Patient-specific, digital and physical three-dimensional (3D) anatomic models created from computed tomography (CT) or magnetic resonance (MR) images have been an important adjunct to medical fields in which disease management relies on thorough understanding of complex anatomic structures. Examples of conditions for which 3D imaging and modeling have been particularly useful include cerebrovascular disease, central nervous system tumors, craniofacial anomalies, and pelvic trauma.1–4 For these conditions, 3D modeling improves diagnostic accuracy, assists in preoperative planning, and enhances teaching of both patients and clinicians. Historically, 3D physical models of congenital heart disease have been created by paraffin wax or silicone casting of autopsy specimens; however, the techniques involved in the generation of these models are time-consuming and can only be performed on postmortem hearts.5,6 Although on-screen 3D renderings of cardiac structures are now routinely available on a standard CT or MR workstation, these renderings cannot be used interactively to simulate placement of medical devices, while the output format derived from these renderings generally does not permit conversion into physical models.
The ability to generate 3D models of patient-specific anatomic structures is of particular value during the development of new medical devices to accurately assess spatial relationships of the device within the human body. We are currently developing the PediPump, which is a small, impeller pump ventricular assist device designed specifically for children.7 As part of the development process, we have created 3D reconstructions of cardiac and other thoracic structures to optimize implantation configurations for this new device. We have also generated physical models using rapid prototyping techniques that employ a flexible polymer to create thin-walled “biomodels” that represent the external shape of the heart and great vessels of specific patients. These techniques will provide the basis for accurate device fitting during the development of the PediPump and may ultimately be useful for preimplantation surgical planning.
Materials and Methods
Computed Tomography Scans
Twelve contrast-enhanced cardiac CT studies obtained from 11 patients with and without congenital heart disease (median age 3 years; range 2 days to 13 years) were used as the source of digital datasets for 3D modeling. These CT scans were archived studies obtained in standard fashion for clinical evaluation; the underlying diagnoses are given in Table 1. Although the exact details of the scanning protocol varied depending on the clinical setting, the following general characteristics applied: a Siemens Sensation-16 spiral CT scanner (Siemens Medical Solutions, Forcheim, Germany) was used with a rotation time of 0.37 s, a collimation of 16 × 1.5 mm, a pitch of 1 with a feed of 24 mm per rotation and a tube voltage of 120 kV or 80 kV. Effective mAs varied according to patient size. The in-plane resolution of the CT datasets ranged from 0.2 to 0.5 mm, and the distance between slices ranged from 0.5 to 2 mm. All the datasets included the cardiac tissue; most included the great vessels and chest wall. This study was approved by the Cleveland Clinic Institutional Review Board.
On-screen 3D Visualization of CT Scans
The DICOM files of the CT datasets were downloaded onto a PC workstation where 3D models of the myocardium, great vessels, and chest wall were generated using Mimics software (Materialise, Leuven, Belgium). Individual two-dimensional images were “stacked” by the software, allowing thresholding operations to be performed for the selection of pixels having gray values that represented the desired anatomic structures. A region-growing algorithm was then used to separate these desired structures from the surrounding tissue. When the anatomic structures were not clearly separated by marked differences in pixel gray values, the desired pixels were manually chosen on individual two-dimensional CT images. Finally, a contour interpolation algorithm was used to calculate the 3D volumes represented by the selected pixels and the on-screen versions were rendered. On-screen models could then be interactively manipulated by rotating, translating, and zooming to determine whether or not further refinement was necessary to achieve accurate representation of the cardiovascular and skeletal structures.
Creation of Rigid Stereolithography Biomodels
Using the refined digital models, rigid and flexible physical models (biomodels) of hearts were produced using a variety of techniques. Rigid biomodels were created using a standard stereolithography machine (SLA 250/30A, 3D Systems, Valencia, CA). This stereolithographic process employed a laser beam to sequentially draw cross-sections of the desired model on very thin layers of photosensitive liquid resin to harden the resin. As each layer was generated, the platform on which the developing model rested was lowered the depth of the next slice thickness (0.1 mm) to submerge the model below the surface of the liquid resin. The process was repeated until the model was complete. The finished model was then removed from the surrounding liquid resin, cleaned with isopropyl alcohol, and postcured in an ultraviolet oven.
