The layer data format of scanners quickly prompted the realization that it should be possible to convert the data to be compatible with RP machine requirements. The first task was to separate the data of interest from the general information available from the scanner. Initial efforts focused on CT scan data, and fixed threshold algorithms were used to locate the edge surfaces of bone in each plane. The region-of-interest boundary was located on a pixel-by-pixel basis. Thresholds located within pixels formed "edgels" that were connected into a continuous periphery. Resulting in-plane data showed severe faceting as straight-line edges were used within image pixels. Periphery of the profiles was smoothed by fitting cubic splines or other higher order curves, yielding acceptable in-plane contours (Fig. 12.2).
Fig. 12.2. Thresholding results in edgel definition. Connected edgels form bone edge contour.
The enlargement shows connection of the pixel edgels (Materialise).
The RP models constructed were based on the coarsely spaced layer data in which a scan layer was repeated on the RP machine until it had built up sufficient material to align with the next scan layer. The resulting structure showed severe staircasing. Overcoming this problem has been a focus of both the European and Japanese programs. Several approaches lead to smoothing of the staircasing. The data between scan layers can be approximated by "morphing" from one layer to the next. Critical transformation locations must be specified. The result will be a smoother profile with a discontinuity at each layer, which can also be smoothed out by fitting splines in the vertical dimension. The results provided data that led to RP models that had excellent appearance and were accurate anatomical models of structures with significant cortical bone, the dense bone structure in the outer areas of heavier bones (Swaelens and Kruth 1993).
Although scan data is typically presented in a two-dimensional format, the film density assigned to a pixel is a measure of the average density measured throughout a volume element (voxel). Consequently, the assigned density associated with the center location of the pixel may actually more closely represent the value at a location on a plane above or below the scan plane. This spatial displacement needs to be taken into account when constructing more accurate models and models with highly figured surface profiles, such as maxillofacial, jaw and tooth, spine, and fine-boned joints of the hand and foot. Another complication that arises in modeling these structures is that when the thickness of the bone structure is smaller than the aperture window for the scanner, the average X-ray density computed for the voxel will be less than the actual bone density, possibly less than the threshold density of the data segregation algorithm. As a result, using fixed threshold data segregation may result in the edgel being located too far into the bone, yielding a measure thinner than the actual bone, or, even more serious, the voxel averaged density may not be sufficient to cross the threshold and a void artifact will be generated.
The medical modeling programs in Europe, Japan, and Australia have concentrated on the application of RP models for diagnostics and surgical planning. Their major effort has been in the development of models for patient-unique structures (as opposed to hip replacements, where a limited set of variations will satisfy all customers). There has been a major focus on maxillofacial and craniofacial reconstruction (Fig. 12.3), jawbone replacement and augmentation, and dental implants. In addition, models of the pelvis, foot, and spine have been examined in some detail. Both the European and Japanese programs recognized the limitations due to the straightforward data interpretation and developed higher order data reduction algorithms that smooth out the surfaces to more closely match the patient's actual geometry. The issue of variable thresholding and void artifacts is a continuing problem. The Australian program is principally focused on surgical applications (Fig. 12.4), mainly relying on European software. In the United States the major thrust of RP modeling has been to hip and knee prosthetics, so current capability has adequately met most needs.
Fig. 12.3. Osteotomy planning of reconstruction for Goldenhar syndrome patient.
(Dr. B. Vanassche, Eeuwfeestkliniek, Belgium).
Fig. 12.4. (Left) A cranial trauma was modeled, along with a defect obtained by a Boolean subtraction of the defect from the mirrored, blended opposite face; (right) a biocompatible insert was molded from an RP master and the fit verified before surgical implantation (Dr. P. D'Urso, University of Queensland, Australia).