Med. Eng. Phys. Vol. 19, No. 1, pp. 90-96, 1997 Copyright 0 1997 Elsevier Science Ltd for IPEMB. All rights reserved

Printed in Great Britain

ELSEVIER PII: S1350-4533(96)00039-2 1350-4533/97 $17.00 + 0.00

Preliminary experience with medical applications of rapid prototyping by selective laser sintering

E. Berry*, J. M. Brown-t, M. Connellf, C. M. Craven& N. D. Effordll, A. Radjenovic* and M. A. Smith*

*Medical Physics, The University of Leeds, Leeds, UK; tApplied Mathematics and Mechanical Engineering, The University of Leeds, Leeds, UK; $Medical Physics, Western General Hospital, Edinburgh, UK; SDepartment of Radiology, St James’s University Hospital, Leeds, UK; [IComputer Studies, The University of Leeds, Leeds, UK

Received 29 November 1995, accepted 10 May 1996

ABSTRACT Rapid prototyping techniques, originally deweloped fi building components from computer aided designs in the motor

industry, are now being applied in medicine to build models of human anatomy from high resolution multiplanar

imaging data such a computed tomography (CT). The established technique of stereolithography and the more recent

selective laser sintaing (SLS) , both build up an object layer by layer. Models have applications in surgical planning, for the design of customised implants and for training. Preliminary experience of using the SLS technique for medical

a@lications is described, addressing questions regarding image processing, data transfe and manufacture. Pilot

models, built from nylon, included two skulls (a child with craniosynostosis and an adult with hypertelorism) and

a normal femur which was modelled fw use in a bioengineering test of an artificial hip. The dimensions of the

models were found to be in good agreement with the CT data from which they were built--fo the childs skull the

difference between the model and the CT data was Less than l.M.05 mm in each direction. Our experience showed

that, with care, a combination of existing softiare packages may be used for data conversion. Ideally, image data of

high spatial resolution should be used. The pilot models generated suf)icient clinical interest for the technique to be

pursued in the mthopaedic$eld. 01997 Elseuier Science Ltd for IPEMB. All rights reserved.

Keywords: Selective laser sintering, computed tomography, three-dimensional imaging, medical imag-

ing, anatomic models

Med. Eng. Phys., 1997, Vol. 19, 90-96, January

Computer visualization techniques for rendering

1. INTRODUCTION

3D reconstructions from high resolution multi- planar imaging techniques such as computed tom- ography (CT) and magnetic resonance imaging (MFU) are well established’*2. These techniques have found applications in radiotherapf and for surgical planning particularly in the craniofacial and maxillofacial fields and for neurosurgery4,5. Systems provide interactive functions which allow the surgeon to experiment with cutting and mov- ing bony materia16,‘; such systems may also be used to obtain measurements for customized pros- thetic implants, often designed by mirroring of the unaffected side. There are occasions when 3D viewing may be inadequate and the surgeon would prefer to be able to handle a physical

where the pyocedure is complex, the anatomy unusual and difficult to visualise on screen or the

model when planning a procedure; this is the case

procedure involves the use of prostheses which have not been included in the computer simul- ation. Physical models are also useful as moulds for the fabrication of customized implants. Early models were built using milling techniques’, but a significant limitation of such methods is the need to avoid undercuts. This means that it is impossible ‘to produce models with certain com- plex geometries where parts of the objects obstruct the access of the milling tool.

The advent of rapid prototypmg technologies overcame these problems. There are several tech- niques, each of which builds up a model layer by layer. These include laminated object manufac- ture (LOM) . stereolithoaraohv and selective laser sintering (SLS) . Laminayed&object manufacturing laminates thin sheets of material such as paper, after each layer is attached a laser is used to cut

Correspondance to: Dr E. Beny, Medical Physics, Wellcome Wing, Leeds General Infirmary, Great George Street, Leeds LSl 3EX, UK

Medical application, of rapid potol@ing: E. Berry et al.

human skull may be defined using around several hundred thousand facets this can lead to extremely large data files. International standards for more compact formats are currently under dis- cussion,

