Variation of the clavicle’s muscle insertion footprints – a cadaveric study

The muscle footprint anatomy of the clavicle is described in various anatomical textbooks but research on the footprint variation is rare. Our goal was to assess the variation and to create a probabilistic atlas of the muscle footprint anatomy. 14 right and left clavicles of anatomical specimens were dissected until only muscle fibers remained. 3D models with muscle footprints were made through CT scanning, laser scanning and photogrammetry. Then, for each side, the mean clavicle was calculated and non-rigidly registered to all other cadaveric bones. Muscle footprints were indicated on the mean left and right clavicle through the 1-to-1 mesh correspondence which is achieved by non-rigid registration. Lastly, 2 probabilistic atlases from the clavicle muscle footprints were generated. There was no statistical significant difference between the surface area (absolute and relative), of the originally dissected muscle footprints, of male and female, and left and right anatomical specimens. Visualization of all muscle footprints on the mean clavicle resulted in 72% (right) and 82% (left) coverage of the surface. The Muscle Insertion Footprint of each specimen covered on average 36.9% of the average right and 37.0% of the average left clavicle. The difference between surface coverage by all MIF and the mean surface coverage, shows that the MIF location varies strongly. From the probabilistic atlas we can conclude that no universal clavicle exists. Therefore, patient-specific clavicle fracture fixation plates should be considered to minimally interfere with the MIF. Therefore, patient-specific clavicle fracture fixation plates which minimally interfere with the footprints should be considered.


Materials and Methods
Reverse engineering workflow. All research was performed in accordance with guidelines and regulations of our institution (KU Leuven/UZ Leuven), and we confirm that informed consent was obtained from all participants who donated their body to the Anatomical Skills Lab (KU Leuven). After approval from the Ethical Committee (Commissie Medische Ethiek, UZ Leuven/KU Leuven) 14 right and 14 left clavicles were dissected from 14 phenol preserved cadavers (7 male and 7 female cadavers). In a first stage all gross fat tissue, ligamentous and soft tissue structures apart from muscle fibers were removed. In the second stage the sternoclavicular and acromioclavicular ligaments were cut, the clavicle was removed from the thorax and the dissection was meticulously continued until only muscle fibers and bone remained. The dissection was performed by the author MH and supervised and assessed by authors PH & SN.
We created a workflow in order to reverse engineer the anatomical specimens in to usable 3D models, which is presented in Fig. 1. First, all clavicles were CT-scanned using a Siemens SOMATOM Definition Flash CT scanner with a 1 mm slice thickness and a resolution of 512 × 512 pixels. The CT images were then imported and segmented in Mimics Research 18.0 (Materialise, Leuven) to create 3D model bones. These STL files (3D mesh file format) were than remeshed using 3-matic 12.0 (Materialise, Leuven) to produce high quality STL files.
The dissected clavicles were then fixed in a frame in an upright position and optically scanned from 18 different angles using a Coord-3 CMM with LC60Dx Laser scanner (Nikon Metrology NV) 20 . The obtained point clouds of the surface model were meshed, optimized and exported as STL files using Focus Inspection v9.3 (Nikon Metrology NV).
Then, while remaining fixed in the frame, a large amount of photos were taken from different angles all around the clavicle (average of 192 photos per clavicle). The photos of each clavicle were imported into Agisoft photoscan (Agisoft, St Petersburg Russia) and exported as 3D photogrammetry PDF files (Adobe).
STL files of the soft tissue and bone were loaded into 3-matic (Materialise, Belgium) and aligned using the global registration module until an optimal alignment was obtained. Alignment of the bone with the soft tissue STL was judged by using the corresponding photogrammetry model. When necessary, small manual adjustments were made until optimal alignment was obtained.
The location of the attachment sites was obtained by subtracting the bone STL from the soft tissue STL. Based on the photogrammetry model, the STL model was manually improved until it represented the appearance of the photogrammetry model.
Basic measurements such as the surface of the muscle footprint and the surface and volume of the clavicle were performed.
After creation of the STL files, one clavicle was non-rigidly registered to the other 13 clavicles using the algorithm based on Danckaers et al. 21 . The mean clavicle was calculated and then this clavicle (further referred to as 'source') was non-rigidly registered to all 14 left and 14 right cadaver clavicles respectively(further referred to as 'target') 22 . Next, the muscle footprints of each cadaver clavicle were copied to this corresponding registered mesh. Since the non-rigid registration algorithm assures one-on-one mesh correspondence between source and target, muscle footprints of the target can also be highlighted on the source by using the vertex indexes. This was repeated for all 14 clavicles and resulted in a surface footprint map. The framework of highlighting the muscle attachment sites is presented in Fig. 2.
The probabilistic area was constructed by averaging the attachment region by the number of muscles contributing to that particular attachment region. This procedure was performed for the left and right set of clavicles.

