Medical 3D printing for vascular interventions and surgical oncology: a primer for the 2016 radiological society of North America (RSNA) hands-on course in 3D printing
© The Author(s) 2016
Received: 12 October 2016
Accepted: 21 October 2016
Published: 25 November 2016
Medical 3D printing holds the potential of transforming personalized medicine by enabling the fabrication of patient-specific implants, reimagining prostheses, developing surgical guides to expedite and transform surgical interventions, and enabling a growing multitude of specialized applications. In order to realize this tremendous potential in frontline medicine, an understanding of the basic principles of 3D printing by the medical professionals is required. This primer underlines the basic approaches and tools in 3D printing, starting from patient anatomy acquired through cross-sectional imaging, in this case Computed Tomography (CT). We describe the basic principles using the relatively simple task of separation of the relevant anatomy to guide aneurysm repair. This is followed by exploration of more advanced techniques in the creation of patient-specific surgical guides and prostheses for a patient with extensive pleomorphic sarcoma using Computer Aided Design (CAD) software.
Keywords3D Printing Aneurysm repair Cancer Segmentation Computer-aided design Orthopedic Surgery Implant Surgical Guide Radiological Society of North America Precision Medicine
In the short interval since the publication of our initial practical medical 3D printing guide for the 2015 annual RSNA meeting , the published literature in this domain has undergone exponential growth. The number of peer-reviewed journal publications has nearly doubled, ever expanding the breadth and scope of the applications of 3D printing in medicine. It is evident that 3D printing is poised to play an important role in transforming the practice of medicine, with applications ranging from fabrication of simple tools to complex tissues and, eventually, organs. Development of familiarity with 3D printing may therefore be of considerable interest to a wide range of medical professionals.
The term “3D printing” has evolved to become synonymous with the terms “rapid prototyping” and “additive manufacturing” within the medical domain, and refers to the process of fabrication of 3D objects through sequential deposition and fusion of matter in a layer-by-layer fashion . A wide range of printing technologies are available to enable the fabrication of 3D models in a range of materials, including plastics, metal alloys, ceramics, and numerous biological substrates supporting living cells. While a more detailed examination of these technologies is covered at length elsewhere, it must be noted that this diversity enables a tremendous range of applications at numerous levels of cost, accuracy, durability, build time, and biocompatibility. Medical 3D printing is therefore nearly universally accessible and applicable.
The process of 3D printing starts with the generation of a printable model in a process that is not entirely dissimilar to typical 3D reconstructions. Typically, cross-sectional imaging is used as an initial step in model creation. Computed Tomography (CT), by virtue of reflecting changes in a single parameter (attenuation), is the preferred modality for this purpose. The acquisition of images specifically for the purposes of 3D printing requires minimizing the size of voxels within the reconstructed images while optimizing contrast administration for adequate visualization of the relevant anatomy.
Once appropriate images are acquired, segmentation, or isolation of the relevant anatomy is undertaken using methodology that is akin to 3D visualization methods available in most radiology departments. The process of segmentation involves creating a set of criteria for the voxels to satisfy in order to be included in a model. These criteria may include connectivity to a seed point, attenuation coefficient within a specific range, or presence within specific manually defined geometric boundaries. Such selection criteria may be set manually or specified automatically as part of applying a segmentation algorithm.
While 3D visualization typically ends at selecting and displaying a set of segmented voxels, further additional model manipulations are required to enable 3D printing. Fundamentally, the collection of voxels occupying a specific set of Cartesian coordinates within a region of interest (ROI) needs to be transformed into a 3D object in a process referred to as tessellation. Tessellation is widely used in computer graphics to approximate shapes using a set of triangles. The more triangles are used, the more refined a shape becomes. Unfortunately, transformation into a 3D object does not ensure model printability or stability. For instance, it may be necessary for the model to be further smoothed, non-printable parts may be removed or mathematically adjusted, and vulnerable areas may be reinforced using a range of automated algorithms.
Further manipulations using Computer-Aided Design (CAD) software may be necessary to enable the development of patient-specific instrumentation, implants, or reconstructions. Given the ubiquity of such tasks, this course will focus not only on segmentation of patient anatomy, but also the de novo creation of personalized medical models.
