The Natural Historian's Guide to the CT Galaxy: Step-by-Step Instructions for Preparing and Analyzing Computed Tomographic (CT) Data Using Cross-Platform, Open Access Software

Synopsis The decreasing cost of acquiring computed tomographic (CT) data has fueled a global effort to digitize the anatomy of museum specimens. This effort has produced a wealth of open access digital three-dimensional (3D) models of anatomy available to anyone with access to the Internet. The potential applications of these data are broad, ranging from 3D printing for purely educational purposes to the development of highly advanced biomechanical models of anatomical structures. However, while virtually anyone can access these digital data, relatively few have the training to easily derive a desirable product (e.g., a 3D visualization of an anatomical structure) from them. Here, we present a workflow based on free, open source, cross-platform software for processing CT data. We provide step-by-step instructions that start with acquiring CT data from a new reconstruction or an open access repository, and progress through visualizing, measuring, landmarking, and constructing digital 3D models of anatomical structures. We also include instructions for digital dissection, data reduction, and exporting data for use in downstream applications such as 3D printing. Finally, we provide Supplementary Videos and workflows that demonstrate how the workflow facilitates five specific applications: measuring functional traits associated with feeding, digitally isolating anatomical structures, isolating regions of interest using semi-automated segmentation, collecting data with simple visual tools, and reducing file size and converting file type of a 3D model.


Introduction
The applications of three-dimensional (3D) visualizations of internal anatomy are varied and vast, spanning a galaxy of analytical possibilities. Recently, the increased ease of gathering such data has led to their widespread adoption in the comparative morphological community. The embrace of this new data type has, in turn, catalyzed many recent biological discoveries, such as revealing brain and muscle activity during bird flight (positron emission tomography scanning; Gold et al. 2016), determining how blood circulates through vasculature (magnetic resonance imaging; O'Brien and Williams 2014; O'Brien 2017), revealing the function of the appendicular skeleton during locomotion and feeding in live sharks (X-ray Reconstruction of Moving Morphology, 3D fluoroscopy coupled with computed tomographic [CT] animation; Camp et al. 2017;Scott et al. 2019) and reconstructing the feeding behavior of long-extinct monsters of the deep (CT imaging of Helicoprion; Tapanila et al. 2013). Other researchers have used 3D digitization to educate and inform. Anatomical models of living and extinct taxa can be built digitally so that students can manipulate, dissect, and scale anatomical structures online (see Rahman et al. 2012;Manzano et al. 2015), used to make 3D prints of missing bones of incomplete physical specimens, or print whole rare or otherwise difficult to acquire specimens for use in teaching comparative anatomy (Gidmark 2019;Staab 2019). For example, the anatomically accurate, 3D printed, vertebrate skull magnetic puzzles by Singh et al. (2019) allow students to understand how different parts of the skull fit together.
Open source efforts like MorphoSource (Boyer et al. 2016;morphosource.org) and DigiMorph (digimorph.org) aggregate thousands of digital 3D models into anatomical libraries and serve them freely to researchers, teachers, and laypersons alike. Like other synthetic, open access approaches to data management and data sharing (Sidlauskas et al. 2009;Whitlock 2011; but see also Hipsley and Sherratt 2019), these repositories encourage data reuse, reanalysis, and reinterpretation, and have ushered in a digital renaissance of comparative morphology.
Most of the 3D images in the online digital libraries result from computed tomography scanning, commonly known as ''CT'' or ''cat'' scanning, which benefits from the quadruple advantages of nondestructivity, shareability, printability, and affordability (Cunningham et al. 2014;Sutton et al. 2014). CT scanning neither invades, modifies, or destroys the original sample. The digital nature of CT data makes it easy to share via open-access platforms and has sparked ''big data'' initiatives, such as oVert (floridamuseum.ufl.edu/overt) and the #ScanAllFishes projects (adamsummers.org/scanallfish). The simplicity of converting CT scans to digital ''surfaces'' allows almost any anatomical structure to be 3D printed, even permitting structures to be artificially warped, scaled, or mirrored to fit experimental or teaching needs (Stayton 2009). Scans can also be converted into digital ''meshes'' which can be used to gather 3D geometric morphometrics data (Lawing and Polly 2010), model the reaction forces on the structures using finite element analysis (FEA; see Hulsey et al. 2008), predict fluid flow around structures using computational fluid dynamics (see Inthavong et al. 2009), study multibody dynamics (Lautenschlager et al. 2016), or render and animate 3D objects (Garwood and Dunlop 2014).
Perhaps most importantly, the decreasing cost, size, and complexity of CT hardware, and the development of open source software like Horos (https:// horosproject.org/) or 3D Slicer (Fedorov et al. 2012;Kikinis et al. 2014, https:// download. slicer. org) has opened access to scientists working outside the biomedical arena. Aspiring digital anatomists no longer need to seek time on the multi-million-dollar, roomsized set-ups in hospitals, but can use desktop machines costing far less. The spread of these smaller systems, often purchased through collaborative interdepartmental funding opportunities, has drastically decreased the cost per study, increased the willingness of researchers to share their data, and caused CT data to explode in popularity, even among scientists who lack access to CT hardware (see Davies et al. 2017). The methods have now transcended biomedical and anthropological research to penetrate fields like organismal taxonomy, paleontology, comparative anatomy, and physiology, as well as biomechanics and biomimetics Divi et al. 2018;Santana 2018;Rutledge et al. 2019). The advantages of biodiversity and taxonomical research cannot be understated as rare, endemic, and understudied taxa can now be shared widely. More open access to specimens allows for systematic hypotheses to be updated, re-examined, and replicated, and each scan preserves an in silico virtual record of morphology for posterity. Metaphorically, each of these virtual specimens can be considered a point of light in a vast and growing constellation depicting the world's biological diversity. Those researchers able to navigate that starfield, which we dub the CT galaxy, will be poised to visualize and analyze biodiversity in ways never before possible.
As is typical when technologies become newly affordable and accessible, the pace of method development has far outstripped the pace of training. Though many researchers and educators have become aware of CT's potential, relatively few have been able to participate in focused training workshops. Strides have been made in establishing best practices in the process of CT scanning itself, and in the curation of 3D data (Davies et al. 2017;Keklikoglou et al. 2019). However, the only available training protocols for analyzing the CT data after they have been gathered have been ad hoc efforts developed within research groups and passed among scientists via email and similar channels. This contribution aims to democratize access to such training by publishing an open-access workflow using freely available and cross-platform software.
Herein, we outline a set of practices in the production, visualization, and analysis of CT data. We have found this workflow saves time and money while maximizing efficiency. We hope that these suggestions tempt the uninitiated to experiment with CT methods for the first time or ease the struggle of learning new techniques. To that end, we focus on those often-tedious nuances of data preparation, formatting, and navigating software that commonly hinder progress in CT-based studies of anatomy, functional morphology, and macroevolution. We also emphasize tools useful for creating pedagogical aids such as 3D prints and images of anatomical structures. Whenever possible, we include steps for data-reduction that help to make processing time more reasonable for older/slower machines, although most any reasonably up-to-date machines (e.g., Mac OS X Lion 10.7.3, Windows 7, Ubuntu 10.10, or newer) can perform all manipulations and analyses herein.