Additional information.
There are no competing financial or non-financial interests for any of the authors regarding the performed research presented in this manuscript.

Results
The muscle footprint surfaces and the surface and volume of the clavicles are shown in Table 1. A mean coverage of 37.5% in right and 35.5% in left clavicles was observed in the original clavicles.
The mean absolute and relative surface areas of the respective male and female clavicle's were compared using a paired t-test. There was a significant difference between the volume and surface area of the male and female clavicle's and the footprint of the left deltoid muscle. (Tables 1 and 2). A paired t-test was also performed to detect significant differences in surface area between left and concomitant right muscle attachment sites but it did not reach statistical significance <0.05 for any of the muscle attachment sites. Detailed results of each muscle for each target can be found in the supplement table.
The mean area (±standard deviation) of the muscle attachment sites for the right and left clavicle on the source can be found in Table 3. The MIF of each specimen covered on average 36.9% of the average right and 37.0% of the average left clavicle.
Visualization of all muscle footprints on the right average clavicle resulted in 72% coverage of the surface, visualizing only each muscle's largest muscle footprint led to 52% coverage, which can be seen in Figs 3 and 4 respectively. All muscle footprints on the left average clavicle resulted in 82% coverage of the surface, each muscle's largest muscle footprint led to 49% coverage.
In Fig. 5 the resulting probabilistic atlas of the right clavicle is shown, in Fig. 6 the resulting probabilistic atlas of the left clavicle is shown.

