Organs in Color: Utilizing Free Software and Emerging Multi Jet Fusion Technology to Color and Surface Label 3D-Printed Anatomical Models

Abstract

3D printing (3DP) is a rapidly evolving innovative technology that has already been utilized for the development of educational anatomic models. Until recently, it was difficult and tedious to create multi-colored models and especially labels due to technological constraints. In this technical note, a comprehensive guide for creating labeled and color-coded anatomic models was created using free software, Blender. We have composed a step-by-step process for taking an existing 3D model and adding labeling and color that is compatible with modern high-quality 3D printing technologies (Multi Jet Fusion). We provided colored and labeled 3D renderings of the surface anatomy of the brain, ventricular system of the brain, the segments of the liver, and coronary arteries as examples of the diverse potential of this technology. Additionally, we 3D printed actual models of the surface anatomy of the brain and ventricles of the brain using HP Multi Jet Fusion to demonstrate the potential of this technology in the creation of anatomic models.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10278-022-00656-1.

Background

3D printing (3DP) is an emerging technology with a multitude of applications in medicine, especially in radiology [1, 2]. 3DP is quickly moving to the forefront of alternative learning for medical students and residents. Multiple studies have found that the use of 3DP models increased medical students’ performance on anatomy evaluations compared to 2D options [3]. Medical students were also more satisfied with their instruction when given 3D models to represent anatomical concepts [4]. Additionally, a survey of medical students found that students believe that 3D printing should be incorporated into their curriculum [5].

Accurate anatomical models are often expensive and difficult to obtain. The use of 3DP is an obvious alternative to these expensive models and can be individualized and customized to the specific needs of learners. Such anatomic models are already freely available online at sites such as Thingiverse or the National Institutes of Health in STL format and could serve as starting points to the process we have described in this paper (NIH) [6, 7]. 3DP machines allow the use of varying materials, increased detail, and, more recently, dynamic color [8]. While 3DP has been shown in multiple studies to be an effective way to enhance anatomical understanding as it relates to surgery [9], there is concern for reducing costs, increasing accessibility, and optimizing the techniques and materials to create very specific and accurate models [10].

Material extrusion, also known as fused deposition modeling, has emerged as the predominant 3DP method as a result of its relative simplicity which in turn has rendered it quite economical. The past decade has seen an explosion in the availability, affordability, and efficacy of the technology. However, this class of 3DP is unable to dynamically swap color or material which requires handholding or the use of a device which can splice together different filaments and produces purge blocks (unusable leftover material). Several companies have developed what are essentially material extrusion printers using a special blank filament onto which an ink cartridge is used to add color as needed. Regardless, these methods do not produce dynamically colored prints and often require merging different STL files into one during slicing which can render them nonfunctional for detailed models.

Another method of color 3DP utilizes powder as a base which is then joined together to form a single cohesive model [11, 12]. Before this adhesive step, a colored resin may be inserted into each layer which results in a dynamically colored model. Binder jetting is one specific example of this which uses a resin to join this powder bed together before a second layer of powder is put on top of the precious one. Notably, this process does not use heat. The ColorJetPrinters (CJP) from 3D Systems is an example of such a system [13]. An alternative method has been developed by HP called Multi Jet Fusion [14]. This uses a similar concept as CJP with additional agents to prepare the powder bed and then heat is applied to join the powder bed together.

A lesser used method for creating dynamic color 3DP is sheet lamination or laminated object manufacturing. This process takes what is essentially printer paper, sends it through a color paper printer to color it, and then proceeds to cut it to shape and laminate it layer by layer to generate a 3D model. The CG-1 from Cleangreen 3D is an example of such a machine [15].

While color 3DP has become technically possible, it is currently limited by two main obstacles. For one, it is expensive and requires special tools, machines, and skilled operators. As this technology matures, we suspect it will become more affordable in line with what has been seen with material extrusion printing. Fundamentally, this also requires that a demand for these devices exist. Additionally, the ability to generate models that are ready for full color 3DP is limited to graphic designers and tech savvy individuals often using expensive and complex software. As such, we have endeavored to produce a technical guide that will allow users already familiar with 3DP to take a preexisting STL file and color it with labels using free software. Our hope is that this will increase interest in this exciting aspect of an emerging field.