Software
This workflow is designed to be completely open to any researcher, educator, or enthusiast. Generally speaking, the only limitation is access to a computer with at least 8 GB of random access memory (RAM), though this depends mostly on the size of the file to be analyzed. For optimal performance, we recommend that the data file not exceed one-tenth to one-fourth the size of the available RAM on your computer. For example, if you have 8 GB of RAM, your data file should be no larger than 0.8 GB (800 MB) to 2 GB. If the file that you intend to analyze is larger than this range, we include a variety of steps below for down-sampling or working around the computationally and/or memory-intensive steps of CT analysis. There are a variety of software programs available to process CT data, and the programs we employ are freely available and cross-platform (Table 1; see also Abel et al. 2012). Before beginning the workflow, ensure that you have installed the latest stable version of the ImageJ (Schneider et al. 2012;Rueden et al. 2017) expansion Fiji (Schindelin et al. 2012, https:// fiji. sc; we use ImageJ v2.0.0 herein) and 3D Slicer (https:// download. slicer. org; v.4.10.2 used herein). We also recommend that users interested in working with 3D surface meshes install MeshLab (Cignoni et al. 2008;Pietroni et al. 2010, www. meshlab. net;v. 2016.12 used herein). If your computer has a dedicated graphics card, you can use it in Slicer to reduce lag time when rendering your data in 3D. The process for telling Slicer to use your graphics card will vary based on your machine, operating system, and the brand of card. Generally, there will be an option in the automatically installed graphics card software (NVIDIA-NVIDIA Control Panel, AMD-AMD Catalyst TM Control Center, etc.) to select which programs you want to use the card by default. Set this up before running Slicer (you will likely have to restart your machine). Alternatively, users can manually designate the graphics card within Slicer in the volume rendering step (see Step 7.a.ii.1, below), but this action must be repeated in every session. Finally, ensure that there is adequate hard drive space on your machine for storing the CT dataset and derivative products thereof. Approximately 10 GB will be adequate for all steps involved in this workflow using Guide to the CT galaxy the example datasets. Users who wish to store and process several CT datasets should consider the size of the datasets with respect to their available hard drive space. Datasets available on MorphoSource range from $200 MB to 10 GB in size, and we recommend a storage capacity of several terabytes (TB) for users wishing to engage in extensive (i.e., high sample size) studies using CT data. Figure 1 illustrates the steps of this workflow. Briefly, the user will acquire a tomographic dataset (Step 1) and read it into the program Fiji, where it can be manipulated to reduce file size before being exported as a single file in Nearly Raw Raster Data (NRRD) format (Steps 2-4). The user then imports the file into the program 3D Slicer, which can visualize the specimen(s) or region(s) of interest. Later steps demonstrate how to measure and landmark morphologies of interest, and/or export data for downstream applications (Steps 5-8).

