An MD&DI March 1997 Feature
Magnetic resonance imaging is not just for diagnostics. Converted into CAD models, MRI data can also create virtual anatomies upon which to build medical devices that are sure to fit patients.
For more than a decade, physicians have used magnetic resonance imaging (MRI) to produce accurate representations of internal and external anatomical structures. With these images, abnormalities such as tumors can be easily detected.
MRI scans converted into vector CAD files can be used to create 3-D anatomical models for product development.
Today, MRI data are being used not only by physicians, but also by medical device manufacturers. Through the use of 3-D computer-aided design (CAD) models derived from MRI data, devices can be designed to accurately conform to the shape of the human body. In particular, medical devices that are inserted or implanted into the body, such as hearing aids, surgical tools, and diagnostic imaging probes, can benefit from the use of anatomical models in product development. By seeing how the device will fit the body early in the design process, product engineers can dramatically improve the ergonomics of these types of products and prevent redesigns.
Anatomical models produced from MRI data can also be used to design prostheses. An abnormal structure, such as an amputated limb, for example, can be scanned by an MRI machine and modeled in a CAD program. By comparing the amputation model to a model of a whole limb, a designer can determine the shape of the needed prosthesis.
To use MRI data to build medical devices, designers need to become familiar not only with what the MRI data actually represent, but also with how to convert them to CAD format and then how to manipulate the CAD files to create 3-D anatomical models. Becoming familiar with these aspects of the process will require an understanding of some technical graphics concepts, such as voxels and vectors.
To produce an MRI representation, magnetic fields and radio waves stimulate atoms in the body, which emit radio signals in response. The durations of these response signals vary among tissue or other organic material types. Cancerous tissue, for example, emits a much longer response than healthy tissue does. A microprocessor uses these variations in response lengths to produce an image of a cross section of an anatomical structure. The cross sections comprise rows and columns of 3-D volume elements known as voxels.1 The depth of the voxels is, of course, set by the thickness of the slice of anatomy that is being modeled. The width and height are determined by the area that is within the scope or boundaries of the MRI scan (the field of view), the overall size of the image, and the number of columns and rows of voxels that are used to generate the image (the imaging matrix). Each voxel is either white, black, or a shade of gray. The shade, or intensity, of the voxel is determined by the duration of the radio-frequency response that is emitted from the corresponding area of the patient's anatomy.
One of the first technical challenges of creating CAD models from MRI data is converting the information generated by the MRI scanner--the location, size, and intensity of every voxel--into a vector format that a CAD program can use.
VOXELS TO VECTORS
To create the MRI representation, the coordinates of the voxels, their size, and their intensity are displayed, or painted, as a series of unrelated dots that form a cohesive picture only because they are very close together. Graphics used by CAD programs, however, are composed of continuous lines, or vectors, that represent equations based on beginning points, ending points, and direction.
There are several techniques for converting voxel data sets into vector graphics. Of course, before a conversion technique can be chosen, the MRI images must be clear enough that the structure of interest can be identified. If the images are sufficiently sharp, then practical factors--such as the availability of computer hardware and software resources; the complexity, clarity, and number of MRI images; and the frequency at which anatomical models need to be built--will determine which conversion technique is most appropriate.
Voxel to Triangulated Mesh. The most direct, automated, and expensive method is to use a set of conversion algorithms that generate vector triangulated mesh surfaces directly from the MRI data set.2,3 Typically, such algorithms evaluate neighboring voxels to determine whether the differences in their intensities are within a specified threshold value. Adjacent voxels that have nearly the same intensity can then be linked to form a vector. Using this method, the computer generates vertices and triangular surfaces from the voxel data sets.
A 3-D model of a medical device is drawn onto a wireframe anatomical model.
Complex computer code can be written to perform this type of conversion. There is also diagnostic imaging software available, although it is costly and requires a significant investment in sophisticated hardware.
Because it is automated, the voxel-to-triangulated-mesh technique is best for manufacturers who plan to convert a large amount of MRI data. One limitation of the method, however, is that CAD files generated by it are made up of collections of flat triangulated planes with no actual curves. The triangles, especially if they are very small, can closely approximate curves, but using very tiny triangles will cause large file sizes and slow processing times, which can have a negative impact on CAD system performance.