As such, this methodology paper aims to provide detailed instruction for how to create colored, labeled models using Blender. We thus created color-labeled models of the ventricular system of the brain, surface anatomy of the brain, the segments of the liver, and of the coronary arteries. We have documented this process so others may be able to recreate it. Additionally, we have physically 3D printed the ventricular system and surface anatomy of the brain using HP Multi Jet Fusion. While 3DP is increasingly used to depict anatomical structures, the use of color and labeling on the physical structure has been less explored. Traditionally, 3DP models are often in one color due to technological limitations; however, for optimal application for anatomical instruction, a color-labeled structure may be more beneficial to medical student and resident learners. It is our belief that 3D color models will be a superior learning tool compared to 2D representations and non-colored/non-labeled 3D models. Our goal is to create a step-by-step method for creating a high-fidelity color model of the ventricular system and surface anatomy of the brain. This will create a rubric and guide for future 3D color printing of complex anatomy which in turn will make complex anatomy easier to comprehend for radiologists, medical learners, and physicians in general.

Methodology

As a general overview, our method involves taking an STL file, importing it into Blender and then creating a UV map of that object (Fig. 1). A UV map is essentially a 2D image that projects onto a 3D surface allowing data such as color to be incorporated onto it. The UV in UV map denotes the axes of the 2D image generated, since XYZ are already utilized in the 3D space. Blender as referenced in this paper is a freely available graphic design software capable of creating 3D models, UV maps, and texture files which provide information on the surface color (and not texture as colloquially defined) of models. Once done, we set up the model for texture painting which allows us to color the object as desired. A stencil feature is then used to insert whatever label is desired in whatever orientation. In our case, a senior medical student prepared and labeled the models which were then reviewed with a board-certified radiologist and necessary adjustments made. The ventricle and coronary arteries were graphically designed by an artist and pre-made available from public resources [6, 7], while the liver segments were custom designed by an artist in collaboration with a radiologist, and the brain was 3D reconstructed from 1-mm slice thickness isovolumetric 3D T1 sequence of a brain MRI.

Fig. 1.

Overview of our project pipeline, starting from a pre-existing model to sending colored and labeled model off to be printed. In brief, Blender is a 3D modeling program that is available free of charge. A UV map is a 2D representation of the surface of a 3D object, with “UV” referring to the two dimensions of the file created since XY are already utilized. A principled BSDF is best explained as the thalamus of this process, linking together the UV map, the model, and the color and labeling; a more technical definition would be essentially a central node that combines multiple layers into one singular object

Blender version 2.93.1 (Stichting Blender Foundation, Amsterdam, The Netherlands) is a free software that was utilized on a windows PC (Windows 10 with AMD Ryzen 5 3600 CPU with 6 cores clocked at 3.6 GHz, AMD Radeon Rx 5700 XT GPU and 48 GB RAM) with the following add-ons: STL format, UV layout, Wavefront OBJ format, Web3D X3D/VRML2 format, gITF 2.0 format, 3D-Print tool box (Supplementary 1) [16]. Per documentation from Blender, the minimum hardware requirements are a 64-bit quad core CPU, 8 GB RAM with a GPU with at least 2 GB RAM. Recommended hardware is a 64-bit eight core CPU with 32 GB RAM and a GPU with 8 GB RAM. The model was imported into a general blender file (“File”—> “Import”—> “STL”). Then navigate to the “UV Editing” tab, ensure you are in “Object Mode,” then “Select”—> ”All” (Fig. 2a). Once the entire model is selected, change to “Edit Mode” (Fig. 2b) and click on “UV” and select “Smart UV Project” (Fig. 2c). Default settings may be used for the smart UV project (Fig. 2d). Once processed, a UV map of the object should appear in the second window on the screen (Fig. 3).

Fig. 2.

UV editing tab. Basic program orientation is shown such as the UV map, tabs, and how to select different modes for model manipulation. The model is selected (a), the mode is changed to the edit mode (b), and Smart UV Project is selected (c and d). Red boxes represent menus that will open and are superimposed on the background image

Fig. 3.

Generated UV map

Before beginning the texture painting process, it is important to assess the UV map itself and ensure no portions are overlapping or closer than they need to be to ensure following steps go smoothly. The “L” key may be used to select an island (continuous portion of vertices on UV map) and the “G” key or the transform tool may be used to reposition portions of the UV Map (Supplementary 2).