Workflow
Step 7.f. specifically outlines the necessary workflow for generating the 3D surface renders for use in eventual 3D printing. The final step of the workflow (Step 9) presents five analytical examples to launch the reader's exploration of practical applications. Workflow steps 1. Acquire CT data Any CT reconstruction outputs a series of grayscale images that make up the CT ''slices'' of a specimen or specimens. The brightness of the pixels making up each image in the resulting tomographic series represents the X-ray attenuation of a given area within the scan, such that areas containing material with high X-ray attenuation (e.g., heavily-mineralized bone) appear white. In addition to the x and ydimensions of the pixels, each slice contains a z-dimension (i.e., thickness) and thus each pixel actually represents a 3D volume of space, known as a voxel. The images in a tomographic series are usually in a standard format (e.g., .TIFF, .BMP, .JPEG, etc.) but they are often converted to a specialty format such as Digital Imaging and Communications in Medicine (DICOM). Whether your data come directly from CT reconstruction software or are downloaded from a CT data repository site such as MorphoSource.org (see Fig. 2A), OSF.io, or DigiMorph.org, you should move or copy the folder that contains the tomographic image series to a working location (we recommend a local file location such as the desktop rather than a remote drive).
If the image series is in any format other than DICOM, locate the resolution/dimensionality data on either the data host website ( Fig. 2A) or in the scanner log file. Note that MorphoSource removes the original scanner log file from their uploaded datasets, but the voxel dimensions can be found in the .CSV file accompanying your downloaded image stack dataset under the ''x res,'' ''y res,'' and ''z res'' columns. For the purposes of demonstrating the steps in our workflow, we will use a CT reconstruction of a pacu specimen (Pisces: Characiformes: Piaractus brachypomus; Academy of Natural Sciences of Drexel University, specimen ID: Fish: 166685), downloaded from MorphoSource.org (MorphoSource ID M15138-27533, see Fig. 2A and Supplementary Video S1). This is a modest-sized dataset ($2.5 GB) that works well on most machines. However, readers whose machines have low available RAM (i.e., <8 GB) may experience lag times in processing this dataset, and we recommend instead that they follow along with a smaller dataset, such as the reconstruction of the sculpin Porocottus allisi (Pisces: Scorpaeniformes; University of Washington, Burke Museum of Natural History and Culture, specimen ID: UW 047873) available from MorphoSource.org (MorphoSource ID M15090-27349; file size: $193 MB). If you do not already have an account and login information for MorphoSource, you will need to create one to download these files. The time taken to download will vary with Internet connections and service providers.

Import your tomographic stack of images into Fiji
While it is possible to import a tomographic image series directly into 3D Slicer, we have found that it is more reliable to first convert the image series into a single, NRRD format file. In fact, all of the optional tasks that are performed in Fiji in our workflow have analogs in 3D Slicer. Likewise, there are many tasks that we perform in 3D Slicer that could be performed in Fiji. However, we have found that the pairing of steps to the two programs outlined herein places each step in the program that performs it optimally. This minimizes instances of crashing and excessive wait times and thus maximizes the efficiency and robustness of the workflow. Familiarity with both Fiji and 3D Slicer greatly behooves the natural historian, especially once they become comfortable enough with the basic steps of CT image processing and begin to explore more advanced techniques.
6. Optimize density threshold 8. Save and/or export 7. Optional data analysis steps in 3D Slicer, see Figure 11 9. Example analyses 1. Acquire CT data from source 2. Import tomographic dataset into Fiji 3. Optional data preperation steps in Fiji, see Figure 3 4. Export tomographic dataset as NRRD format Guide to the CT galaxy (a) Open Fiji, go to ''File,'' then ''Import,'' and select ''Image sequence.'' (b) Navigate to the folder containing your tomographic image stack and select the folder (Mac) or any image within the folder (Windows), and press ''Open.'' (c) Next, Fiji will present you with a window of ''Sequence Options,'' where you can customize your import. If they are not checked already, check the box for ''Sort names numerically'' and ''Use virtual stack.'' Ensure that the ''Increment:'' is set to ''1'' and that the ''Scale:'' percent is set to ''100.'' Press ''OK'' (Fig. 2B).

Load NRRD volume into 3D Slicer
(i) Note: If desired, it is possible to reduce the file size of your stack through downsampling, but do not attempt to do so here. See Step 3.e below.
(ii) Note: The use of the virtual stack reduces the time it takes to read-in the dataset, and we have found this helpful in saving time when cropping images. However, advanced users may wish to adjust parameters of the images (e.g., brightness and contrast) in Fiji. These steps are beyond the scope of this workflow, but for such users, we do not recommend using the virtual stack option, as this can introduce system errors when attempting to modify the image parameters of large datasets. For these advanced users, or users attempting to analyze datasets with file sizes larger than the available memory (RAM) on their computer, Supplementary Script S1 will enable FIJI to crop and/or adjust image parameters of image sequences with large file sizes.