The principles of 3D printing in medicine are best appreciated through practical hands-on experiences covering a broad range of applications. Therefore, this guide will provide the foundational knowledge to broadly cover the segmentation of relevant anatomy on CT-derived Digital Imaging and Communications in Medicine (DICOM) images followed by 3D printable model creation and concluded by patient-specific reconstructions and surgical guide design.
The United States Food and Drug Administration (FDA) classifies medical 3D printing software into design manipulation software that enables medical device design and modification and build preparation software that enables the conversion of the digital design into a file format that is 3D printable, or Standard Tessellation Language, or STL file in this case . To illustrate the use of the former, we shall apply Mimics inPrint and 3-matic Medical software (Materialise, Leuven, Belgium) while the latter will be represented by Polyjet Studio (Stratasys Ltd., MN, USA).
The Mimics inPrint software facilitates the processing of 2D image data acquired from axial imaging (CT, MRI) in order to create 3D printable models. While the vast majority of functionality is present in this accessible package, certain higher-order operations, including various segmentation algorithms present in more advanced packages such as Mimics (Materialise, Leuven, Belgium), are omitted for simplicity and accessibility.
3-matic Medical is a CAD package tailored for design of medically relevant models. This software enables the manipulation of patient-derived 3D models as well as creation and redesign of entirely new models in the context of patient anatomy. The application of this package in anatomical reconstruction and medical device creation will be explored in this work.
Finally, Polyjet studio enables the preparation of the 3D models represented as a set of connected triangles in an STL file format for printing on a commercial 3D printer. This software enables the exploration of factors such as part placement, material costs, build times, and material selection in order to optimize the 3D printing process. While the Polyjet studio is dedicated specifically to Polyjet printers capable of producing multi-colored prints, the general interface and approach are fairly representative of the experience in setting up a 3D printing task.
In the first case, we shall examine two common scenarios where 3D printing holds the potential of improving patient care: pre-procedure planning for vascular interventions with an example of iliac artery aneurysm repair and creation of surgical tools for neoplasm resection on a patient with a large pleomorphic sarcoma.
Iliac artery aneurysms are mostly seen in association with aortic aneurysms. Isolated iliac artery aneurysms are rare, involving <0.1 % of population . Most common etiology is atherosclerosis, while less common causes include connective tissue disorders, trauma, dissection, or infection. The most commonly involved site is the common iliac artery, followed by the internal iliac artery . Aneurysms of external iliac artery are relatively rare.
Although most patients are asymptomatic, some present with symptoms due to mass effect on adjacent ureters, nerves, or veins. Rarely, aneurysmal rupture will lead to an acute presentation. Asymptomatic aneurysms are treated when they enlarge beyond 3–3.5 cm in diameter or if they become symptomatic [6–8]. Although treatment can be performed by open surgery or by an endovascular approach, the latter is presently favored due to its safety and lower complication rates, except in patients with compressive symptoms [9, 10].
Depending on the location and extent of the aneurysm, endovascular treatment involves deployment of stent graft with or without internal iliac artery embolization or surgical bypass. Since internal iliac embolization can lead to complications including buttock claudication, bowel ischemia, sexual dysfunction, and tissue necrosis, branch iliac artery grafts are also used to preserve flow, especially while treating bilateral aneurysms.
An initial CT angiogram is performed prior to the procedure to define the anatomy and select the appropriate device. Pertinent information sought includes assessment of access (common femoral artery) for its diameter (large enough to accommodate delivery system), presence of plaques or calcifications, which can hinder access. Iliac artery diameters at landing zones (typically 15–20 mm length) are measured in true short axis using multiplanar reconstructions to select the size of the stent graft (typically oversized by 10- 15 % or as per manufacturer’s recommendation). Iliac vessel tortuosity is also assessed as it can pose a challenge during the procedure.
A 3D printed model can therefore be quite beneficial in preoperative planning and teaching of aneurysm repairs. The aneurysm model produced using this process may be used to demonstrate the mechanism of stent deployment in endovascular aneurysm repair. Using this case as a guide, learners will become familiar with segmentation techniques used routinely in 3D printing to expose relevant anatomy, as well as extrusion techniques necessary to reconstruct the anatomy of any contrast-opacified structure, including vasculature, heart, or hollow viscus.
The second case demonstrates the design and surgical planning involved in the creation of a prosthetic hip implant for a patient with an extensive lytic lesion within the left iliac crest secondary to a soft tissue sarcoma. This patient was diagnosed with undifferentiated pleomorphic sarcoma (giant cell subtype).