Discussion
The goal of our study was to describe the variability of muscle attachment sites through the development of a probabilistic atlas of the muscle footprints on the clavicle. The large differences in muscle footprint areas, as shown by the standard deviation, already indicate their variability. The difference between coverage by all MIF and the average MIF area, shows that MIF location varies strongly as well.
Many previous studies described the bony anatomy of the clavicle and searched for constants in its highly variable anatomy in order to produce better fitting implants 13,14,22 . Simon Lambert et al. described an implant preferred pathway (IPP) which was defined as a continuous linear region where the least possible soft tissue disruption would be necessary for plate fixation 13 . The location of this IPP was based on expert opinion, defined on one clavicle and then registered through a non-rigid registration method to 174 clavicles. Fatah et al. described the bilateral direction asymmetry of clavicles using a statistical bone atlas 23 . They calculated a mean bone and used this to localize five muscle attachment sites based on a standard anatomical textbook. These sites were then propagated across their entire sample to generate correspondence between homologous anatomical sites. Our results support the existence of an implant preferred pathway as defined by Lambert et al. but there is an important remark to make. Lambert and Fatah both assumed that the muscle footprint and implant preferred pathway is not related to the shape of the bone and that a universal clavicle exists, which is in contradiction to our results, as no general IPP could be identified and a lot of variation is present in the muscle footprints on the mean clavicle. The creation and usage of the 3D muscle insertion footprint models in this study allowed the authors to exactly describe the location of the muscle on the bone. As the clavicle is a bone (or free shape from a mathematical/ engineering point of view) with a very large shape variation, the position of a point/surface/muscle footprint cannot be easily described nor can it easily be extrapolated to another bone. The usage of a SSM and the according Non-Rigid-Registration allowed us to compare the position of the muscle insertion footprints on these highly variable shapes (clavicles) with each other. This reverse engineering technique requires the usage of high quality 3D models. Non-Rigid Registration of the muscle attachment sites: First, the source (mean) mesh was transformed to the target mesh through a non-rigid registration method. The transformed mesh was then nearly identical to the target mesh (registered mesh). The muscle footprints of the target mesh were then indicated on the registered mesh.Through the one-to-one mesh correspondence between source and the registered meshes, muscle attachment sites in the registered mesh can also be highlighted in the mean mesh using vertex indices.. This was repeated for every 14 target meshes in order to create a muscle footprint atlas on the source (mean) mesh.    24 . In their study they used 8 cadaver specimens on which the muscle attachment regions were defined and a muscle probabilistic atlas was constructed. For each muscle in the probabilistic atlas, they created an average muscle footprint and used this to predict patient-specific muscle footprints. They suggested that the probabilistic atlas could be useful to estimate these patient-specific attachment regions. Prediction of the clavicle's muscle footprint will be a topic of future research.
Two recent meta-analyses investigated the effect of anteroinferior versus superior plating of the clavicle and they both concluded that plating along the superior and anteroinferior aspects of the clavicle leads to similar operative outcomes with respect to union, nonunion, malunion, and implant failure 25,26 . Only the meta-analysis by Nourian et al. concluded that patients in the superior plating group had a significantly higher probability of suffering from symptomatic hardware as compared to patients in the anteroinferior plating group 25 . Baltes et al. looked in detail at the influence of plate positioning on hardware removal rate but could not conclude that it leads to reduced implant removal rates 27 . From an anatomical point of view our results support a superior plating   position as this would lead to a less extensive muscle dissection. We do believe however that plate prominence is key to reduce soft tissue irritation. Galdi et al. compared the results of plates positioned anteroinferiorly of 2.7 mm vs 3.5 mm and concluded that the 2.7 mm plates resulted in excellent clinical outcomes but with a higher cosmetic applicability and a lower implant removal rate 28 .
Havet et al. described the periosteal vascularization of the middle one third of the clavicle 29 . They showed that main blood supply was periosteal and arose from two different origins. First, the posteroinferior surface of the clavicle is vascularized by the middle part of the suprascapular artery in all their cases. Secondly, in 75% of the cases the anterosuperior surface of the middle third was supplied by anastomosing vessels between the medial and lateral branches of the thoracocarcomial trunk. Because of this periosteal vascularization they conclude that there is a high risk of disruption of the periosteal blood supply in case of severely displaced fractures. These anatomical findings explain why in several randomized trials where operative vs conservative treatment of more than 2 cm displaced fractures resulted in non-union of the conservatively treated fractures 30,31 . Research has shown that muscle is capable of supplying osteoprogenitor cells in cases where the periosteum is insufficient, and the muscular osteoprogenitors possess similar osteogenic potential to those derived from the periosteum 32 . We believe that successful clavicle fracture treatment relies on anatomical preservation, meaning bony stability and soft tissue recovery in order to prevent non-union. Furthermore, disruption of muscular footprints accompanied by compromised perfusion contributes to the risk of complications 13 .
Goudie et al. tried described the relationship between clavicle shortening and functional outcome based on 3D-CT measurements 33 . They concluded that a reduced clinical outcome was predominantly mediated by the development of a non-union and see the prevention of non-union as the holy grail in successful clavicle fracture treatment. Although we agree with their findings, we believe that the reduced outcome is not only related to non-union but also to severe malunion which can lead to a change in preload of the muscles attaching to the clavicle and its long-term effect on arthrosis of the nearby glenohumeral joint 34,35 .
The strength of our study lies in its innovative character and the use of reverse engineering and non-rigid registration to describe the clavicle's soft tissue anatomy. To our knowledge, this is the first study that describes the muscle footprint variation of the clavicle. It's major weakness is the relatively small number of specimens used. This was due to the time consuming reverse engineering process and the technical pitfalls to standardize the process that came along with it. Although only a relatively small number of clavicles were used we were still able to demonstrate the large variability in muscle footprint anatomy.
In conclusion, our anatomical description encourages the use of fixation plates or intramedullary devices that respect the muscle footprints for the surgical treatment of displaced midshaft clavicle fractures. Soft tissue respecting fixation devices could lead to an even larger reduction of fracture non-union compared with conventional fixation devices and conservatively treated fractures. Since the probabilistic atlas showed the high variability of clavicle muscle footprints, patient-specific clavicle fixation plates that not only take the shape of the bone but also the position of the muscle footprints into account could potentially reduce surgery time, improve anatomical alignment, preserve muscle length and preload, and reduce the rate of non-union. We believe that our results create a basis for an improved approach in clavicle fracture research and optimizing the currently existing hardware.