Optional steps-data preparation
There are several optional steps available within Fiji that serve to prepare the data for analysis in 3D Slicer. Use the decision tree illustrated in Fig. 3 to decide which (if any) optional steps are appropriate for your dataset and your intended analyses thereof.
(a) Specify voxel size: For use when any of your downstream analyses may include length. This step is highly recommended. Note: this step is usually not necessary if your tomographic dataset is in DICOM format.
(i) Locate the x, y, and z-dimension length of your tomographic dataset. If your data come from the output of a CT reconstruction, the voxel/pixel size is indicated in the log file of the reconstruction (e.g., ''Image Pixel Size (um)¼39.989574''; in this case, it is implicit that this is the length of each dimension). If your data come from an online repository such as MorphoSource.org, this information may be indicated in the specimen data ( Fig. 2A). (ii) If necessary, convert the units so that a single number is present on the left side of the decimal place. For example, if the pixel size is reported as 39.989574 mm, convert it an arbitrary unit that represents Â10 À7 m. For our example analysis, we will refer to this unit as a ''pym,'' and the pixel size of the above example would be 3.9989574 pym. The voxel dimensions for the pacu specimen are given as 0.04 mm ( Fig. 2A), so would be represented as 4 pym. The pixel size of the sculpin reconstruction is ''0.0299 mm'' and we would represent this as 2.99 pym. This step is critical for avoiding arbitrary scaling and rounding issues in 3D Slicer, especially for users working without a dedicated graphics card in their machine. (iii) In Fiji, click on the window that contains the image stack data that you opened in Step 2. (1) Go to: ''Image,'' then select ''Properties.'' (2) In the window that opens, change the ''Unit of length'' to whichever is most appropriate for your data (e.g., pym), and change the pixel/voxel dimensions to the appropriate dimension of your data. Press ''OK'' (Fig. 4).
(b) Digitally isolate your specimen/area of interest in the z-dimension: For use when working with a large volume of data and/or when you are interested in only a portion of your CT dataset (e.g., you are interested in only skull but have scan data for the entire skeleton). This step helps reduce file sizes and increases processing speed.
(i) Locate the upper and lower bounds of your area of interest in the z-dimension by scrolling through the image stack using the scrub bar at the bottom of your image stack window (Fig. 5A). (ii) Record the image number for each bound ( Fig. 5A and B). (iii) Create a substack of just the images that contain your region of interest.

Export tomographic dataset as NRRD format
Need to measure length?
Only interested in one area of long axis of specimen (e.g., head)?
Only interested in one area of wide axis of specimen (e.g., single bone)?
Want to visualize the 3D structure of your specimen?
File size > 2GB? Step 3 may be useful for your dataset and intended analyses.

YES
Guide to the CT galaxy (v) The substack that you specified will open in a new stacks window titled ''Substack'' followed by the range that you specified in parentheses. (vi) Use this window for all additional steps.
(1) Note: It may help to close the original stack window to avoid confusion, though leaving it open is mostly harmless. (2) Note: If attempting to analyze a dataset whose file size is larger than your available RAM, see Workflow Step 2.c.ii. (c) Digitally isolate your specimen/area of interest in the x, y-dimension: For use when working with a large volume of data and you are interested in only a portion of your CT dataset (e.g., you are interested in only a single side of a bilaterally symmetric structure such as the cranium). This step may prove ineffective for highly 3D (e.g., coiled, spiraled) specimens, and user discretion is warranted in such instances.
(i) Select the ''Rectangle'' tool from the ''(Fiji Is Just) ImageJ'' toolbar ( Fig. 6A). (ii) Use the rectangle tool to select an area of your scan that encompasses all of your specimen/area of interest.
(1) Use the rectangle tool on any image in your image sequence that contains your specimen (Fig. 6B).
(2) Use the scroll bar at the bottom of the window to visually check all images that contain your specimen to ensure that your highlighted region is not too large or too small (Fig. 6C). (a) Adjust borders of your rectangle as necessary.
