003c/em\u003e(12), 2013\u0026ndash;2022. https://doi.org/10.1249/01.MSS.0000099092.83611.20\u003c/li\u003e\n\u003cli\u003eBodine, S. C., Roy, R. R., Meadows, D. A., Zernicke, R. F., Sacks, R. D., Fournier, M., \u0026amp; Edgerton, V. R. (1982). Architectural, histochemical, and contractile characteristics of a unique biarticular muscle: The cat semitendinosus. \u003cem\u003eJournal of Neurophysiology\u003c/em\u003e, \u003cem\u003e48\u003c/em\u003e(1), 192\u0026ndash;201. https://doi.org/10.1152/jn.1982.48.1.192\u003c/li\u003e\n\u003cli\u003eBolsterlee, B. (2022). A new framework for analysis of three-dimensional shape and architecture of human skeletal muscles from in vivo imaging data. \u003cem\u003eJournal of Applied Physiology\u003c/em\u003e, \u003cem\u003e132\u003c/em\u003e(3), 712\u0026ndash;725. https://doi.org/10.1152/japplphysiol.00638.2021\u003c/li\u003e\n\u003cli\u003eB\u0026uuml;chler, P., \u0026amp; Farron, A. (2004). Benefits of an anatomical reconstruction of the humeral head during shoulder arthroplasty: A finite element analysis. \u003cem\u003eClinical Biomechanics\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(1), 16\u0026ndash;23. https://doi.org/10.1016/j.clinbiomech.2003.09.009\u003c/li\u003e\n\u003cli\u003eDice, L. R. (1945). Measures of the Amount of Ecologic Association Between Species. \u003cem\u003eEcology\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e(3), 297\u0026ndash;302. https://doi.org/10.2307/1932409\u003c/li\u003e\n\u003cli\u003eFahn-Lai, P., Biewener, A. A., \u0026amp; Pierce, S. E. (2020). Broad similarities in shoulder muscle architecture and organization across two amniotes: Implications for reconstructing non-mammalian synapsids. \u003cem\u003ePeerJ\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e, e8556. https://doi.org/10.7717/peerj.8556\u003c/li\u003e\n\u003cli\u003eFedorov, A., Beichel, R., Kalpathy-Cramer, J., Finet, J., Fillion-Robin, J.-C., Pujol, S., Bauer, C., Jennings, D., Fennessy, F., Sonka, M., Buatti, J., Aylward, S., Miller, J. V., Pieper, S., \u0026amp; Kikinis, R. (2012). 3D Slicer as an image computing platform for the Quantitative Imaging Network. \u003cem\u003eMagnetic Resonance Imaging\u003c/em\u003e, \u003cem\u003e30\u003c/em\u003e(9), 1323\u0026ndash;1341. https://doi.org/10.1016/j.mri.2012.05.001\u003c/li\u003e\n\u003cli\u003eForsberg, D., Rathi, Y., Bouix, S., Wassermann, D., Knutsson, H., \u0026amp; Westin, C.-F. (2011). Improving Registration Using Multi-channel Diffeomorphic Demons Combined with Certainty Maps. In T. Liu, D. Shen, L. Ibanez, \u0026amp; X. Tao (Eds.), \u003cem\u003eMultimodal Brain Image Analysis\u003c/em\u003e (Vol. 7012, pp. 19\u0026ndash;26). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-642-24446-9_3\u003c/li\u003e\n\u003cli\u003eGazielly, D. F., Gleyze, P., \u0026amp; Montagnon, C. (1994). Functional and anatomical results after rotator cuff repair. \u003cem\u003eClinical Orthopaedics and Related Research\u003c/em\u003e, \u003cem\u003e304\u003c/em\u003e, 43\u0026ndash;53.\u003c/li\u003e\n\u003cli\u003eGerber, C., Fuchs, B., \u0026amp; Hodler, J. (2000). The results of repair of massive tears of the rotator cuff. \u003cem\u003eJBJS\u003c/em\u003e, \u003cem\u003e82\u003c/em\u003e(4), 505.\u003c/li\u003e\n\u003cli\u003eHotelling, H. (1933). Analysis of a complex of statistical variables into principal components. \u003cem\u003eJournal of Educational Psychology\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(6), 417\u0026ndash;441. https://doi.org/10.1037/h0071325\u003c/li\u003e\n\u003cli\u003eIsensee, F., Jaeger, P. F., Kohl, S. A. A., Petersen, J., \u0026amp; Maier-Hein, K. H. (2021). nnU-Net: A self-configuring method for deep learning-based biomedical image segmentation. \u003cem\u003eNature Methods\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(2), 203\u0026ndash;211. https://doi.org/10.1038/s41592-020-01008-z\u003c/li\u003e\n\u003cli\u003eJenkinson, M., Bannister, P., Brady, M., \u0026amp; Smith, S. (2002). Improved optimization for the robust and accurate linear registration and motion correction of brain images. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(2), 825\u0026ndash;841. https://doi.org/10.1016/s1053-8119(02)91132-8\u003c/li\u003e\n\u003cli\u003eJenkinson, M., \u0026amp; Smith, S. (2001). A global optimisation method for robust affine registration of brain images. \u003cem\u003eMedical Image Analysis\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e(2), 143\u0026ndash;156. https://doi.org/10.1016/S1361-8415(01)00036-6\u003c/li\u003e\n\u003cli\u003eJones, D. K., Griffin, L. D., Alexander, D. C., Catani, M., Horsfield, M. A., Howard, R., \u0026amp; Williams, S. C. R. (2002). Spatial Normalization and Averaging of Diffusion Tensor MRI Data Sets. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(2), 592\u0026ndash;617. https://doi.org/10.1006/nimg.2002.1148\u003c/li\u003e\n\u003cli\u003eKhandare, S., Arce, R. A., \u0026amp; Vidt, M. E. (2022). Muscle compensation strategies to maintain glenohumeral joint stability with increased rotator cuff tear severity: A simulation study. \u003cem\u003eJournal of Electromyography and Kinesiology\u003c/em\u003e, \u003cem\u003e62\u003c/em\u003e, 102335. https://doi.org/10.1016/j.jelekin.2019.07.005\u003c/li\u003e\n\u003cli\u003eKruse, A., Schranz, C., Tilp, M., \u0026amp; Svehlik, M. (2018). Muscle and tendon morphology alterations in children and adolescents with mild forms of spastic cerebral palsy. \u003cem\u003eBMC Pediatrics\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(1), 156. https://doi.org/10.1186/s12887-018-1129-4\u003c/li\u003e\n\u003cli\u003eLieber, R. L., \u0026amp; Frid\u0026eacute;n, J. (2000). Functional and clinical significance of skeletal muscle architecture. \u003cem\u003eMuscle \u0026amp; Nerve\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(11), 1647\u0026ndash;1666. https://doi.org/10.1002/1097-4598(200011)23:11\u0026lt;1647::aid-mus1\u0026gt;3.0.co;2-m\u003c/li\u003e\n\u003cli\u003eNarici, M. V., Maganaris, C. N., Reeves, N. D., \u0026amp; Capodaglio, P. (2003). Effect of aging on human muscle architecture. \u003cem\u003eJournal of Applied Physiology\u003c/em\u003e, \u003cem\u003e95\u003c/em\u003e(6), 2229\u0026ndash;2234. https://doi.org/10.1152/japplphysiol.00433.2003\u003c/li\u003e\n\u003cli\u003ePapenkort, S., B\u0026ouml;l, M., \u0026amp; Siebert, T. (2021). Architectural model for muscle growth during maturation. \u003cem\u003eBiomechanics and Modeling in Mechanobiology\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(5), 2031\u0026ndash;2044. https://doi.org/10.1007/s10237-021-01492-y\u003c/li\u003e\n\u003cli\u003ePark, H. (2003). Spatial normalization of diffusion tensor MRI using multiple channels. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(4), 1995\u0026ndash;2009. https://doi.org/10.1016/j.neuroimage.2003.08.008\u003c/li\u003e\n\u003cli\u003ePark, J. E., Seong, Y.-J., Kim, E. S., Park, D., Lee, Y., Park, H., \u0026amp; Rha, D. (2019). Architectural Changes in the Medial Gastrocnemius on Sonography after Nerve Ablation in Healthy Adults. \u003cem\u003eYonsei Medical Journal\u003c/em\u003e, \u003cem\u003e60\u003c/em\u003e(9), 876. https://doi.org/10.3349/ymj.2019.60.9.876\u003c/li\u003e\n\u003cli\u003ePearson, K. (1901). On lines and planes of closest fit to systems of points in space. \u003cem\u003eThe London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(11), 559\u0026ndash;572. https://doi.org/10.1080/14786440109462720\u003c/li\u003e\n\u003cli\u003ePrinold, J. A., Masjedi, M., Johnson, G. R., \u0026amp; Bull, A. M. (2013). Musculoskeletal shoulder models: A technical review and proposals for research foci. \u003cem\u003eProceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine\u003c/em\u003e, \u003cem\u003e227\u003c/em\u003e(10), 