Creation of Flexible Rapid Prototype Biomodels
Flexible 3D biomodels were created using a ZPrinter 310 System (Z Corporation, Burlington, MA). Initially, files from on-screen models were modified using Magics RP (Materialise, Leuven, Belgium) to create a thin-walled representation of the exterior surface of the myocardium and great vessels. To manufacture the physical models more easily, each digital model was transected. The two model “halves” were then generated with the ZPrinter, which creates 3D physical models using an ink-jet printer head to apply a liquid binder to thin layers of a fine, starch-based powder to create the shape of a cross section of the desired model. As each layer of the model was created, a new layer of powder was spread across the top of the model. This process was continued until the model was completed. Upon completion, the thin-walled model halves were taken out of the starch-based material, and loose powder was removed using compressed air.
The model halves were then infiltrated or coated with different elastomeric materials, creating several versions of flexible biomodels with a range of material characteristics. Some models were painted with a two-part polyurethane (Synair Corporation, Chattanooga, TN), which was absorbed by the starch-based material. After the polyurethane cured, the model halves were bonded to one another using the same polyurethane. Other versions were created by bonding the separate starch-based parts together and using them as a mold. The mold was coated with a thin layer of two-part silicone rubber (MED 4210 NuSil, Carpenteria, CA). After the silicone cured, the coated mold was placed in a water bath to remove as much of the inner mold material as possible leaving the outer flexible heart model intact.
Twelve 3D models of the cardiac structures and the chest wall were generated from 11 patients. On-screen renderings of these models could be rotated in three dimensions to allow viewing from any angle. Figure 1 demonstrates representative anterior and lateral views of an on-screen digital model obtained from an infant with extensive malformation of the aortic arch (double aortic arch with bilateral aortic arch interruption).8Figure 2 depicts a rigid stereolithographic biomodel generated from this infant, demonstrating the accuracy and detail of the technique in modeling the complex arch anatomy that is present. Figure 3 shows the same stereolithographic biomodel after it was painted different colors to allow easy identification of different structures. Figure 4 demonstrates a flexible biomodel from a child with normal cardiac anatomy.
The ability to produce 3D reconstructions of digital imaging files has been an important addition to the clinical management of conditions that affect complex anatomic structures. For example, the diagnostic approach to trauma or other pathology of craniofacial structures, the cerebral vasculature, and the pelvis has been enhanced by the ability to generate patient-specific 3D models of this anatomy.1–4 In addition to positively impacting the diagnostic approach, 3D models influence treatment by providing an opportunity to perform detailed and precise preoperative planning. Some techniques developed to augment the usefulness of on-screen modeling, such as attempts to generate stereoscopic, on-screen reconstructions to provide a truly interactive “3D feel” of these images, have not been widely adapted.9,10 However, the ability to generate 3D plastic reconstructions of complex anatomic structures (biomodeling) has been particularly useful.1,11 D’Urso et al.1 reported that the creation of these biomodels greatly facilitated patient education, diagnosis, and operative planning when applied to complex cerebrovascular surgical cases.
The present study represents a novel application of digital 3D modeling for the physical replication of cardiac structures while accurately reproducing congenital heart lesions. Historically, autopsy specimens were used to create paraffin wax models from patients with congenital heart disease. Similar techniques for creating anatomically accurate models of hearts involved casting postmortem heart specimens in silicone rubber.5,6 These wax or silicone models have tremendous educational value, and extensive collections of these models have been used as illustrative examples of the entire range of congenital heart disease; however, these models involve time-consuming techniques and are not generally applicable unless an autopsy has been performed. In addition, the process of casting hearts in wax or silicone necessarily involves destruction of the modeled heart; if lost or damaged, these models cannot be replaced. The modeling techniques described in the present study involve manipulation of standard CT images obtained as part of routine clinical evaluation and may be applied to most digital image datasets created by CT. Thus far, our work has focused primarily on representing the exterior surfaces of the cardiac anatomy in our physical models, but techniques are being developed to create physical models with accurate internal geometry as well.
The 3D modeling techniques that we have described may be particularly useful for preoperative planning. The surgical approach to complex anatomic details in patients with congenital heart disease may be facilitated by on-screen “virtual surgery.” The benefit of being able to examine and manipulate a physical model may be enhanced when the material characteristics of the model accurately represent that of the original anatomic structure. Biomodels made of flexible polymers may be of great use in this regard. These polymer models possess tactile characteristics similar to those of human tissue, providing a substantial advantage over rigid stereolithography or wax models. In addition to allowing ex vivo visual evaluation of complex anatomic structures, these flexible physical models may be used to provide more accurate depiction of the tactile and spatial relationships to aid in preoperative planning.