(iii) Crop the image stack to eliminate all the area outside of your rectangle. (1) Go to ''Image,'' and select ''Crop.'' (2) Note: If attempting to analyze a dataset whose file size is larger than your available RAM, see Workflow Step 2.c.ii. (d) Examine the 3D volume of your cropped image stack: Use this step to visualize the 3D structure(s) contained within your image stack. This is useful for verifying that any previous digital dissection did not unintentionally remove any anatomical structures of interest. This step uses the ''3D Viewer'' plugin (Schmid et al. from the default value of ''2'' to a higher number (e.g., !8) to decrease the amount of time it will take your computer to load the 3D volume rendering (Fig. 7A).
(1) Note: This step will decrease the resolution of the rendering but will not affect the underlying slice data. (iii) Optional: Change the threshold value for the opacity of the volume rendering to highlight the denser materials (e.g., bone) in your scan.
(2) Go to ''Edit,'' and select ''Adjust threshold.'' (3) Slide the ''Threshold:'' scroll bar until the rendering highlights the material with the density of your choice ( Fig. 7B and C). (4) Press ''OK.'' (iv) Note: The volume rendering is a rotatable 3D area. Users with Microsoft Windows operating systems have reported issues with the rotation axis of the 3D volume in the ImageJ 3D Viewer. Until these issues are resolved by developers, Windows users can get around this issue by grabbing with the mouse within the ImageJ 3D Viewer window but outside of the 3D volume bounding box to rotate the area (i.e., click and drag in the black space surrounding the bounding box to properly rotate the area within the bounding box). Guide to the CT galaxy (e) Reduce the file size of your image stack by down-sampling: This step maintains the dimensionality of your specimen but reduces the resolution and thereby file size of the data. This can affect the visualization of minute structures on your specimen but may be necessary for downstream processing in programs that struggle with large file sizes (e.g., file sizes >2GB will crash 3D Slicer on most computers with 8GB of RAM). Advanced users working with large file sizes (or limited RAM) are encouraged to explore the program SPIERS (see Table 1), which can produce 3D models without loading the data into memory.
(i) The file size of your current image stack is indicated at the top of the stack window (Fig. 8A). (ii) To reduce the size, go to ''Image,'' then ''Stacks,'' then ''Tools,'' and select ''Reduce '' (iii) The default reduction factor is ''2'' (Fig. 8A), this reduces the number of slices and the size of your dataset by half and any single voxel in the new dataset will comprise the average value of a 2 Â 2 Â 2 cube of voxels in the original dataset. (1) Tip: Simply divide the current file size of your dataset by the target file size to calculate the reduction factor. For example, if your current file size is 4.5 GB, and your target size is 1.5 GB, use a reduction factor of ''3.'' (iv) When the reduction process is complete, the new number of slices and the new file size will replace the old values at the top of the stack window (Fig. 8B). (v) Note: if you did NOT set the dimensionality data for voxel size (i.e., Step 3.a above), your voxel dimensions will be given in ''pixels'' by default and you will need to manually change the voxel depth for your image stack after the reduction process is complete.
A B C Fig. 7 Visualization of CT data with a 3D volume rendering of the image stack from Workflow Step 3.c. The resampling factor (A) does not modify the underlying image data but decreases the resolution of the visualization in order to reduce loading time.
We recommend a resampling factor of between 2 (small datasets and/or fast computer hardware) and 10 (large datasets and/or slow computer hardware). Adjust the threshold from its initial value (B) until the anatomy of interest is clearly visible (C). Illustrates Workflow Step 3.d. Fig. 8 Reducing the file size of the dataset using a reduction factor applied to the image stack from Workflow Step 1. The initial file size and the reduction factor are each highlighted with a red box in (A). The resulting file size from applying the reduction factor is highlighted with a red box in (B). Illustrates Workflow

