Reference data on in vitro anatomy and indentation response of tissue layers of musculoskeletal extremities

The skin, fat, and muscle of the musculoskeletal system provide essential support and protection to the human body. The interaction between individual layers and their composite structure dictate the body’s response during mechanical loading of extremity surfaces. Quantifying such interactions may improve surgical outcomes by enhancing surgical simulations with lifelike tissue characteristics. Recently, a comprehensive tissue thickness and anthropometric database of in vivo extremities was acquired using a load sensing instrumented ultrasound to enhance the fidelity of advancing surgical simulations. However detailed anatomy of tissue layers of musculoskeletal extremities was not captured. This study aims to supplement that database with an enhanced dataset of in vitro specimens that includes ultrasound imaging supported by motion tracking of the ultrasound probe and two additional full field imaging modalities (magnetic resonance and computed tomography). The additional imaging datasets can be used in conjunction with the ultrasound/force data for more comprehensive modeling of soft tissue mechanics. Researchers can also use the image modalities in isolation if anatomy of legs and arms is needed.


Background & Summary
The musculoskeletal system is a multi-layer support and protective structure composed of skin, fat and muscle. The tissue properties of individual layers and composite structure influence the musculoskeletal system's ability to respond to mechanical loading, in some cases, loads that may result in injury. Extremities are prone to injuries that range from gun-shot wounds or blast injuries 1,2 to minor cuts and bruises and require surgical intervention depending on severity. High fidelity surgical simulations can enhance surgical training by representing real tissue characteristics.
The past two decades have seen significant advancements in virtual reality haptic feedback simulations that aim to increase the effectiveness of surgical training 3 , and long before that, soft tissue characterization was possible using continuum mechanics techniques 4,5 . However, a foundational database containing accurate tissue thickness and mechanical properties did not exist until recently 6 . To obtain accurate tissue thickness and mechanical properties, non-destructive techniques have been used, e.g., imaging. Ultrasound is a cost-effective and non-invasive imaging modality best used to investigate soft tissue properties. If used in conjunction with force feedback the user can account for the applied load deforming tissue during imaging 7 . Computer tomography (CT) and magnetic resonance (MR) imaging are also used to investigate soft tissue properties and, although more cost prohibitive, allow accurate measurement of tissue thickness. These imaging modalities also have the added benefit of being three-dimensional (3D) image volumes and can be segmented to reconstruct detailed 3D anatomical models.
In a recent and similar study completed by the authors 6 , instrumented ultrasound and anthropometric measurements were performed to quantify multi-layer and aggregate tissue thickness as well as characterize the mechanical properties of the musculoskeletal extremities in vivo. The goal of the current study was to expand and improve existing datasets by performing more extensive experimentation in vitro. Additional imaging modalities were included to allow tissue thickness comparison or individual analyses by implementing methods of spatial association via implanted fiducial markers.
A related dataset 8,9 examined surgical tool kinetics and kinematics as well as surface strain measurements using the same specimens from the study presented in this data descriptor. Briefly, three instrumented surgical tools, including a scalpel, retractor and forceps, as well as an indenter, were used to perform seventeen different surgical acts on the skin, fat, and muscle layers of cadaveric legs. The data collected includes forces acting at the tool tip and motions of the tool with respect to the specimen. Surface strain measurements were also captured for indentation, cutting, and pinching surgical acts.

Methods
Specimen overview. The right arm and leg of nine cadaver specimens (each set from the same donor) were obtained with the approval the Human Research Protection Office of the U.S. Army. The donors consisted of four females and five males with an average age of 51.6 years (range 39-65 years) and an average body mass index (BMI) of 23 (range 19-25). Exclusion criteria included any past injuries or surgeries performed on extremities, as well as having a BMI outside the range of 18 to 29. The criteria limited adverse visual artifacts and ensured the full depth of soft tissue to be imaged via ultrasound, which, unlike MR and CT imaging, has depth limitations. Experimentation workflow. Experimentation started with specimen preparation, followed by CT imaging (Fig. 1), and finally MR imaging (Fig. 1). The specimen was then stored in a custom-made fixture (Fig. 2) in a walk-in refrigerator overnight due to the time required for imaging. The next day, anthropometric measurements were recorded followed by ultrasound imaging.