1041\u0026ndash;1057. https://doi.org/10.1177/0954411913492303\u003c/li\u003e\n\u003cli\u003eRaffelt, D., Tournier, J., Crozier, S., Connelly, A., \u0026amp; Salvado, O. (2012). Reorientation of fiber orientation distributions using apodized point spread functions. \u003cem\u003eMagnetic Resonance in Medicine\u003c/em\u003e, \u003cem\u003e67\u003c/em\u003e(3), 844\u0026ndash;855. https://doi.org/10.1002/mrm.23058\u003c/li\u003e\n\u003cli\u003eRoig, M., Shadgan, B., \u0026amp; Reid, W. D. (2008). Eccentric Exercise in Patients with Chronic Health Conditions: A Systematic Review. \u003cem\u003ePhysiotherapy Canada\u003c/em\u003e, \u003cem\u003e60\u003c/em\u003e(2), 146\u0026ndash;160. https://doi.org/10.3138/physio.60.2.146\u003c/li\u003e\n\u003cli\u003eRoura, E. (2015). Multi-channel registration of fractional anisotropy and T1-weighted images in the presence of atrophy: Application to multiple sclerosis. \u003cem\u003eFunctional Neurology\u003c/em\u003e. https://doi.org/10.11138/FNeur/2015.30.4.245\u003c/li\u003e\n\u003cli\u003eSalhi, A., Burdin, V., Boutillon, A., Brochard, S., Mutsvangwa, T., \u0026amp; Borotikar, B. (2020). Statistical Shape Modeling Approach to Predict Missing Scapular Bone. \u003cem\u003eAnnals of Biomedical Engineering\u003c/em\u003e, \u003cem\u003e48\u003c/em\u003e(1), 367\u0026ndash;379. https://doi.org/10.1007/s10439-019-02354-6\u003c/li\u003e\n\u003cli\u003eShur, N. F., Creedon, L., Skirrow, S., Atherton, P. J., MacDonald, I. A., Lund, J., \u0026amp; Greenhaff, P. L. (2021). Age-related changes in muscle architecture and metabolism in humans: The likely contribution of physical inactivity to age-related functional decline. \u003cem\u003eAgeing Research Reviews\u003c/em\u003e, \u003cem\u003e68\u003c/em\u003e, 101344. https://doi.org/10.1016/j.arr.2021.101344\u003c/li\u003e\n\u003cli\u003eSiebert, T., Tomalka, A., Stutzig, N., Leichsenring, K., \u0026amp; B\u0026ouml;l, M. (2017). Changes in three-dimensional muscle structure of rabbit gastrocnemius, flexor digitorum longus, and tibialis anterior during growth. \u003cem\u003eJournal of the Mechanical Behavior of Biomedical Materials\u003c/em\u003e, \u003cem\u003e74\u003c/em\u003e, 507\u0026ndash;519. https://doi.org/10.1016/j.jmbbm.2017.07.045\u003c/li\u003e\n\u003cli\u003eSmith, R. E., Tournier, J.-D., Calamante, F., \u0026amp; Connelly, A. (2015). SIFT2: Enabling dense quantitative assessment of brain white matter connectivity using streamlines tractography. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e119\u003c/em\u003e, 338\u0026ndash;351. https://doi.org/10.1016/j.neuroimage.2015.06.092\u003c/li\u003e\n\u003cli\u003eSu-Lin Lee, Tan, E., Khullar, V., Gedroyc, W., Darzi, A., \u0026amp; Guang-Zhong Yang. (2009). Physical-Based Statistical Shape Modeling of the Levator Ani. \u003cem\u003eIEEE Transactions on Medical Imaging\u003c/em\u003e, \u003cem\u003e28\u003c/em\u003e(6), 926\u0026ndash;936. https://doi.org/10.1109/TMI.2009.2012894\u003c/li\u003e\n\u003cli\u003eSutherland, A. M. T., Lynch, J. T., Serpell, B. G., Pickering, M. R., Newman, P., Perriman, D. M., \u0026amp; Kenneally-Dabrowski, C. (2023). Statistical shape modelling reveals differences in hamstring morphology between professional rugby players and sprinters. \u003cem\u003eJournal of Sports Sciences\u003c/em\u003e, \u003cem\u003e41\u003c/em\u003e(2), 164\u0026ndash;171. https://doi.org/10.1080/02640414.2023.2204269\u003c/li\u003e\n\u003cli\u003eThevenaz, P., Ruttimann, U. E., \u0026amp; Unser, M. (1998). A pyramid approach to subpixel registration based on intensity. \u003cem\u003eIEEE Transactions on Image Processing\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(1), 27\u0026ndash;41. https://doi.org/10.1109/83.650848\u003c/li\u003e\n\u003cli\u003eTournier, J.-D., Calamante, F., \u0026amp; Connelly, A. (2007). Robust determination of the fibre orientation distribution in diffusion MRI: Non-negativity constrained super-resolved spherical deconvolution. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e35\u003c/em\u003e(4), 1459\u0026ndash;1472. https://doi.org/10.1016/j.neuroimage.2007.02.016\u003c/li\u003e\n\u003cli\u003eTournier, J.-D., Calamante, F., \u0026amp; Connelly, A. (2012). MRtrix: Diffusion tractography in crossing fiber regions. \u003cem\u003eInternational Journal of Imaging Systems and Technology\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e(1), 53\u0026ndash;66. https://doi.org/10.1002/ima.22005\u003c/li\u003e\n\u003cli\u003eTournier, J.-D., Calamante, F., Gadian, D. G., \u0026amp; Connelly, A. (2004). Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(3), 1176\u0026ndash;1185. https://doi.org/10.1016/j.neuroimage.2004.07.037\u003c/li\u003e\n\u003cli\u003eTournier, J.-D., Smith, R., Raffelt, D., Tabbara, R., Dhollander, T., Pietsch, M., Christiaens, D., Jeurissen, B., Yeh, C.-H., \u0026amp; Connelly, A. (2019). MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e202\u003c/em\u003e, 116137. https://doi.org/10.1016/j.neuroimage.2019.116137\u003c/li\u003e\n\u003cli\u003eTran, V.-D., Nguyen, T.-N., Ballit, A., \u0026amp; Dao, T.-T. (2023). Novel Baseline Facial Muscle Database Using Statistical Shape Modeling and In Silico Trials toward Decision Support for Facial Rehabilitation. \u003cem\u003eBioengineering\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(6), 737. https://doi.org/10.3390/bioengineering10060737\u003c/li\u003e\n\u003cli\u003eUus, A., Grigorescu, I., Pietsch, M., Batalle, D., Christiaens, D., Hughes, E., Hutter, J., Cordero Grande, L., Price, A. N., Tournier, J.-D., Rutherford, M. A., Counsell, S. J., Hajnal, J. V., Edwards, A. D., \u0026amp; Deprez, M. (2021). Multi-Channel 4D Parametrized Atlas of Macro- and Microstructural Neonatal Brain Development. \u003cem\u003eFrontiers in Neuroscience\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e, 661704. https://doi.org/10.3389/fnins.2021.661704\u003c/li\u003e\n\u003cli\u003eUus, A., Pietsch, M., Grigorescu, I., Christiaens, D., Tournier, J.-D., Grande, L. C., Hutter, J., Edwards, D., Hajnal, J., \u0026amp; Deprez, M. (2020). Multi-channel Registration for Diffusion MRI: Longitudinal Analysis for the Neonatal Brain. In Ž. \u0026Scaron;piclin, J. McClelland, J. Kybic, \u0026amp; O. Goksel (Eds.), \u003cem\u003eBiomedical Image Registration\u003c/em\u003e (Vol. 12120, pp. 111\u0026ndash;121). Springer International Publishing. https://doi.org/10.1007/978-3-030-50120-4_11\u003c/li\u003e\n\u003cli\u003eVeraart, J., Fieremans, E., \u0026amp; Novikov, D. S. (2016). Diffusion MRI noise mapping using random matrix theory: Diffusion MRI Noise Mapping. \u003cem\u003eMagnetic Resonance in Medicine\u003c/em\u003e, \u003cem\u003e76\u003c/em\u003e(5), 1582\u0026ndash;1593. https://doi.org/10.1002/mrm.26059\u003c/li\u003e\n\u003cli\u003eVeraart, J., Novikov, D. S., Christiaens, D., Ades-aron, B., Sijbers, J., \u0026amp; Fieremans, E. (2016). Denoising of diffusion MRI using random matrix theory. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e142\u003c/em\u003e, 394\u0026ndash;406. https://doi.org/10.1016/j.neuroimage.2016.08.016\u003c/li\u003e\n\u003cli\u003eVidt, M. E., Santago, A. C., Marsh, A. P., Hegedus, E. J., Tuohy, C. J., Poehling, G. G., Freehill, M. T., Miller, M. E., \u0026amp; Saul, K. R. (2018). Modeling a rotator cuff tear: Individualized shoulder muscle forces influence glenohumeral joint contact force predictions. \u003cem\u003eClinical Biomechanics (Bristol, Avon)\u003c/em\u003e, \u003cem\u003e60\u003c/em\u003e, 20\u0026ndash;29. https://doi.org/10.1016/j.clinbiomech.2018.10.004\u003c/li\u003e\n\u003cli\u003eVlachopoulos, L., L\u0026uuml;thi, M., Carrillo, F., Gerber, C., Sz\u0026eacute;kely, G., \u0026amp; F\u0026uuml;rnstahl, P. (2018). Restoration of the Patient-Specific Anatomy of the Proximal and Distal Parts of the Humerus: Statistical Shape Modeling Versus Contralateral Registration Method. \u003cem\u003eJournal of Bone and Joint