Voxel to Vector Polygon. A less expensive and less automated technique uses the raster-to-vector conversion functions that are available in vector-based graphics packages, scanner conversion packages, or mapping and geographical information system software.
Specialized applications, or utilities, are available to save each voxel-based image in a graphic format that is compatible with the vector conversion software to be used. Common formats for saving these raster graphics are bitmap (.BMP) and graphics interchange format (.GIF) for the PC, or picture (.PICT) and tagged image file format (.TIFF) for the Macintosh.
Each graphic file is then imported into the conversion software, which draws a polyline along the edge of the anatomical structure of interest. To help the software select the anatomy of interest, the user can convert the graphic to a negative or monochrome image before conversion to vector format. The vector outlines are exported in a CAD-compatible format, typically the design exchange format (.DXF).4 A batch utility program can then be used to put the .DXF files into the CAD drawing application that will be used.
This conversion technique is not fully automated, because some user intervention and setup is required to enable the software to correctly outline the structure of interest for each image. Therefore, this method is best used when only a moderate number of images are to be converted.
Manual Vectorization. The least expensive way to convert MRI data to vectors is to import a raster graphic file directly into the CAD system and then simply draw a polyline spline, or series of lines and arcs directly over the anatomical structure of interest.
To prevent the creation of excessively large files when using this method, it is best to insert one image at a time into the drawing, trace the image, and then save the drawing to a separate file. Then the designer can delete the raster image and repeat the process for the other images in the data set. This technique should be used when only a few images are being converted.
CAD MODEL BUILDING
After the MRI graphics have been converted to a CAD-compatible format, the next step for a device designer is to create a 3-D surface model of the anatomy. If the voxel-to-triangulated-mesh conversion process was used, then this step will be unnecessary, since the model is already rendered in 3-D. But if either of the two less-expensive methods, the voxel-to-vector-polygon or manual vectorization techniques, was used, then the designer will be left with a collection of 2-D drawing files that must be combined to form a 3-D model.
To create a 3-D surface model, a designer must first build a wireframe model of the anatomy by individually inserting each drawing file into the CAD environment, positioning them relative to each other based on the center point of their fields of view.5 As each graphic is inserted into the CAD system, it must be scaled to ensure that the size of its field of view corresponds to the size of the MRI field of view. As MRI graphics are converted to a graphic file format they are often reduced, so when they are inserted into the CAD application, they much be expanded again. There are now programs, such as AutoLISP, that can automate placing the graphics in the CAD program, expanding them, and orienting them to a center point.5
A surface mesh developed from a wireframe model does not have the smooth transitions of a complex surface model.
When this process is complete, the outlines of the fields of view are deleted and a 3-D wireframe model of the anatomy remains. The components of this wireframe model can then be used to generate the 3-D surface of the anatomy and thus create a nonuniform rational B-spline (NURBS) surface model.
Surface Model Development. One way to generate the 3-D surface is to use the CAD program to create a ruled surface mesh between each of the adjacent entities of the wireframe model. This method is somewhat time-consuming, because separate surfaces must be created between each of the adjacent polyline entities of the wireframe model. Also, smooth transitions cannot be made between the adjacent surfaces.
Complex Surface Model Development. Some CAD systems are capable of generating complex surface models from the wireframe model. This type of 3-D model differs from typical surface model development in the number of entities that can be used to generate a surface and in the type of surface created. The advantages of complex surface modeling are that surfaces can be created quickly and that the surfaces are smooth. This complex modeling method produces the shape and contours of the anatomical structures more accurately than simple surface rendering does.
Solid Model Development. The wireframe model can also be used to generate a 3-D solid model. Because the anatomical drawings that make up the wireframe model can be extremely complex and nearly impossible to parametrically define, it is best to avoid solid modeling software packages that do not allow unconstrained sketches to be used to generate solid features.