P articularly in relation to CAD data

exchange ‘. An alternative format, the common layer interface (CLI) based on slices rather than facets, has been devised within a BRITE EURAM project (BE52’78: Rapid Prototyping in the Auto- motive Industry) but has not yet entered general usage. Software is available commercially to con- vert segmented image data to slice formats such as CLI, complete with smoothing functions and the addition of support structures for stereolithog- raphy (CT-Modeller, Materialise) . No suitable commercial packages were available for our sinter- ing equipment (Sinterstation 2000, DTM Corporation) but we wished to assess the feasi- bility of the technique and produce models to stimulate clinical interest-preferably within a few months and without a large financial burden. The questions which needed to be addressed fell into three categories: image processing, data transfer and manufacture.

and remove unwanted areas. The result is a physi- cal model. In stereolithography the object is built layer by layer by the polymerization of selected regions of a tank of UV sensitive resin, resulting in a translucent model. The polymerization is per- formed selectively, using a laser to illuminate an area defined from CT data. Each layer fuses to the one below, allowing the creation of the complex type of structure not achievable by milling. Software is available to generate support struc- tures for regions of the object such as overhangs which need supporting during the build. These structures are removed once the model is com- plete. Stereolithography models have been shown to have a precision for a human skull model5 of f1.3 mm and have been used quite widely in max- illofacial sur tering 8

ical planning’“*“. Selective laser sin- (SLS) * also involves the selective use of a

laser to build up a model layer by layer, see Figwe f. The material used is a fine powder (at present, nylon, polycarbonate or wax) whose particles adhere and solidify (or sinter) under laser illumi- nation. SLS has the advantage over stereolithogra- phy that support structures are not required, as the unsintered powder provides support during the build. SLS models are opaque, and the devel- opment of materials is an area of strong current research interest. Powders which give a metal modellg or are biocompatible, such a hydroxya- patite’“, are under research and development. ,

Although SLS models have been built using thresholded CT slice data directly’“, most SLS installations require data to be provided in the STL format which gives a description of a closed faceted surface 16. This compute r model is then sliced for the build. The STL format is verbose; the coordinates of each vertex are listed for each face with which it is associated, and hence are repeated at least three times. As a model of a

Laser

1. Image processing Could data from the segmentation modules in our existing image processing package be out- put in a contour format amenable to conver- sion to STL format? What degree of user interaction would be required in the segmentation process? Was it necessary to perform segmentation on images with the maximum available spatial and grey level resolution, or could the demands be relaxed to reduce data storage requirements and processing times? What further processing may be necessary to improve results?

2. Data transfer Was public domain software available which could be used to convert contour information to a closed faceted surface?

3. Manufacture Does the process result in a model of the cor- rect dimensions? Are there any practical points to be learned?

2. METHOD

2.1. Examples

The pilot examples chosen were the skull of a year-old child diagnosed as having craniosy- nostosis and a section of an adult femur which could be used to aid the testing of an artificial hip in the Bioengineering Laboratories at the Univer- sity of Leeds.

Fiie 1 The selective laser sintering process. A roller levels a thin layer of powder in the build area (A), a laser then scans a selected area of the powder, consolidating the powder as it does so (B), for- ming a new layer of the object.Wherk the new layer overlaps the previous layer a bond is automatically formed. The volume in which the model is being built is then lowered and the powder feed car- tridges raised to supply fresh powder (C) and the ororess r~near~=d ----- m-r-----

until the model is complete

2.2. Imaging

In both cases images were acquired on the Sie- mens Somatom Plus SCT scanner at St James’s University Hospital in Leeds and transferred bv network Across-the city to the University site. tij

91

Medical applications of rapid prototyping: E. Berq et al.

scans were transverse. For the child’s skull the in- plane pixel size was 0.44 mm and a slice recon- struction interval of 1 mm, with a pitch of 1, was used18. The data set used for the femur was not specially acquired for this purpose and covered the whole of the pelvis, the in-plane pixel size was 0.63 mm and the reconstruction interval was 3 mm.