Specimen preparation. When specimens were received, they were kept in a freezer at −20 °C and removed twenty-four hours prior to experimentation. A custom-made fixture (Fig. 2) was used to support the specimen and maintain consistent and neutral anatomical position throughout testing and transportation. The disarticulated limbs did not naturally rest in anatomical supine position, so the femoral and humoral heads were rigidly attached to height adjustable 3D printed mounts on the fixture. Additionally, the foot and hand were secured All CT and MR imaging modalities of both leg (top 3 rows) and arm (bottom 3 rows). Row labels are as follows: T1: T1 weighted magnetic resonance, FS: T1 weighted magnetic resonance with fat saturation, CT: computed tomography. The sagittal cross sections are shown to the left (viewing from medial to lateral), and axial to the furthest right (viewing from inferior to superior). Directional arrows show where the axial sections were taken along the length of the sagittal plane. Note the leg MR images were separated into upper and lower segments due to the limited field of view of MR machines. Also shown in the fourth row (CT arm) is an axial cross section of a phantom object used in all CT scans.
www.nature.com/scientificdata www.nature.com/scientificdata/ to further limit movement during transportation and imaging. The leg fixture also included a height adjustable rod for the greater trochanter so that it would rest lower than the femoral head, much like the typical anatomical supine position. Each leg and arm were equipped with hollow, 3D printed, MR and CT visible spherical markers (see Figs. 1-3) having a 10.0 mm radius and secured with nylon screws. The leg had twelve markers (six on the femur and six on the tibia) and the arm had seven markers (four on the humerus and three on the radius) (Fig. 3). The number of markers used in each was limited by the size of the humerus and radius as well as the desire for not disrupting the region of interest. A minimum of three markers was needed to establish a coordinate system, but more were used to reduce registration errors between imaging data and motion tracking system.
The specimens were equipped with two sets of smart cluster triads composed of nine LEDs each, by Optotrak Certus (Northern Digital Inc., Waterloo, Ontario) after CT and MR imaging. For the leg, the triads were placed at the femoral head and on the mid-shaft of the tibia (Fig. 2). For the arm, the triads were placed on the humeral head and distal end of the radius (Fig. 2). Placement of the triads was chosen to limit interference with desired ultrasound locations while remaining visible to the Optotrak camera system during experimentation.
cT and MR image acquisition. Both CT and MR protocols were selected to enhance soft tissue clarity and tissue differentiation of the skin, fat, and muscle layers. CT imaging was conducted at the Cleveland Clinic, Fig. 2 The custom fixture used for imaging. Two variations are shown, (a) with arm attachments and (b) with leg attachments. Each is fully adjustable to accommodate anatomical differences in a variety of specimens. The foot and hand were secured with Velcro, while the femoral and humeral heads were screwed in to a 3D printed receiver with nylon screws. All fixture materials were non-metal to reduce imaging artifacts in CT and MRI. Also shown are the instrumented ultrasound and Optotrak LED triads. www.nature.com/scientificdata www.nature.com/scientificdata/ Cleveland, Ohio using a SOMATOM Definition Edge 128 scanner (Siemens Healthcare, Erlangen, Germany) with a 0.6 mm slice thickness and no gap using a B40 reconstruction kernel. A phantom was placed in the region of interest with three rods (0, 500, and 1000 mg HA/cm 3 ). The axial in-plane resolution ranged from 0.51 mm to 0.59 mm depending on the reconstruction size of the area of interest. The entire specimen i.e., upper and lower appendage, was scanned in a single pass.
MR imaging was conducted at University Hospitals, Cleveland, OH using a 3T Skyra (Siemens Healthcare, Erlangen, Germany) with a 2 mm slice thickness, 1 mm gap, and an axial in-plane resolution of 0.5 mm. Two imaging protocols were used; T1 weighted without fat saturation and T1 weighted with fat saturation (FS). The repetition time (TR) and echo time (TE) were 500 ms and 11 ms respectively. The specimens were scanned by segment i.e. upper/lower arm/leg in twenty-five slice increments due to the limited field of view of MR imaging with the desired slice thicknesses. Each segment was scanned twice before moving to next section to include both T1 weighted and T1 weighted with FS. Depending on the size of the specimen, each segment was scanned in four to five sections. The specimen was not moved between section scans to maintain alignment in a single imaging coordinate system. A custom Python script (http://www.python.org) concatenated the sections automatically so that the each segment could be viewed as one continuous scan. anthropometric measurements. Circumference and length measurements were recorded for each specimen using a soft tape measure. Each appendage was nominally divided into an upper and lower segment, e.g. the upper arm was elbow to humoral head. At each segment, the circumference and the distance to the most superior landmark were recorded at the proximal, middle, and distal locations in addition to the total length (Fig. 3). The landmarks used were identical to our previous in vivo study 6 with exception to the inferior lateral acromion process, where the current study used the head of the humerus (the scapula was removed prior to experimentation).