Next navigate to the “Shading” tab and select “New” (Fig. 4a). This will insert a new “Principled BSDF” which is a central node onto which various layers can be assigned to a single model. The only layer necessary for this use case is an “Image texture” which can be found in “Add”—> “Image”—> “Image Texture” (Fig. 4b). Now we must create an image that will be superimposed on the UV map and allow us to texture paint the model (Fig. 4c). Name the image, set its dimensions (we utilized 8192 × 8192), and assign a background color. A higher dimension will result in a model with higher resolution which may be necessary for labeling and precise coloring. It is suggested to select a background color that will either make it easy to see which portions of the model have been colored or one that will minimize the need to color the model. The “Color” icon on the inserted “image texture” box may now be dragged to the “Base Color” on the “principled BSDF” (Fig. 4d). The model can now be painted in the “Texture Painting” tab (Fig. 5).

Fig. 4.

Shading tab. A new principled BSDF is created (a). Then an image texture node is generated (b) and an image to overlay the UV map is created (c). Click and drag the color node to the base color node (d). Red boxes represent menus that will open and are superimposed on the background image

Fig. 5.

Texture Painting tab. Model may now be readily painted. Shown is the coloring of the third ventricle which is initially blotted (a) to localize it on the UV map (b). The UV map can then be roughly colored (c) and results in a model with clean margins (d). Falloff is shown; for this process we used the steepest one which is rightmost in the yellow circle (e). Red boxes represent menus that will open and are superimposed on the background image

When texture painting, some useful features may be utilized to aid the process in Blender. One of these is the paintbrush “falloff” which can alter the intensity of the color at the edges of the brush (Fig. 5e). This can be gradual or abrupt allowing for greater control depending on the situation. For greater precision in coloring models with a clear color transition, a large brush size may be selected and to blot the model with the desired color which allows you to localize it to the UV map and directly color it on the UV map (Fig. 5a–c). This allows you to color in the lines so to speak and without needing to rotate the model and cover all angles (Fig. 5d).

In terms of inserting labels onto the model, a few different methods exist. Creating raised or sunken letters, called embossing and engraving respectfully, is already commonly used and easily produced using conventional extrusion printers. More relevant to our guide is utilizing the UV map to label our models. If a drawing tablet or other such input device is available, it is simple to adjust the brush size and falloff as needed to simply handwrite in the label. A much more standardized approach is stenciling. To do this, an image must be prepared in a separate program, which involves creating a transparent background and using a text feature to insert text. If a transparent background is not used, it will be stenciled onto the model with the text. For this project we used a free software known as GIMP, which is a graphics editor with image manipulation functions that allow the creation of the necessary file (Supplementary 3) [17]. Alternatives to GIMP would include more advanced software such as Adobe Photoshop and Illustrator. When generating this file, the background should be made transparent or else it will also be stenciled in when applied to the final model. Then text may be added to the file which can then be exported as a.png and imported into blender. A.png file is a graphics file format that allows for incorporation of transparent pixels/sections, useful for our use in surface labeling. Ensure the text is the desired color before exporting. Once in Blender, it can be resized and rotated until it is correctly positioned on the desired face using the “CTRL + Shift” keys with the left mouse button (Fig. 6). The brush can be used to apply the stencil. Since background color is kept, this process can be replicated with more complex images such as QR codes.

Fig. 6.

Stenciling. Stencil can be imported into blender (a) and should have the settings shown (b). Once imported it should overlay the model and be positioned using the CTRL and Shift keys with the left mouse button. Using the brush will insert the stencil on the face

Once these previous steps have been completed, it is up to the user to review the completed model for labeling and anatomical accuracy. Currently and to our knowledge, there is no system in place within blender or any other program to correct the authors’ labeling, coloring, or the anatomical correctness of the original file imported. This process, as detailed, does not alter the overall or surface geometry of the original preexisting model. Therefore, if anatomic or clinical accuracy is desired, it needs to be done before the model is colored and labeled. In our experience, a label down to 2 mm in height is legible as seen in our model of the brain with submillimeter letter spacing of the printed labels (Fig. 12).

Fig. 12.

The surface anatomy of the brain in color. Model generated from MRI demonstrating the ability of the software to process complex models that are biologically accurate with a high level of fidelity. Close examination of this model will reveal punctate regions of color from different regions which correlates to overlap on the UV map. Smallest labels are less than 2 mm in height. Model is 3.8 cm × 6.4 cm × 10 cm

The final step of this process involves exporting the completed model from Blender into a program that can convert the saved model from.blend (the specific file type used by Blender) to a more general file type usable by most other applications. We utilized freely available Microsoft 3D Builder for this purpose and converted the file into a 0.3mf file type before sending our models to the 3D printer.