Pleomorphic sarcoma is the most common type of high grade soft tissue sarcoma in the adults [11, 12]. It often presents as an aggressive, large, high grade sarcoma of the extremity. Prognostic risk factors include tumor size, depth, and proximal location. This soft tissue sarcoma is thought to be derived from a primitive mesenchymal cell capable of differentiating into histiocytes, fibroblasts, myofibroblasts, and osteoclasts. The etiology of the tumor remains unknown. Prior radiation therapy is likely a risk factor. The most common clinical presentation is an enlarging painless soft-tissue mass in the thigh, typically 5–10 cm in diameter. The majority of tumors are intramuscular.
When it arises within the pelvis, detection may be limited due to its deep location, and the patient may only seek medical attention when the tumor is quite large. Although it may metastasize to lungs or local lymph nodes, this neoplasm typically expands and recurs regionally to produce a lesion which may visually appear to be well circumscribed but may actually have extensive soft tissue invasion . The treatment of choice is wide local excision with adjuvant radio- and chemotherapy, for a cumulative 5 year survival on the order of 50–60 % of all high-grade pleomorphic sarcomas [13–15].
This case will primarily focus on CAD software (3-matic) to introduce the learners to the patient-specific surgical planning and reconstruction principles that may then be adapted to a wide range of cases, ranging from craniotomy implants to surgical planning with custom guides for osteotomies involving all bones of the appendicular and axial skeleton.
Mimics inPrint keyboard and mouse shortcuts
Scroll Wheel Drag (2D View)
OR Shift right click + drag
Pan: Move the mouse while keeping the center button pressed
Right Click + Drag
Zoom: click and drag the right mouse button to zoom in and out.
Arrow Up/Scroll Up (2D View)
Go to next slice
Arrow Down/Scroll Down (2D View)
Go to previous slice
Page Up (2D View)
Skip 10 slices upward
Page Down (2D View)
Skip 10 slices downward
Hover mouse over a view, then SPACE
Change chosen image to/from full screen.
Switch between two window states
CTRL + Z (Everywhere)
Undo the previous action.
CTRL + Right Click + Drag (2D View)
Adjusts contrast window in images
Right Click + Drag (3D View)
Rotate a 3D shape.
CTRL + Right Click + Drag (3D View)
Zoom into or out of 3D shape.
Shift + Right Click + Drag (3D View)
Pan the 3D shape around a scene.
Patient 1: endovascular aneurysm repair simulation
Task A: segmentation
WHAT YOU ARE DOING: Segmenting the aorta and iliac vessels as well as the bones. In segmentation, specific voxels satisfying specific attenuation criteria are selected to create a region of interest. You will isolate the vessels and bones in this step.
WHY YOU ARE DOING IT: To identify the voxels for conversion into a 3D model.
HOW TO DO IT: We will be using a widely available tool, Thresholding, to select the desired voxels based on attenuation within a specified Hounsfield Unit (HU) range. This step will create a Region of Interest (ROI), an intermediate model that requires further manipulations to be printed. A list of the ROIs you have created appears in the first pane of the Project Management Toolbar (bottom left).
We shall begin by importing the images into our project from a DICOM folder. The DICOM data can be obtained from the OsiriX DICOM image library . To do this, select the File menu, then New From Disk, followed by selecting the DICOM Data folder, pressing the Next button in the dialog that appears, and finally Convert. This will load all the DICOM images into the workspace.
Segmentation is a task that simply creates a set of criteria for voxels to fulfil in order to be included in a model. The first set is created by the thresholding function – a range of HU values are set to select the relevant voxels. When we decided to “keep the largest region” after thresholding, we selected the largest set of contiguous voxels. With the split function, we are providing a number of seed points for the software to automatically extend in order to create two dissimilar contiguous regions, separating the anatomy we wish to see in our model from the background.
Note: To expand the axial image window to the full screen, simply press the spacebar while hovering over it with your mouse.
Thus far, we have segmented the contrast-opacified arteries within the patient’s lower abdomen and pelvis. This has created an ROI or mask, which is simply a collection of voxels that satisfy a set of criteria, which in this case constitute 1) 140-250HU density range, 2) membership within the largest contiguous collection of voxels satisfying density criteria, and 3) algorithmic separation of ROI based on user input. This is quite similar to segmentation that is typically performed in 3D rendering on a daily basis. Note that this collection of voxels is not yet a 3D printable model. In order to create a printable model, we need to convert this collection of voxels into an STL model and verify that this STL model is printable.