Ultrasound image acquisition. Ultrasound imaging was performed using an ACUSON S3000 TM (Siemens Healthcare, Erlangen, Germany), instrumented 7 with a six degree-of-freedom (6-DoF) Nano25 load cell (ATI Industrial Automation, Apex, North Carolina), a VN-100 inertial measurement unit (IMU) (VectorNav, Dallas, Texas), and an optical tracking smart cluster (Northern Digital Inc., Waterloo, ON) (Fig. 2). Depending on the depth of tissue to be imaged, a 9L4 transducer (head dimensions 45 mm × 15 mm) or 14L5 transducer (head dimensions 45 mm × 8 mm) was used. The probe used for each trial is identified in the trial specific state configuration file.
A gravity compensation algorithm was used in a custom program based on simVITRO ® LabVIEW packages (Cleveland Clinic, Cleveland, Ohio) to offset the weight of the ultrasound probe in any orientation. The attached IMU and load cell first measured the probe weight in mid-air, in six orientations, to determine a spatial relationship between probe weight measured by the load cell and orientation. The weight of the probe in any orientation became a function of orientation, therefore in any orientation the probe weight could be subtracted. This method of weight compensation was previously used and validated 6,7 .
Just prior to ultrasound imaging, the 3D printed spherical markers on the specimen were digitized in the motion capture coordinate system to spatially relate the optically tracked ultrasound to the imaging coordinate systems. In addition, a bone coordinate system was setup for each segment, by digitizing bony landmarks (Table 1), to spatially relate the optically tracked ultrasound and digitized points to the relevant bone position. The upper and lower appendages were treated as separate rigid bodies.
Soft tissue thicknesses and indentation response were captured through freehand anatomical ultrasound imaging and indentation ultrasound imaging, respectively. Anatomical ultrasound imaging (Fig. 4) was performed at the anterior, posterior, medial, and lateral positions at proximal, middle, and distal regions of each upper and lower leg and arm, totaling 24 per appendage. The applied load during anatomical imaging was minimized to limit soft tissue deformation and was considered minimal if the measured load was less than 2N; measured by the attached load cell. Indentation ultrasound imaging (Fig. 4) was performed at four locations per arm and per leg, the central anterior and central posterior region of each segment. The indentation trials were www.nature.com/scientificdata www.nature.com/scientificdata/ performed immediately following the respective force trials of the same location to ensure minimal changes in location. It is important to note that the loading rate of the ultrasound probe was not controlled, due to the freehand nature of ultrasound imaging. A limitation of this dataset is the lack of indentation response collected at varying loading rates. While this data may be important to capture the full mechanical behavior of the soft tissue indentation response, the data included in this dataset provides a realistic field assessment of the indentation response without the need for a sophisticated loading mechanism.
The total time of each image acquisition was set to eight seconds. Once the target region e.g. upper arm was located, the probe was oriented so that the target bone (region dependent, shown in Table 2) was visible. The long axis of the probe was oriented along the bone, therefore, the bone appeared as a flat and bright landmark spanning the width of the image. The depth, gain, and focus were adjusted so that the bone was visible and the multi-layer delineation between skin, fat, and muscle was as clear as possible.