2.3. Image processing

Image segmentation was performed using tools in the AnalyzeTM (7.0) biomedical image processing package . lg Preliminary tests showed that the struc- ture of the skull could not be successfully seg- mented without significant operator interaction, so, to reduce the time needed for interactive seg- mentation and the associated data storage requirements the skull data were converted to 8 bits and to cubic voxels on a 256x256 matrix by grey level interpolation. This had the effect of reducing both the grey level resolution and the in-plane spatial resolution. In contrast, with the smaller data set and rather less complex bony structure of the femur, data processing was perfor- med using the full 16 bits, and grey level interp- olation was performed to increase the apparent spatial resolution to an in-plane pixel size of 0.42 mm. The slice thickness was kept at 3 mm during segmentation. Because data had not been acquired specially for building the model femur it did not extend far enough inferiorly and the shaft of the model would have been too short for use in laboratory measurements. The shaft length on the model was increased by replicating the low- est slice a number of times. Segmentation was accomplished slice by slice using a combination of region growing and manual tracing, and text files containing the coordinates of contours from each slice were generated.

A 3D computer reconstruction was generated from each data set by volume rendering with the threshold set at the level used for the initial region growing stage of the segmentation. The recon- struction was compared with the model to help identify artefacts introduced by the processing.

2.4. Data transfer

Software in the C language was written to pre-pro- cess the contours by removing or reporting inter- sections and converting the file format to that required for the next stage of processing. Intersec- tions can occur when coordinates are repeated at the end of a contour, when single pixel-wide spurs occur, or where edge tracking is confounded by a narrow isthmus or adjacent objects. A faceted surface representation was created from the con- tour data using the public domain tool Nuages (Bernhard Geiger, INRIA, France); other tools suitable for performing this task include the marching cubes algorithm20 and utilities in many CAD/CAM packages. Additional software was writ- ten in the C language to convert the output to both ASCII STL and binary STL formats. The pro- cedure is shown schematically in Figure 2.

CT Slices

“.c-> ,̂ convetion to STL format

Generation of faceted surface computer model

Model buiff by selective laser sintering

Figure 2 Schematic diagram showing the steps in the process of cre- ating a model from CT data by selective laser sintering

2.5. Manufacture

Models were built from nylon powder on the Sin- terstation 2000 (DTM Corporation, Austin, TX). The layer thickness was 0.1 mm and beam diam- eter 0.3 mm. Prior to the build the computer model was scaled using standard techniques to compensate for material shrinkage during the build. The scaling factor used depends on the material used and the size of the model; for the child’s skull the factor was 4.55% in the x and y directions and 1.5% in the .z (slice) direction. For the femur, the computer model was also scaled to compensate for the anisotropic coordinate system. Each model was built together with a number of different objects for economy, and the build ses- sion lasted between 12 and 48 h. Once sintering was complete, models were broken out from the unsintered powder using the hands, brushes and a powder blaster (Figure 3).

Validation of the dimensions of the first model (the child’s skull) was performed by making measurements in three orthogonal directions on both the model and the CT scans on which it was based.

Figure 3 The unsintered powder is removed from a model in the breaking out procedure

Medical applications of rapid pototyping: E. Bemy et al.

3. RESULTS

The manufactured nylon models are shown in Fig- ure 4.

3.1. Image processing

The model and 3D computer reconstruction of the child’s skull are shown in Figure 5, it can be seen that some surface detail has been lost in the segmentation and faceting process, for example the sagittal and coronal sutures are seen more clearly on the computer reconstruction. The 8 bit data set used for the child’s skull was found to have adequate grey level resolution for segmen- tation and it was concluded that the 16 bits retained for the femur were not necessary. The higher spatial resolution of the femoral data was found to ease segmentation and as expected gave higher fidelity in the model.

3.2. Data transfer

Although the data transfer process was generally successful in generating an STL file acceptable to the Sinterstation, the surfaces produced had defects which were introduced by the surface faceting process in Nuages, where the internal sur- face had been wrongly joined to the outer one. Triangular artefacts measuring approximately 2 mm by 1 mm by 0.5 mm deep were apparent bn the top of the child’s skull and on the head of the femur.

3.3. Manufacture

Comparative measurements from CT scans and the model of the child’s skull showed that the model was consistently larger than expected. On investigation this was found to be due to the build mistakenly being performed using a voxel size of 1 mm when the true value was 0.88 mm. This was an error in communication and not an error in the CT scan parameters or the manufacturing process. Sizes corrected for the error (Tubk I) show similar agreement in the three directions, with the difference between the model and the CT data less than l.OkO.5 mm in each direction.