Task B: creation of prinable STL models
What are STL models?
STL models define a 3D geometry as a collection of triangles that collectively describe a shape. As the number of triangles used increases, the approximation of reality by the model improves, but the model becomes more computationally expensive to create and manipulate. The mere act of describing a model as a set of triangles does not ensure printability, however. Further processing is required to ensure that the shapes that are represented in the STL model are manifold shapes and can exist as non-abstract geometrical constructs. While this is outside the scope of this tutorial, non-manifold shapes include entities such as planes, lines, and points, which have no thickness, cannot exist in the real world, and will likely lead to errors in printing.
After inspecting the model, you will notice that a uniform wall was made around the intraluminal contrast ROI. Since we would like to simulate stent placement, it is necessary to open the vessels by cutting off the closed ends. To do this, select the Cut function.
Patient 2: soft tissue sarcoma excision and personalized implant design
For this patient with extensive soft tissue sarcoma invading into the osseous structures of the pelvis, we shall first need to design a patient-specific prosthetic implant. To do this, we would require to mirror and duplicate the healthy hemipelvis exactly at the point of excision. In order to allow for precise excision that spares healthy tissue, we would also need to create cutting guides to direct the wide excision of this neoplasm in a manner that allows subsequent placement of the custom implant flush with the excision site while optimizing the resection volume.
The images for this project may be obtained from the Cancer Imaging Archive  by searching for the patient TCGA-QQ-A5V2 and retrieving corresponding CT images (no direct link is available). The preparation of these CT images for segmentation is identical to that in Task A for Patient 1.
Task A: segmentation
Task B: isolation of relevant pathology
The axial view is best suited for selection of the background (rejected) and foreground (desirable) features in this case. The sacrum will be the background, while the remainder of the bones of the pelvis will be the foreground. You need not paint all of the axial images – two well-placed images will typically suffice. If your selection results in more remnant fragments than necessary, more axial images may be annotated. After you are satisfied with your result, press the green checkmark in the Split dialog.
Task C: mirroring the healthy hemipelvis
Mirroring is a basic approach used in the reconstruction of patient anatomy in cases of severe pathologic unilateral deformities. Since humans exhibit plane symmetry through the midline sagittal plane, the non-pathologic side can be mirrored and computationally overlapped with the diseased side to provide rapid patient-specific reconstruction. This technique is widespread, from craniotomy plate creation to mandibular reconstruction, for example. In this work, we shall use the healthy hemipelvis as a template for the hemipelvis with the sarcoma. For this, we will need to first separate the two halves of the pelvis, and then reflect and overlap the healthy hemipelvis with the diseased one.
After the part has been created, the software will display the Edit Part menu. Here, we shall select to Cut the part into two. Before doing this, however, let us align the view to best prepare the cut. Go to View in the menu toolbar, and then select 3D Viewports > Views > Front. Once this is set up, ensure that the Skeletal part is in the selection menu, that the Method is set to Cut, and draw a rectangle separating the two halves of the pelvis, as shown. Press the green checkmark button once this is complete.
You will notice that a Right_Mirrored part has now been created. Visualizing the two parts together as above, you will notice that their alignment is not complete, limiting implant creation. Computer Aided Design (CAD) software, 3-matic will be necessary to complete this task. To transfer our work, simply save the project from the File menu.
Task D: computer-aided implant design with 3-matic
You will notice that the 3-matic window has a Menu Bar where its numerous functions can be found, a 3D view to visualize the selected models, Object Tree where the manipulated models appear, Operations Tab which allows us to set parameters for various functions, and the Logger to display pertinent status updates (Fig. 21).
Task E: computer-aided cutting guide design
Let us now take a moment to rename all model parts for clarity. Select the smaller part of the Right_Mirrored model and rename it to Implant. To do this, simply single-click the name of this object in the Object Tree so that it changes to an editable field, type in the new name and press Enter. This is the reconstructed, patient-specific implant that will be printed in titanium and implanted after the resection. Because we have completed our manipulations with it, we shall now hide it by right-clicking it in the Object Tree and selecting Hide. Likewise, rename the healthy part of the left hemipelvis as Left. The object tree should therefore contain only two parts: Left and Implant.