To create the solid model, each drawing is simply extruded to a depth equal to the anatomical slice thickness. The disadvantage of this technique is that it generates a model with abrupt steps in its surface where two extruded sections come together. Another limitation is that the sections generated by this technique are extruded along a straight plane, which is not characteristic of most anatomical structures. However, the advantage of producing a solid model is that it can be analyzed to determine mass, volume, centroid, and other physical properties of the structure.
There are several ways MRI representations can be used in product design.
General Reference 2-D. The simplest way to use MRI data is to import the graphic file of the image into a 2-D CAD environment, scale it to the appropriate size, and use it as a reference to develop orthogonal views of the medical product. But this method is not helpful for developing a medical device that has complex shapes that must match those of anatomical structures.
General Reference 3-D. A better way to use MRI data for device design is to use a 3-D wireframe model as a reference. This method offers flexibility, because it allows the design engineer to use almost any type of CAD tool to generate the product design. However, it is not appropriate for a design in which the surfaces of the product and of the anatomy must match one another very closely.
Base Surfaces. A more sophisticated technique is to use portions of the NURBS surface model generated from the wireframe as the base surface of the product. After the product is designed onto the surface model, the anatomical model is simply trimmed away to leave only the product. One limitation of this type of surface modeling is that it may require a sophisticated modeling package or application. Also, although it efficiently combines the product with the anatomical model and maintains the complex surfaces that accurately model the anatomy, this type of model cannot be revised later as easily as can parametric solid models.
Boolean Solid Operations. Using a solid 3-D anatomical model, several Boolean operations can be used to build solid models of devices. Most CAD systems can perform a Boolean subtraction of the shape of the anatomical model from the device. Features of a medical product can also be added to the solid anatomical model by performing Boolean unions between the models. Using Boolean operations would be ideal for product modeling, except that the abrupt steps in the solid anatomical model, which are a result of the model-building technique, are transferred to the product model by the Boolean operations.
Integrated Surfaces and Solids. The most sophisticated and most accurate method of transferring anatomical model contours and shapes to a product model is to use a CAD system capable of integrating complex surface modeling with parametric solid modeling. Using a CAD system with these capabilities, the design engineer, taking advantage of advanced modeling capabilities and parametric relationships, can develop a solid product model and then integrate the shape of the anatomy into the product by performing a cutting operation on the solid to remove the complex surface anatomical model from the product. Changes to the product model can be made quickly because the solid model is parametric, with the complex surfaces providing the most accurate representation possible of the anatomical structures.
The biggest limitation to the use of MRI data in medical device design is inaccuracy of the anatomical images, which is caused by MRI scanning itself as well as by the processing of the images. Slight motions of the subject during scanning can cause the image to be inaccurate. Therefore, modeling rapidly moving structures, such as the heart, is difficult, although new technologies for this are under development. Certain other factors, such as chemical shift, can also affect MRI scans. More inaccuracy is also introduced during processing because raster-to-vector conversion software is not exact.
Data from MRI systems can be used not only for diagnostics, but also to help create medical devices that closely match patient anatomy. There are limitations to the accuracy of the MRI data and the CAD modeling of them, but as the technologies improve, this kind of processing can only continue to become faster and more precise.
MRI technology can be an especially valuable tool for designers of devices that are meant to conform to the shape of the human body. Use of CAD modeling of MRI data in device design can significantly reduce the product development cycle by taking ergonomic and human factor requirements into account up front and by reducing trial-and-error iterations necessary to fit the product to patient anatomy.
1. Kaut C, MRI Workbook for Technologists, New York, Raven Press, 1992.
2. Cline HE, Lorensen WE, Ludke S, et al., "Two Algorithms for the Three-Dimensional Reconstruction of Tomograms," Med Phys, 5(3):320-327, 1988.
3. Wallin A, "Constructing Isosurfaces from CT Data," IEEE Computer Graphics and Applications, 11(6):28-33, 1991.
4. Byrnes D, "From Raster to Vector: A Road Map," CADalyst, 11(7):40-50, 1994.
5. Kristoff J, "Converting MRI Scans to CAD Models for Device Engineering," in MD&M West Conference Proceedings, Santa Monica, CA, Canon Communications, pp 103-109, 1996.