Results

Once finished, the model may be exported in the desired format. It can be opened in 3D Builder to verify that it is satisfactory and ready to be printed (Figs. 7, 8, 9, and 10). The initial STL file, finished Blender file, and final exported file have all been provided as supplemental materials. Our models were printed on a HP Fusion 580 color 3D printer and dipped in paraffin as a protective coating but otherwise unaltered (Figs. 11 and 12).

Fig. 7.

Final model with color-coded ventricles, labels, stenciled QR code, and handwritten title. QR code links to the Wikipedia page for the ventricular system. Shown in 3D Builder

Fig. 8.

Brain with labeled surface anatomy; ready to print

Fig. 9.

Liver segments with labeled surface anatomy; ready to print

Fig. 10.

Coronary arteries with labels; ready to print

Fig. 11.

The ventricular system in color. Stylized model generated in house demonstrates ability of printer to create fine details on thin structures and the ability to insert a functional QR code and handwritten labels. Thinnest region of cerebral aqueduct is 3.3 mm. Model is 10 cm × 12 cm × 10 cm

We have generated colored, labeled anatomic models of the surface anatomy of the brain, ventricles of the brain, the lobes of the liver, and coronary arteries from what was originally a standard STL file used routinely for 3DP and have documented and presented this process from beginning to end in Blender, a freely available software. We detailed the process by which we generated a UV map of the model and then generated an image texture to allow us to color these models. We then demonstrated the process of coloring using the model itself and the UV map and have provided the subsequent fully color models. Previously generated models of the liver and coronary arteries [18] were surface labeled and made ready to be color 3D printed using the described methods.

To generate the clearest and most helpful anatomic models, it is also important to ensure that the structure is labeled and not just color coded. We have shown here methods not commonly utilized in 3DP, essentially writing onto the surface itself and utilizing a stencil to precisely put the desired 2D object onto the 3D surface. We have shown this same method may be applied with QR codes, allowing it to have significantly more utility. This, in conjunction with Blender’s wide applicability with the generation of 3D models, allows the end user to adjust the model as desired. This makes it uniquely suited for more complex and less-distinct anatomy such as the surface anatomy of a brain, the model of which was created from an MRI scan (Fig. 8).

Discussion

This methodology paper aims to provide detailed instruction for how to create colored, labeled 3D-printed anatomic models using Blender. The demonstration of a method by which to label/color previously generated STL files and render them ready for labeled/color 3DP is an exciting development. The explosion in 3DP over the past decade has resulted in an abundance of 3D models that have been created through the years. Although still nascent, we expect color 3DP to mature over time and become much more available and affordable. In anticipation of this, we have illustrated this process of manipulation of pre-existing models using freely available software which can be utilized by physicians, anatomists, and medical students for their unique use case scenarios.

Cost is one of the major obstacles in the widespread adaptation of 3DP, especially color 3DP considering it is still an emerging technology. While the cost of machines capable of printing models like those shown in this paper is substantial, it is generally not modifiable by the consumer and can usually be outsourced. For instance, numerous rapid prototyping companies exist which can be contracted to actually 3D print the model. At the same time, the cost for the development of the model is variable and dependent primarily on the labor and software costs. The labor cost is depending on who is actually developing the model. For instance, if an attending physician is taking time while at work to develop a model, the opportunity cost will be quite substantial compared to that of a medical student. Furthermore, the time needed to create a model is variable based on the complexity of the model and experience of the user. Our model of the ventricle for instance could be colored in an hour by a user familiar with Blender, perhaps an afternoon if one is Blender naive but computer savvy, whereas coloring the model of the surface anatomy of the brain is a much more involved process that would take days to complete. The last major modifiable cost for model development is the selection of the software that will be used. Professional software such Autodesk 3ds Max requires $1785 for an annual subscription, whereas less versatile programs such as Blender which we utilized are freely available. It should be noted that there is an educator/student subscription for Autodesk 3d Studio Max which is free of charge for first year of use.