Figure 5 (2mqmrison of a) a Sd computer rwonstn~rtion and b) a photo of the model

The large percentage error is made up from con- tributions from the voxel size of the CT volume and model measurement accuracy. The mean dif- ference was 0.7kO.9 mm.

4. DISCUSSION

Figure 4 The Leeds child’s skull (left), the Edinburgh skull (right) and thr femur (front)

4.1. Image processing

Modules in our existing image processing package were found to be suitable for use in this work, but required considerable user interaction for the examples chosen. During segmentation some fea- tures which can lead to problems with faceting, sllch as the occurrence of small spurs, were easily

93

Medical applications of rapid prototyping: E. Berry et al.

Table 1 Comparative measurements from CT scans and the model

CT of skull (scaling as for model)

Model Difference

Total height (mm) 162.01kO.4 161.W0.3 l.oH.5 AP at nasium (mm) 137.5f0.4 137.5kO.3 O.OkO.5 Maximum lateral size (mm) 147.5f0.4 148.5fo.3 l.Of0.5

dealt with. Objects which narrow to an isthmus of one pixel wide (and one or more pixels long) were a constant difficulty in this work and the resulting contours usually had to be edited manu- ally. This problem arises when the edge-tracking algorithm defines the contour coordinates in terms of the centre of the pixel, leading to repeated values if tracking passes along both sides of a single pixel line. Using an alternative defi- nition, such as the mid-edge point definition which is im lemented in a more recent version of AnalyzeTM (5.5)) avoids this difficulty. No change in contour definition can help automated seg- mentation, however, if partial volume effects lead to apparent gaps in bone, if metal artefacts obscure detail in the CT image or disease has rendered tissue boundaries indistinct: at present segmentation is a skilled task and may require interaction from surgeon or radiologist. By- products of the pre-processing include data suit- able for multiple object volume rendering and CAD-style models which could be of use in finite element analysis studies or for virtual reality simul- ations.

The reduced spatial resolution used for the child’s skull had a clear detrimental effect on the resulting model and it was concluded that the maximum spatial resolution compatible with avail- able disk space should be used, in spite’ of the resulting increased processing time. For both ster- eolithography and SLS the limiting factor is the accuracy of the CT data and of the segmentation, as the manufacturing processes themselves are accurate2’ to 0.25 mm. CT spatial resolution is highest for closely space reconstructions, such as an interval of 1 mm with a slice width greater than 2 mm and a pitch of one22.

The model of an adult skull shown at the right of Figure 4 was built using data from the University of Edinburgh of an adult patient with hypertelor- ism. The data had previously been used to pro- duce a model from stereolithography. In this case the contouring algorithm used for segmentation was based on a method of sub-pixel interpolation described by Diamond23 and the resulting model has a notably smoother surface than the pilot models. This smoothing feature was introduced because in spite of grey level interpolation resulting from pre-processing, raw contours tend to have an outline with discrete, stepwise features (Figure 6) which does not correspond with the smooth shape expected from our knowledge of human organs. Smoothing of the individual con- tours can also be accomplished b approximation methods as described by Udupa J4. , a similar rou- tine is available in Nuages, but was not

94

Figure 6 Schematic diagram to illustrate stepwise features on a con- tour and the effect of smoothing. Approximation methods (a) smooth the contour and reduce the number of points. Interpolation (b) gives s&pixel resolution and a smoother shape

implemented on the models of the child’s skull or the femur. Alternative methods have been described by Heffernan and Robb25 and Hill- manz6. Further refinement is nossible using: inter- slice smoothing olation27’28.

using &ape-based ?nterp-

4.2. Data transfer

The use of a public domain package to facilitate data transfer allowed us to proceed to model building reasonably quickly, but the software chosen did introduce artefacts which reduced the surface quality of the models. Surface rendering by faceting has tended to fall out of favour in medical imaging and volume rendering tech- niques have taken its place. This is because of the difficulties with faceting when the number of con- tours changes from slice to slice, as it often does, and the tendency for unrealistic-looking patterns of facets24. Work continues on surface techniques; promising recent developments are designed to reduce the volume of the data set whilst preserv- ing detail by reducing the number of redundant faces2g,30 and in fitting smooth surfaces to con- tour data31.