A B
Step 3.e. Guide to the CT galaxy (3) Under the ''Length'' submenu, change the ''Suffix'' from the default value of ''mm'' to the unit that you specified in Step 3.a. (e.g., for our data, we would set the suffix to ''pym''). (a) Note: We recommend users also change the ''Precision'' level under the ''Length'' submenu from its default value of ''3'' to a value of 5-10. This will decrease errors and loss of information due to rounding. (4) Note: Users with a dedicated graphics card can designate its use as a default setting here by selecting ''Volume rendering'' from the side menu, then changing the ''Default rendering method:'' to ''VTK GPU Ray Casting,'' changing the ''Default quality'' to ''Normal,'' and changing the ''GPU memory size'' to match the GPU memory on their machine.

Optimize image contrast
This step adjusts the contrast of the image and is useful for any downstream step where visually differentiating structures are useful, such as trimming and editing segmentations (e.g., Step 7.c). However, it does not alter the underlying data; it simply alters how those data are visualized. Adjust the maximum value so that the bone or other high-density material is clearly visible and distinct but that fine structures (e.g., sutures) are distinguishable and not washed-out by too high of contrast. Adjust the minimum value so that the specimen is clearly distinct from the background (see Fig. 10B).

Data visualization and analyses
There are many useful tasks and analyses available in 3D Slicer. Figure 11 illustrates a decision tree for selecting among the tasks that we have found to be most common and useful. Many of the analyses within 3D Slicer can be performed directly on the tomographic image series or on a 3D visualization of the specimen(s) therein. For users with very limited RAM and/or processing (i.e., Central Processing Unit [CPU]) speed, skipping Steps 7.a-7.c and taking measurements and landmark coordinate data directly from the slices is a way to avoid the computation and memory-taxing processes involved in 3D visualization.
(a) Volume rendering: This step creates a 3D visualization of the dataset and allows the operator to assign different values of opacity and color to materials of different density. It is useful for data exploration, measuring, and counting anatomical structures (see Step 9.d below). Volume renderings cannot be used for 3D printing other downstream processes that take place outside of 3D Slicer such as FEA. Volume rendering can be computationally taxing, especially on older machines, and some users may experience frustrating lagtimes when attempting to visualize even modest-sized datasets. If your machine has a dedicated graphics card, using it will drastically reduce lag and other difficulties associated with volume rendering (see Step 7.a.ii.1 below). Alternatively, many of the same operations performed on volume renderings (e.g., measuring anatomical structures) can be performed on surface renderings (Step 7.b below), which do not tax the CPU nearly as much (but typically require more RAM than volume renderings in Slicer). (i) Click on the ''modules'' dropdown menu and click on ''Volume Rendering.'' (ii) In the ''Volume Rendering Module,'' tweak the inputs until you can see the anatomical structures of interest (see Fig. 12): (1) If you have a dedicated graphics card, change the rendering settings so that Slicer uses the graphics card rather than your CPU to render your data. This will increase the performance of your machine drastically for all steps related to volume rendering. (a) In the ''Display'' sub-menu, click the dropdown menu titled ''Rendering,'' Guide to the CT galaxy and change ''VTK CPU Ray Casting'' to ''VTK GPU Ray Casting'' (Fig. 12A). (b) Expand the ''Advanced '' sub-menu by clicking on it (Fig. 12A). (c) Click on the ''Techniques'' tab (Fig. 12B).
(i) Click on the ''GPU memory size:'' drop-down menu and select a unit of memory that is close to but not greater than that of your dedicated graphics memory. For example, if you have an ''Intel Iris 1536 MB'' graphics card, you would select ''1.5 GB'' from the drop-down menu (Fig. 12B).  Guide to the CT galaxy off the specimen bounding box and 3D axis labels (Fig. 13A). This screen also allows you to change the background color from ''Light Blue'' (default color), to ''Black'' or ''White.'' There are many other useful features contained in the ''Pin'' window. One of which is the blue and red sunglasses button, which allows the user to project the image using anagram or other specialty-glasses-enabled schemes. When you are satisfied with your view of the specimen, export an image using the camera icon (Fig. 13B). Your image will be saved by default at the resolution of your screen. To change the resolution of the image, change the ''Scale factor:'' to higher (increased resolution) or lower (decreased resolution) than the default value of ''1.0.'' When you press ''OK,'' your image has been taken, but not yet saved. Go to the ''File'' drop-down menu and select ''Save.'' Here you can save your snapshot by checking only the box for your labeled snapshot (see Step 8 below). (b) Segment bone or other dense material(s) of interest using a density threshold: This step is useful for creating 3D models of anatomy that can be used for fine-scale digital dissection (see Step 7.c), measuring (see Step 7.d), and/or placing landmarks (see Step 7.e) on anatomical structures. The segmentations produced in this step can be used to create surface renderings that can be exported as 3D meshes and used for 3D printing and/or downstream analyses in other programs (see Step 7.f). While we use a density based ''threshold'' to create a segmentation here, there are several other options within 3D slicer for creating segmentations. We have found the threshold-based approach to be the simplest and most accessible option, especially for new users. However, we encourage readers to explore the other options once they become comfortable with the basic steps outlined here. density material that you wish to include in your segment. Very low values of the lower bounds will cause your segment to include lower-density material while high values of the lower bounds will result in only denser material being included. (1) Note: While most users working with fresh or preserved specimens have little need to adjust the upper bounds of the threshold range beyond what is described above, users working with fossil data may wish to adjust the upper threshold bounds to eliminate undesirable high-density materials in the surrounding matrix. Users working with fossil data especially are encouraged to review Sutton et al. (2014). (v) To get a closer look at the effect of changing your threshold range, change the view of your workspace so that you are only looking at one of the slice views of your data (Fig. 14A). The default view is called ''Conventional'' and includes a 3D window on top and a sagittal, coronal, and axial view below. Change the view to that of the sagittal slice (the red window below) by clicking the Slicer layout button to reveal a dropdown menu of different view options. (1) Select ''Red slice only'' (Fig. 14B).