The wrapping operation fills all holes and defects inside our model and creates a solitary watertight surface that would be easily amenable to 3D printing. In this operation, the model is checked to ensure that non non-manifold geometry is present in it. This is important because all objects within the real world are manifold: it is impossible to manufacture a perfect plane (single fold) of 0 thickness, or two shapes that touch in a single, mathematically discrete dot – these are abstract concepts. The wrapping operation ensures that the geometry of our model is not only watertight, devoid of internal cavities, and smooth, but also that it can exist in the real world!
This operation will result in a new wrapped model, Left_wrapped. Hide the Left model and duplicate the Left_wrapped model by right-clicking it and selecting Duplicate. We will now operate on this duplicated model (Left_wrapped_duplicate) to create a surgical guide that adheres to patient anatomy.
Landmarks and the Surgical Guide: Notes from the OR
In our case, the orthopaedic surgeon consulted for this case would like to position the patient in the right lateral decubitus position. This exposes the lateral border of the left pelvis for resection. The surgeon would then proceed to resect the diseased soft tissue and dissect the healthy musculature from the smallest possible portion of the lateral aspect of the iliac bone. In doing so, the surgeon would attempt to preserve the abdominal wall if possible, while minimizing the amount of muscle tissue detached from the lateral aspect of the left pelvic bone while skeletonizing it. The surgeon would then prefer to use the anterior superior iliac spine as the first and most obvious landmark for surgical guide placement, followed by the posterior superior iliac spine as the second point of alignment. If the surface of the surgical guide conforms to the lateral surface of the iliac bone while leaving a 1 mm margin of error for anticipated minimal imperfections while skeletonizing it, the surgical guide is anticipated to fit adequately. Once aligned to all landmarks, the inscriptions on the implant are again verified to ensure appropriate placement. The guide is then secured in place by positioning pins placed within appropriate holes placed on the implant at the design stage, and the guided osteotomy can begin.
We shall now use this fragment to carry out a Boolean subtraction, which will create a fragment that adheres to the lateral surface of the iliac bone and mimics the patient’s anatomy closely, assisting the surgeon in its placement. Therefore, we will simply move this fragment lateral to the iliac bone and then remove its overlap with the iliac bone, leaving a 1 mm margin to accommodate any imperfections of surgical resection.
These models can now be exported as STL files (File > Export > STL) and printed. A further overview of 3D model printing is discussed in Additional file 1.
As the applications of 3D printing in medicine continue to expand, familiarity with the basic operations and tasks involved in this process becomes increasingly important. By the virtue of their existing functions, some radiologists may very well subspecialize to emerge as physicians responsible for integrating the input from other specialists to design procedure guides and generate medical implants in the future. In this work, our goal has been to familiarize the medical professionals with this nascent field and start the dialogue in this revolutionary development in personalized medicine that has the potential of transforming the practice of medicine entirely.
Anterior superior iliac spine
Digital imaging and communications in medicine
United States Food and Drug Administration
Magnetic resonance imaging
Region of interest
Radiological Society of North America
Standard Tessellation Language
No funding sources to declare for this study.
AS and LC conceived the design, analyzed the data, drafted, edited and critically revised the manuscript. AG and TH contributed significant portions of the manuscript. AW provided customized illustrations. LC carried out data collection and analysis. AS, AG, AW, CT, SK, EA, DM, LC interpreted the data and drafted the manuscript. All authors have read and have given final approval of the version to be published. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Consent for publication
The DICOM patient images used this this publication are obtained from open anonymized online repositories publishing patient data with patient consent. Statements to this effect are available on the respective websites of the OsiriX Image Library  and the Cancer Imaging Archive .
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
- Giannopoulos AA, Leonid C, Adnan S, Aili W, Wilfred D, Ekin A, Chris H, Nicole W, Todd P, Dydynski PB, Rybicki DMFJ. 3D printed ventricular septal defect patch: a primer for the 2015 Radiological Society of North America (RSNA) hands-on course in 3D printing. 3D Printing in Medicine. 2015;1:3.View ArticleGoogle Scholar
- Mitsouras D, Liacouras P, Imandzadeh A, Giannopoulos A, Cai T, Kumamaru K, et al. Medical 3D printing for the radiologist. Radiographics. 2015;35(7):1965–88.View ArticlePubMedPubMed CentralGoogle Scholar
- United States Food and Drug Administration. Technical Considerations for Additive Manufactured Devices. Draft Guidance for Industry and Food and Drug Administration Staff. http://www.fda.gov/downloads/medicaldevices/deviceregulationandguidance/guidancedocuments/ucm499809.pdf. Accessed 8 Oct 2016.