In addition to facilitating education, diagnosis, and surgical planning, these modeling techniques have great value in the preclinical development and evaluation of emerging medical technology. The ability to virtually or physically manipulate accurate detailed anatomic models is particularly important during the development phase of new medical devices. Warriner et al.12 described the usefulness of on-screen 3D reconstructions for preoperative planning prior to implantation of ventricular assist devices in adults. This approach was considered to be especially useful in small patients to determine whether implantation of existing devices was feasible. Further, the authors of this study noted that 3D studies based on patient-derived CT or MR imaging might be expected to demonstrate improved accuracy over existing methods such as cadaver fitting studies used during the development of implantable cardiovascular devices. We are currently developing the PediPump, which is a miniaturized rotary blood pump designed to serve as a ventricular assist device for children. The PediPump is extremely small, measuring only 7 mm in diameter, which will allow its use even in newborn patients. We predict that the ability to model this device in relation to cardiac structures and the chest wall will be vital to determine anatomic fit as the device undergoes further development. On-screen 3D renderings and biomodeling are already useful in developing cannulation strategies for the PediPump, particularly for the smallest patients. Finally, we envision that at the time of clinical application, the modeling provided by these tools will allow patient-specific preoperative planning on a case-by-case basis, which will be especially important in providing support strategies for the wide range of anatomic variation encountered in congenital heart disease, such as abnormalities of the great arteries and veins.
We have developed a system of 3D modeling that includes the generation of both on-screen models and patient-specific biomodels with tactile properties and flexibility that mimic human tissue. This system has the potential to aid in preoperative planning and create models that may be used for educational purposes. We have found these techniques to be particularly useful in the development of the PediPump, a new medical device designed to provide circulatory support for children.
The authors thank Davorin Skender for manufacturing the 3D printed models.
1. D’Urso PS, Thompson RG, Atkinson RL, et al
: Cerebrovascular biomodelling: a technical note. Surg Neurol
52: 490–500, 1999.
2. Falchi M, Rollandi GA: CT of pelvic fractures. Eur J Radiol
50: 96–105, 2004.
3. Mortele KJ, McTavish J, Ros PR: Current techniques of computed tomography: Helical CT, multidetector CT, and 3D reconstruction. Clin Liver Dis
6: 29–52, 2002.
4. Papadopoulos MA, Christou PK, Christou PK, et al
: Three-dimensional craniofacial reconstruction imaging. Oral Surg Oral Med Oral Pathol Oral Radiol Endod
93: 382–393, 2002.
5. Cook AC, Fagg NL, Allan LD: Use of casts in the necropsy diagnosis of fetal congenital heart disease. Br Heart J
68: 481–484, 1992.
6. Kilner PJ, Ho SY, Anderson RH: Cardiovascular cavities cast in silicone rubber as an adjunct to post-mortem examination of the heart. Int J Cardiol
22: 99–107, 1989.
7. Duncan BW, Lorenz M, Kopcak MW, et al
: The PediPump: A new ventricular assist device for children. Artif Organs
29: 527–530, 2005.
8. Asnes JD, Joshi R, Kirsch J, White RD, Duncan BW: Interrupted aortic arch (type B) with bilateral ductus arterosi and bilateral aberrant subclavian arteries. Circulation
2005 (in press).
9. Hunziker PR, Smith S, Scherrer-Crosbie M, et al. Dynamic holographic imaging of the beating human heart. Circulation
99: 000–6, 1999.
10. Ribas GC, Bento RF, Rodrigues AJ Jr: Anaglyphic three-dimensional stereoscopic printing: Revival of an old method for anatomical and surgical teaching and reporting. J Neurosurg
95: 1057–1066, 2001.
11. Wurm G, Tomancok B, Pogady P, et al
: Cerebrovascular stereolithographic biomodeling for aneurysm surgery (Technical Note). J Neurosurg
100: 139–145, 2004.
Copyright © 2006 by the American Society for Artificial Internal Organs
12. Warriner RK, Haddad M, Hendry PJ, Mussivand T: Virtual anatomical three-dimensional fit trial for intra-thoracically implanted medical devices. ASAIO J
50: 354–359, 2004.