The specimen remained in its fixture for accessing the medial, anterior and lateral positions. It was then rotated such that the posterior faced upwards to reveal six posterior imaging locations. Spatial alignment. The spherical markers, which can be seen in Fig. 3, were used to establish a relationship between the coordinate system of motion capture system, bone coordinate system, ultrasound acquisition and the anatomical imaging coordinate systems (CT and MR). The boundaries of the spheres were extracted from the MR and CT images using image segmentation (3D Slicer 10 -http://www.slicer.org), resulting in STL surfaces in each image coordinate system. A custom Python script was used to perform the spatial alignment for each segment between the various coordinate systems, i.e. digitized points from bone coordinate system (DG) to CT, DG to MR, and MR to CT. First, the digitized points on the spheres were transformed from the global motion capture coordinate system (set by the camera) to the appropriate bone coordinate system (where the origin is at the knee or elbow). A sphere was fit to both the STL surfaces and the digitized points and a transformation  www.nature.com/scientificdata www.nature.com/scientificdata/ matrix was established between the bone coordinate system and the image coordinate system using the singular value decomposition method 11 . This transformation matrix, along with the others saved during data collection, allow the user to examine the data (positions, forces/moments, images) in their desired coordinate system. The alignment allows users to compare images in the same location across all three modalities. As previously mentioned, the appendages were nominally divided in to upper and lower segments. Each segment was treated as a rigid body and each segment had its own set of spherical markers. The project wiki provides additional details on the workflow of the spatial alignment process (https://simtk.org/plugins/moinmoin/multis/Specifications/ DataAnalysis/Registration).
Thickness analysis. The multi-layer tissue thicknesses were determined using previously established ultrasound thickness measurement methods 6 . Briefly, the graphic user interface was used to display the ultrasound frame(s) at the minimum applied force for anatomical trials and throughout the compression portion of the indentation trials. In each image, moveable red dots were placed at each of the boundaries to represent skin, fat, and muscle thicknesses. The script was modified to store ultrasound position and orientation, in addition to the forces, moments, and thicknesses for each frame in an XML format. Soft tissue mechanics may be extracted from the summarized data, using the deformation defined by the thickness change during indentation and the applied force. In addition, the experiment motion may be re-created virtually or physically using the position and orientation data of the ultrasound probe in relation to the limb.

Data Records
A set of naming conventions was created to help organize and later navigate all raw and derivative data 12 . Each de-identified donor was given a unique ID and identical folder structure. It should be noted that the structure is nearly identical in name and description as the in vivo reference data 6 with several additional folders and modifications to descriptions to reflect the in vitro data. An asterisk (*) denotes an addition or modification to the format of the in vivo data structure.
Raw data. 1. Data* -Probe load and orientation data in raw and transformed (ultrasound probe tip coordinate system with weight compensation) representations. Additionally, Optotrak motion data was included (.tdms). For clarification, this directory was taken directly from the data collection software, however it does not contain all forms of data collected within this dataset (see additional directories below). It was not re-named to avoid manual data entry errors. 2. Configuration -Contains the sensor and state files used during data collection for each data file (.cfg). The sensor and state files provide the structure and description of the binary data file (.tdms) for the raw and transformed data, respectively. Also contains the subject configuration file [.cfg and .xml, which provides the file locations (state.cfg, sensor.cfg, data.tdms) of each data collection site and de-identified subject information (demographics in Table 3   www.nature.com/scientificdata www.nature.com/scientificdata/ system. Additionally, this figure includes six plots depicting Optotrak motion (translation and rotation) of the bone, ultrasound probe, and ultrasound probe relative to bone for each trial. Spatial alignment: • Marker STL's -Segmented spherical markers of both CT and MR used to align limb segments (STL) • Quality Check -The distances between each marker, radius of each marker, and summary statistics provided in a text document (TXT) • Ultrasound Positions -Provides the ultrasound location in CT and MR for each ultrasound trial in the form of a visual toolkit file. This can be used with the provided Python scripts to show the CT or MR image in the same location at the same angle as recorded by ultrasound (XML).
Thickness analysis: • TissueThickness -Sub-directory that contains all thickness analysis.
• UltrasoundManual -Manual identification of tissue boundaries from ultrasound imaging of each trial (XML) • ThicknessPNG -Provides image of the first ultrasound frame that was analyzed (corresponding to minimum force or initiation of indentation, for anatomical and indentation trials, respectively) and a plot of force magnitude vs. thickness (skin, fat, muscle, and total).
Raw and derivative data files exist for each trial with the following root naming convention: "Experiment Run Number"_"Donor ID"_"Limb Segment"_"Location"_"Test Type"-"Trial Number Index" • Experiment Run Number: 3-digit number that gets auto-incremented for each tdms file (order in which data are collected for each specimen). • Donor ID: 3-digit number that gets assigned to each donor in the order that they were acquired. . if the location is tested more than once the subsequent data files will be -2, -3, etc.