(vi) Now we can clearly see the effects of changing our threshold range for this slice. Adjust your threshold until as much of the bone is captured (it will change color to whatever you have selected for your segment). (1) Tip: Set your threshold value initially by lowering it until speckles of segment (as indicated by the segmentation color) begin to appear in unwanted areas of your specimen (e.g., soft tissue such as the lens of the eye if present; see Fig. 14B). Next, raise the threshold value just until all of these undesirable spots disappear (Fig. 14C). Next, check for areas in your structure of interest that are thin and adjust the threshold as necessary to ensure that all areas are encapsulated by the segment. It may not be possible to set a threshold that perfectly captures your anatomical feature of interest, but the segment can be trimmed or expanded using the eraser or paintbrush tools (respectively) to make fine adjustments to the area included in the segment and match it to your structure of interest (see Step 7. c below). (vii) Press the ''Apply'' button ( Fig. 14C). (c) Isolate regions of interest from segmentation: This step is useful for isolating small and/or complex structures such as individual bones from a skeleton, or a single specimen from scan data that contain multiple specimens. (i) Visualize the 3D structure of your segment.
(3) To reposition your specimen/segment in the 3D window, select the ''None'' tool in the ''Effects'' section. Using your left mouse button will rotate specimen, holding down the ''Shift'' key while dragging with the left mouse button with reposition your specimen, the right mouse button will zoom in and out. (ii) Select the ''Scissors'' button in the ''Effects'' section ( Fig. 15B).
(1) This tool can be used on either the 3D view or any of the slice views and has several options available, perhaps the A B Fig. 15 Isolating a region of interest from a segmentation. Using the 3D view of the segmentation (A), extraneous structures are selected and eliminated using the ''Scissors'' tool (B). Illustrates Workflow Step 7.c.
Guide to the CT galaxy most useful for us are the ''Erase inside'' and ''Erase outside'' options under the ''Operation:'' section. Keep in mind that the erasure applies to the entire image stack, so use this tool carefully, especially when used within a slice view. (a) Note: To remove unwanted areas of the segmentation from only a single slice, select the ''Erase'' button in the ''Effects'' section. This tool can be time-consuming to use but is invaluable for fine-scale cleanup of a segmentation. (iii) To erase part of your segment that is not of interest, click the ''Erase inside'' option, then encircle the undesired region with the scissor tool (Fig. 15B). (1) Note: If you have a region of interest that is fairly uniform in shape, it may be useful to start the cleanup process by first using the scissor tool with the ''Erase outside'' option selected before switching to the ''Erase inside'' option for further cleanup. (1) The value of the measurement (i.e., the length) will appear next to a line connecting the two points of your measurement.
The measured length will also appear in the ''Annotations'' box, next to the measurement (it will be given a default name) in the ''Value'' column ( Fig. 16B).
(2) Note: the measurement will be indicated in the units that you specified in Step 3.a. If you followed our example, this unit is ''pym'' (see Step 5.f if 3D Slicer reports this value in an undesirable unit). If you are interested in comparing this measure to measures taken other units (e.g., mm), you will need to convert your measures to a common unit. For example, we measured the orbit diameter of the pacu specimen as 1297.8 pym, which is equivalent to 12.978 mm. The orbit of the A B C Fig. 16 Measuring anatomical structures using the ''ruler'' tool in 3D Slicer. Use the ruler tool (A) to place two points on a structure of interest (each point indicated with a red arrow in (B)). To increase the visibility of the line drawn by the ruler, click the ruler icon in the ''Edit'' column, then expand the ''Advanced'' submenu by clicking on it, next click the ''Lines'' tab and adjust the ''Width'' value using either the slider bar or by inputting the desired number directly. To keep organized, we recommend giving each measurement a descriptive name (C). Illustrates Workflow Step 7.d.
sculpin specimen is $308.2 pym, which is equivalent to 3.082 mm. (v) Optional: If you will be making multiple measurements, you can keep track of them in the ''Annotations'' window. You can change the name of a measurement to reflect what it is measuring (e.g., ''Orbit Diameter''), hide measurements from the 3D viewer, delete measurements that were not satisfactory, etc. (Fig 16C). (e) Adding markers to anatomical landmarks of interest: This step can be used for capturing and exporting 3D coordinate values for anatomical landmarks. These values can be used to calculate distances between the landmark points and/or used in geometric morphometrics studies. Landmarks can be placed in either the 3D view (on either a volume or surface rendering) or in any of the slice views. Users interested in advanced landmark placement and analysis (including the use of sliding semi-landmark curves, etc.) are encouraged to explore the ''SlicerMorph'' extension for 3D Slicer (Rolfe et al. 2020;slicermorph.github.io). (1) A description of the landmark will appear in the ''Markups'' module window, which includes an auto-generated name for the landmark as well as its coordinates. (a) Note: Slicer was created to work with medical data. The coordinate system, ''RAS,'' is short for the human anatomical coordinate system (R: left toward Right, A: posterior toward Anterior, S: inferior toward Superior). More info here: https:// www. slicer. org/ wiki/ Coordinate_ systems.
(2) Tip: The ''Scale'' slider bar controls the size of the markers. Slide it to a size that is easy for you to see on whichever window you are using to mark anatomy. Guide to the CT galaxy (iv) Tip: Give each landmark a descriptive name by double-clicking on each auto-generated name and replacing it (Fig. 17C). (f) Export surface/3D mesh object: This step exports the segmentation that was created in Workflow Step 7.b as a 3D mesh object that can be saved and read into other software packages. Three-dimensional mesh objects are the basis for many downstream applications. They can be 3D printed (or milled) from ceramics, plastics, or even metal and used to test how shape affects performance of certain morphologies like teeth, jaws, or filtering apparatuses (Kolmann et al. 2015;Divi et al. 2018). Mesh objects can also be used for gathering 3D geometric morphometrics data (e.g., Sherratt et al. 2014Sherratt et al. , 2019Buser et al. 2018;Evans et al. 2019aEvans et al. , 2019bSelig et al. 2019).
(iv) For ''Output type,'' there are two options: (1) ''Labelmap'' exports the segmented area as a labeled area in the 3D space of your image stack. This "map" can then be used to isolate areas of your image stack (e.g., using the "Mask Scalar Volume" module). (a) This format lends itself to plugins such as BoneJ for ImageJ (Doube et al. 2010; http:// bonej. org/) to calculate shape properties such as second moment of area of the 3D shape.
(2) ''Models'' exports a 3D surface model, which is composed of points (vertices) in 3D coordinate space that are connected to one another by lines (edges). The triangles formed between the lines connecting each point are called ''faces'' and can be assigned properties such as color (the default color in 3D Slicer is ''white'').
This function is what you will use to generate a surface model for eventual 3D printing. (v) Change the ''Output type'' to whichever suits your needs or whichever you wish to export first (Fig. 18A). (vi) Click the ''Export'' button ( Fig. 18A).