- Brunkwall J, Hauksson H, Bengtsson H, et al. Solitary aneurysms of the iliac arterial system: an estimate of their frequency of occurrence. J Vasc Surg. 1989;10:381–4.View ArticlePubMedGoogle Scholar
- Ferreira J, Canedo A, Brandão D, Maia M, Braga S, Chaparro M, Barreto P, Vaz G. Isolated iliac artery aneurysms: six-year experience. Interact Cardiovasc Thorac Surg. 2010;10(2):245–8.View ArticlePubMedGoogle Scholar
- Huang Y, Gloviczki P, Duncan AA, Kalra M, Hoskin TL, Oderich GS, McKusick MA, Bower TC. Common iliac artery aneurysm: expansion rate and results of open surgical and endovascular repair. J Vasc Surg. 2008;47(6):1203–10.View ArticlePubMedGoogle Scholar
- McCready RA, Pairolero PC, Gilmore JC, Kazimer FJ, Cherry Jr KJ, Hollier LH. Isolated iliac artery aneurysms. Surgery. 1983;93:688–93.PubMedGoogle Scholar
- Santilli SM, Wernsing SE, Lee ES. Expansion rates and outcomes for iliac artery aneurysms. J Vasc Surg. 2000;31:114–21.View ArticlePubMedGoogle Scholar
- Patel NV, Long GW, Cheema ZF, Rimar K, Brown OW, Shanley CJ. Open vs. endovascular repair of isolated iliac artery aneurysms: A 12-year experience. J Vasc Surg. 2009;49(5):1147–53.View ArticlePubMedGoogle Scholar
- Albertini JN, Favre JP, Bouziane Z, Haase C, Nourrissat G, Barral X. Aneurysmal extension to the iliac bifurcation increases the risk of complications and secondary procedures after endovascular repair of abdominal aortic aneurysms. Ann Vasc Surg. 2010;24(5):663-9.View ArticlePubMedGoogle Scholar
- Nascimento AF, Raut CP. Diagnosis and management of pleomorphic sarcomas (so-called “MFH”) in adults. J Surg Oncol. 2008;97(4):330–9.View ArticlePubMedGoogle Scholar
- Matushansky I, Charytonowicz E, Mills J, Siddiqi S, Hricik T, Cordon-Cardo C. MFH classification: differentiating undifferentiated pleomorphic sarcoma in the 21st Century. Expert Rev Anticancer Ther. 2009;9(8):1135–44.View ArticlePubMedPubMed CentralGoogle Scholar
- Le Doussal V, Coindre JM, Leroux A, Hacene K, Terrier P, Bui NB, Bonichon F, Collin F, Mandard AM, Contesso G. Prognostic factors for patients with localized primary malignant fibrous histiocytoma: a multicenter study of 216 patients with multivariate analysis. Cancer. 1996;77(9):1823–30.View ArticlePubMedGoogle Scholar
- Salo JC, Lewis JJ, Woodruff JM, Leung DH, Brennan MF. Malignant fibrous histiocytoma of the extremity. Cancer. 1999;85(8):1765–72.View ArticlePubMedGoogle Scholar
- Zagars GK, Mullen JR, Pollack A. Malignant fibrous histiocytoma: outcome and prognostic factors following conservation surgery and radiotherapy. Int J Radiat Oncol Biol Phys. 1996;34(5):983–94.View ArticlePubMedGoogle Scholar
- Clark K, Vendt B, Smith K, Freymann J, Kirby J, Koppel P, Moore S, Phillips S, Maffitt D, Pringle M, Tarbox L, Prior F. The Cancer Imaging Archive (TCIA): maintaining and operating a public information repository. J Digit Imaging. 2013;26(6):1045–57.View ArticlePubMedPubMed CentralGoogle Scholar
- OsiriX DICOM Image Library. http://www.osirix-viewer.com/datasets/DATA/MACOESSIX.zip. Accessed 8 Oct 2016.