Additional characters are appended to the root name for further data descriptions. Example file names for a single trial are included in Table 4.
CT and MR imaging files use the following naming convention: "Donor ID"_"Limb_Segment"_" Modality".nii   The files used for spatial alignment are found in a Registration folder. Example file names for a single donor are included in Table 5.
The project website (https://simtk.org/projects/multis) contains all specifications as well as experiment notes. The related dataset 8 on surgical tool forces is also described on the project website.

technical Validation
The ultrasound images, load cell forces, and Optotrak motion tracking represent three separate instruments that must be temporally synchronized to be of significant use. The temporal synchronization of force and ultrasound imaging was established using a custom analog signal routed through each system 7 . Further validation was required to ensure temporal synchronization between the Optotrak motion and instrumented ultrasound system. The motion data was routed through its native software, NDI First Principles (Northern Digital Inc., Waterloo, Onterio), to the custom LabVIEW (National Instruments, Dallas, Texas) program used to record all force and orientation data. A tap test was performed to determine the temporal difference between force and motion. The load cell was physically and instantaneously tapped with enough force to move the instrumented ultrasound system, thereby causing a noticeable change in motion and force. A custom Python script was developed to find the exact time each spike in motion and force occurred. The mean of three trials was 43.0 +/− 16.3 ms. The force was measured with a sampling rate of 1,000 Hz and motion roughly 33 Hz, therefore the error was expected to be close to that of the lower rate (33 Hz or 30 ms).
The error of spatial alignment in each limb segment was analyzed by calculating the vector between all pairs of markers (e.g. the lower arm pairs include R1 to R2, R1 to R3, and R2 to R3). The distance (magnitude of the vector) was compared between each coordinate system, i.e. DG to MR, DG to CT, and MR to CT. In addition, the angle difference was calculated using the dot product between the line segments in each coordinate system. The average distance differences as well as angles across all thirty-six segments are presented in Table 6.
The thickness analysis methods, including software and observers, were identical to the authors previous work 6 . In the previous study, repeatability was evaluated using all anatomical and indentation trials from 5 random patients and the inter-and intra-observer variability was measured using two observers. The inter-observer mean and standard deviation thickness measurements for anatomical trials was 0.36 ± 0.38, 0.78 ± 1.10, 0.65 ± 1.11, and 0.32 ± 0.38 mm for skin, fat, muscle, and total tissue thickness respectively. Likewise, the intra-observer was 0.16 ± 0.17, 0.49 ± 1.19, 0.48 ± 1.19, and 0.19 ± 0.19 mm. The thickness change during indentation was measured from the beginning of indentation to maximum force. The inter-observer mean and standard deviation thickness change for skin, fat, muscle, and total thickness was 0.15 ± 0.19, 0.29 ± 0.32, 0.31 ± 0.31, and 0.42 ± 0.37 mm, respectively. Likewise, the intra-observer mean and standard deviation was 0.10 ± 0.12, 0.24 ± 0.24, 0.32 ± 0.25, and 0.29 ± 0.23 mm.  Table 5. Example filenames for spatial alignment data.  Table 6. The mean and standard deviation of distance differences, percentage differences, and angle differences when comparing the thicknesses measured in the bone coordinate system, CT and MR via the segmented spherical markers. Note: DGCT = digitized points to CT. DGMR = digitized points to MR. MRCT = MR to CT.

Usage Notes
All data can be found on the active working version of the data management site: https://multisgamma.stanford. edu/. A static version of the complete dataset is also provided 12 .
More detailed specifications including but not limited to Python scripts, hardware integration and design, the custom LabVIEW program and more can be found on the project website (https://simtk.org/plugins/moinmoin/multis/). A source code repository with downloadable scripts and data can be found at https://simtk.org/ scm/?group_id=1032.

Code availability
The source code repository is available at (https://simtk.org/projects/multis). Additionally, two static packages containing thickness analysis software and all data/source code used in the technical validation section are available (https://simtk.org/frs/?group_id=1032). In order to use the provided scripts, one needs to install Python 2.7 with Anaconda (https://docs.anaconda.com/anaconda/). The two additional libraries required to read ultrasound and data files respectively are DICOM and npTDMS. Additional libraries used in the provided scripts can be installed using Anaconda ("conda install" command).