Saving data
(a) Click the ''Save'' button in the upper left corner of the 3D Slicer window (Fig. 18B). (i) Click the ''Change the directory for the selected files'' button at the bottom of the window (Fig. 18B) and specify a directory for storing your files. (ii) If you added landmarks, custom volume properties, ROI crops, segments, or settings and would like to change the name(s), you can do so here. (iii) If you exported a 3D surface model, the default file type is ''.vtk.'' We recommend changing the file type to a more standard format, such as ''.obj,'' ''.stl,'' or ''.ply'' (Fig. 18B). These file types are the standard formats for 3D printing or refining models prior to 3D printing (using programs like MeshLab). (iv) Check all boxes that contain files and settings you wish to save. Fig. 18 Exporting the 3D segmentation as a surface mesh (A) and saving data files (B). We recommend changing the default surface mesh format (highlighted with a red box in (B)) from the default.VTK to a more standard format, such as .OBJ, before saving. The ''Scene'' file line is indicated with a red arrow and the file format drop down menu for both the ''Scene'' and the 3D surface mesh are each indicated with a red box (B). Illustrates Workflow Steps 7.f and 8.

20
(v) Tip: If you would like to save all of your files as a single file that you can easily share with colleagues, find the ''Scene.mrml'' file ( Fig. 18B).
(1) Click on the ''File Format'' drop down menu for the scene file and select ''Medical Record Bundle'' to save all of your files and settings under one scene file. This format facilitates easy sharing of data, but is not recommended for archiving your work. (b) Note: It is also possible to create a medical record bundle by clicking the ''Create a Medical Record Bundle containing this scene'' button, which is shaped like a wrapped present and is located in the upper left corner of the ''Save Scene and Unsaved Data'' pop up window. Clicking this button will automatically change the scene file type to ''Medical Record Bundle'' or ''Medical Reality Bundle'' depending on your Slicer version and dependencies.
(2) Note: This operation will not save any segmentations as separate 3D surface models, or export any other file separately (e.g., measurements, landmark coordinates, etc.) so if you intend to do so, perform Step 8.i.4 (above) and 8.ii (below) without creating a Medical Record Bundle. You can perform both of these tasks, but they will need to be performed separately. (c) Press the ''Save'' button ( Fig. 18B)

Conclusion
Here, we have outlined steps that will help a researcher begin their journey through the CT galaxy. This workflow was developed through our own exploration and should provide researchers a means to test many of their own hypotheses without getting lost in the space of possibilities. Beyond its use in the fields of comparative anatomy, evolution, and functional morphology that we have highlighted, this workflow could be easily adapted to fields such as paleontology, paleoanthropology, archaeology, museums and heritage, biomedical research, mineralogy, and geology, to name but a few. As with all great frontiers, there is much more to explore, seek out, and more places to boldly go in years to come.
A CT on the edge of forever The potentially unlimited lifespan of CT data makes them useful not only to the researcher(s) who made the initial scan, but also to future researchers who may ask questions that the original scanner would never have considered. Collections and researchers should consider these future applications when choosing the license under which they share data, as retaining a strict copyright on the scanned image and its derivates (as is the standard policy of some prominent museums) may severely limit the ability of future researchers to reuse scans or data resulting from those scans. While it is beyond the scope of this article to discuss the nuances of the many types of licenses available, we note that many data aggregators (including GBIF and VertNet) recommend some version of a Creative Commons license for biodiversity data, and suggest that other options stifle reuse and may even be legally unenforceable. When the data are made open access under a Creative Commons license, the number of future researchers that may examine them, and thereby the number of future studies that may come from them are virtually limitless. Until such licenses become ubiquitous, researchers seeking to reuse scans from data aggregators should check carefully the terms under which each digitization has been shared, and take pains to request permission for reuse for any scans published under a restricted license. There are many potential applications of CT data that are only beginning to be explored by natural historians. For example, the ImageJ extension BoneJ can be used to calculate biomechanical attributes (e.g., second moment of area) of anatomical structures (see Workflow Step 9.f above; Rutledge et al. 2019). The use of CT-based models to gather 3D geometric morphometrics data has been widely embraced (see Workflow Steps 9.e and 9.f; Zelditch et al. 2012;Sherratt 2014), and offers important advantages over more traditional 2D geometric morphometrics (Buser et al. 2018). Such models can also be used to construct digital models for biomechanical analysis through applications such as FEA (Jaecques et al. 2004;Oftadeh et al. 2016;Stayton 2018). Pairing CT with other bio-imaging techniques like histology or material testing has great potential in the visualization and interpretation of complex anatomies, as well as making sure digital models (i. e., FEA) are accurately mimicking structural complexity (Jayasankar et al. 2017;Lessner et al. 2019;Seidel et al. 2019;Wilken et al. 2019). Another application of CT data that has great potential for research in comparative biology is the ability to estimate bone mineral density (Cann 1988;Schreiber et al. 2014). By including samples of materials with known density in their scans, researchers can estimate the density of the bone mineral in their specimens and compare the relative density of anatomical structures across large numbers of specimens. The use of contrast-enhancing elements for staining soft tissue, particularly when paired with histology, is also on the forefront of CT-based natural history and functional morphology studies (see Workflow Step 9.c; Pauwels et al. 2013;Descamps et al., 2014;Gignac et al. 2016;Hongjamrassilp et al. 2018). Finally, there is more to explore even in the way that researchers gather CT data. A recently described method of ultra-high resolution CT-based 3D reconstruction (''X-ray histotomography,'' see Ding et al. 2019) shows great potential for expanding the field even further.
Researchers, educators, and enthusiasts can use the tools, techniques, and demonstrations provided in this workflow to acquire, process, and analyze the great wealth of CT data that is being shared over the Internet. While we concede that we are not the guardians of the one true way of navigating the CT galaxy, we do think that our workflow will save users a lot of time, and hopefully keep them from giving up and going mad. Above all, we hope that our approach will reduce panic, and help readers launch their own galaxy quests. Anything less would be illogical. for generously providing the Supplementary Script for analyzing large format data in ImageJ/Fiji and we further thank Dr. Garwood and two anonymous reviewers for their thoughtful commentary and suggestions, which served to greatly improve the manuscript. We look forward to hearing about the readers' journeys in papers still unwritten, and until then, so long and thanks for all the fishes.

Funding
National Science Foundation (NSF) grant DEB-1745267 to B.L.S., DEB-1701665 and DBI-1759637 to A.P.S., and NSF 1712015 to M.A.K. supported the development of this workflow and manuscript. Example data associated with Step 9.c (Supplementary Data S1) courtesy of Oregon State University's microCT facility, a user facility developed with support from the Major Research Instrumentation Program of NSF's Earth Sciences (EAR) directorate under award #1531316.