Surgery\u003c/em\u003e, \u003cem\u003e100\u003c/em\u003e(8), e50. https://doi.org/10.2106/JBJS.17.00829\u003c/li\u003e\n\u003cli\u003eYushkevich, P. A., Piven, J., Hazlett, H. C., Smith, R. G., Ho, S., Gee, J. C., \u0026amp; Gerig, G. (2006). User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e31\u003c/em\u003e(3), 1116\u0026ndash;1128. https://doi.org/10.1016/j.neuroimage.2006.01.015\u003c/li\u003e\n\u003cli\u003eZhang, Y., Herbert, R. D., Bilston, L. E., \u0026amp; Bolsterlee, B. (2023). Three-dimensional architecture of the human subscapularis muscle in vivo. \u003cem\u003eJournal of Biomechanics\u003c/em\u003e, \u003cem\u003e161\u003c/em\u003e, 111854. https://doi.org/10.1016/j.jbiomech.2023.111854\u003c/li\u003e\n\u003cli\u003eZhang, Y., Herbert, R. D., Bilston, L. E., \u0026amp; Bolsterlee, B. (2024). Three‐dimensional architecture and moment arms of human rotator cuff muscles in vivo: Interindividual, intermuscular, and intramuscular variations. \u003cem\u003eJournal of Anatomy\u003c/em\u003e, joa.14050. https://doi.org/10.1111/joa.14050\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Rotator cuff muscles, fibre orientation, muscle architecture, musculoskeletal imaging, diffusion-weighted imaging, multi-channel registration","lastPublishedDoi":"10.21203/rs.3.rs-4683327/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4683327/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMeasurements of muscle architecture are crucial for understanding muscle function but are often difficult to obtain in human muscles \u003cem\u003ein vivo\u003c/em\u003e. This study aimed to create population-averaged atlases of human rotator cuff muscle shape and muscle fibre orientations from anatomical magnetic resonance images (MRI) and diffusion-weighted images (DWI), and to utilize these atlases to predict muscle fibre orientations from anatomical MRI data alone. An image registration framework was applied to co-register anatomical MRI and DWI data of 11 male and 9 female subjects into sex-specific common spaces, forming the basis for the atlases. The accuracy of registration was quantified using Dice coefficients, angular correlation coefficients (ACCs), and angular differences. The same metrics were used to assess the capability of the atlases to predict fibre orientations for subjects not included in the atlas construction, via leave-one-out cross-validation. The results showed that individual male and female image data were accurately registered into their respective atlas spaces, with high Dice coefficients (0.888 ± 0.002 for males, 0.856 ± 0.021 for females) and consistent angular alignment as evidenced by the ACCs and angular differences. Predicted fibre orientations for out-of-sample subjects closely matched those derived from DWI images, exhibiting improved smoothness and coverage (ACC: 0.909 ± 0.011 for males, 0.942 ± 0.011 for females; angular difference: 13.8 ± 1.3° for males, 11.2 ± 1.2° for females). These findings demonstrate that population-averaged atlases not only enhance muscle architecture reconstructions but also enable the accurate prediction of muscle fibre orientations using only anatomical MRI scans.\u003c/p\u003e","manuscriptTitle":"Prediction of rotator cuff muscle fibre orientations using a population-averaged atlas generated with anatomical and diffusion-weighted magnetic resonance images","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-29 21:18:08","doi":"10.21203/rs.3.rs-4683327/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1da569ce-cd44-4f72-bd75-26adb42f45c3","owner":[],"postedDate":"July 29th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-09-24T20:08:14+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-29 21:18:08","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4683327","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4683327","identity":"rs-4683327","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}