While in this guide we have demonstrated this process using Blender, it is important to note that this is not the only application that this process can be done on. We have previously also used Meshmixer and 3D Paint for this process with acceptable results [19, 20]. Meshmixer is an older program that is no longer supported by its developer, Autodesk, but still available. While it can be easily colored without the need for the generation of a UV map, labeling it often results in much increased file size and significantly increased processing time due to the need to re-mesh portions of the model to allow higher resolution of the label and does result in distortion of the model surface. This is because this process involves converting the surface of the model into a mesh of triangles. Each triangle is then fully collared in by the confines of that triangle. This is not a major problem when the user desires to simply add color to the surface of the model but does become an issue when labeling. This is because the surface needs to be re-meshed which means the triangles already present on the surface have to be subdivided into smaller and smaller triangles which also results in surface distortion (albeit often minor distortion) resulting in a much bigger file and is computationally time intensive. Furthermore, in order to create enough triangles to result in a model with enough resolution for a clear label, these factors limit the resolution and practical feasibility of this software as a go-to option for the creation of a labeled color 3D-printed model.

3D Paint does not have the described limitation of Meshmixer and until recently was a default program on Windows systems so would not require users to download and install a new program. Regardless, at this time it is still freely available and easily downloadable. Compared to Blender it involves less steps to produce a finished product and is more user friendly in the sense that there are not as many options, tools, and menus to navigate through. Furthermore, the user does not need to generate a UV map in order to color the surface. The issue with 3D Paint arises with precision and how labeling is handled by the software. Likely as a result of its relative simplicity, labels subjectively do not appear as sharp compared to Blender and also actually cut through the model, appearing on the opposite face limiting its applicability. Thus, as the case with Meshmixer, this prevents us from recommending this software except for very simple minimally labeled models.

While Blender is a powerful software capable of professional level design and model manipulation, it is only superior when compared to other free software. Inevitably when a software is free, compromises must be made. In our experience, Blender tends to be more burdensome and have limited features compared to a true professional-grade program such as Autodesk 3D Studio Max. While it is unlikely that these extra features would limit users for this specific use case scenario, it is worth mentioning it increased functionality. Simultaneously, it is likely that users who are interested in this space professionally are already familiar with 3D Studio Max, so it is important to note that it is indeed a viable alternative to Blender that would be able to create a comparable end product.

One major instance in which Blender leaves room to be desired is its generation of UV maps. In general, this process works well especially with objects generated by hand such as our model of the ventricular system. In such a case, the UV map molds adequately and allows very easy texture painting. However, we found this was not the case for our model of the brain’s surface anatomy which resulted in a large amount of very tiny triangles, many of which are adjacent to one another, in the UV map resulting in artifact in the final model due to overlapping regions of color too small to manually fix (Supplement 6). The impact of this on the model may be minimal such as in our case but may still be noticeable on close examination and if larger regions of the UV map are overlapping; however, in that case it would likely be trivial to repair.

The stencil tool shown is especially helpful as it can be used for the imprinting of any 2D image onto a 3D model which we demonstrated with a QR code linking the model, for example, to the Wikipedia page for the ventricular system (Fig. 6). This also allows the user to link to whatever website desired, such as one created specifically for the model, a quiz to test medical students on the portions of the model, or even virtual reality applications.

Previously, color-labeled models have not been well developed within the 3D printing industry [21]. Adding color to 3DP is very complex and can affect the quality of the model itself [22]. While color 3DP studies have been previously created with instructions for how to create one’s own model [23], our methods differ from previous methodologies to especially include the ability to add surface labeling in a detailed step-wise fashion.

For future application, the benefits of using a 3D-printed model with color to teach anatomy must be contrasted with 3D non-colored models and also with standard 2D teaching. We initially plan to conduct a study using a previously generated non-labeled white conceptual skull base model among radiology residents and fellows [24] and comparing it with a surface-labeled color-coded equivalent 3D model. Additionally, further studies on the cost analysis associated with color 3D printing should be explored to define realistic future applications of color 3D printing at medical institutions, both for educational purposes and in the clinical realm.

Conclusion

Color 3DP with surface labeling capability has promise in various surgical, radiologic, and educational use cases. We have demonstrated that converting pre-existing models made for 3DP into models ready for dynamic color 3DP is not only possible but relatively simple with freely available tools. Color 3DP has the potential to become a useful tool for creating anatomically precise models with easily identifiable subcomponents that can be used for trainee instruction with respect to complex anatomical principles.

Supplementary Information

Below is the link to the electronic supplementary material.

Declarations

Conflict of Interest

The authors declare no competing interests.