Experience with the pilot models led to the preparation of a pro-forma to be completed and passed on with the data at each stage of the pro- cessing, this will avert communication problems related to scaling and other important features in the future. A further possible area of uncertainty is related to the definition of the coordinate sys- tem. Analyze TM has its origin in a location corre- sponding to the posterior in transverse CT scans. The origin in the Sinterstation is anterior; a differ- ence which, if not allowed for, results in a model

with right and left reversed. To check the orien- tation of the model geometry and general appear- ance before the build the faceted model output by Nuages can be viewed with a visualization package such as Geomview (The Geometry Center, Minne- apolis, USA).

4.3. Manufacture

Within measurement errors the model was of the size expected confirming that the shrinkage cor- rection and scaling of the SLS manufacturing method was performing as expected. A practical aspect noted was that since close objects may fuse, objects should be placed well apart in the build volume if it is desired to separate parts of the model, such as the femur from the acetabulum.

The long-term stability of the sintered powders of SLS bodes well for the models retaining their geometry over a period of time. The SLS process has the advantage that no support structures are required and hence it is possible to build a com- plete skull; with stereolithography it may be neces- saw to build the top of the skull separately. In addition, where automatic support structure gen- eration systems have been used it can be difficult to identify which structures should be removed after the build, and thus can lead to apparent anomalous anatomic features on the model. SLS polycarbonate models can be used as patterns for investment casting, thus extending the possible applications. In many cases the SLS models are preferred to stereolithography models, because the opaque material looks and feels more like bone. In situations where one might want to fol- low the course of an internal channel the trans- lucent resin material may be more appropriate. There remains considerable potential for the development of new materials for SLS both to make them more suitable for multi-stage casting processes and also for directly implantable bio- compatible materials. Interest here is focused on sinterable hydroxyapatite-based materials. Similar materials are already in use for prosthesis manu- facture by conventional mean2’ and for implants based on stereolithographic model shape”. Fused deposition modelling (FDM) is an alternative rapid prototyping technique which allows differ- ent materials to be used for different parts of the model, and indeed recent developments in stereo- lithography allow for two colour models which can depict for example, the extent of lesions more clearly. Currently the whole SLS model must be built of the same material, which is a drawback, but perhaps in the future specially designed materials could result in a model with dif‘ferent parts having different properties.

The cost of SLS models is high, due to equip- ment and materials costs and the labour intensive image processing stage. For example build costs for the child’s skull were 21350, inclusive of machine depreciation, materials, labour costs and overheads; a commercial bureau charge may be taken to be higher. It can be argued that the manufacturing costs are offset by time savings at surgery and will fall with experience and technical

developments. Image processing time was increased by the developmental aspects, and is estimated to have been around one person-week for the child’s skull because of the manual con- tour correction required. However, experienced and active centres in medical stereolithography such as those in Australia34 and Belgium (each has now built about 100 models) report much lower image processing times, of the order of hours. Progress is being made in these centres towards providing a same-day service which would broaden the potential applications to include trauma surgery. To encourage uptake, a database is currently under development by the European Action on Rapid Prototyping, which will include case studies from many centres, but mainly stereo- lithography models. As the information collected is anecdotal, there remains scope for a health technology assessment study in this area.

The accuracy of these initial models was adequate for orthopaedic applications, and led to further work where a pelvis was built to assess the practicality of bone grafting in planning a total hip replacement. Here, the opportunity to hold the model in the hand and view it from various angles in a natural fashion was appealing, and offered immediate comparison with the prosthesis to be used in the surgery. Possible future uses include building models of soft tissue organs such as the heart and vessels, for example, of aortic aneurysms for training surgeons in the placement of stems. Scaled models of bone structure are planned from microscope data. Applications extend outside the surgical and medical field and we have recently built an SLS model of a Bronze Age skull found buried in sand on the Isle of Lewis. A museum plans to reconstruct the skin surface using clay to provide an exhibit, similar work has been reported of the use of a stereoli- thography model in anthropolo$“.

Although a number of difficulties remain with selective laser sintering it is likely that with experi- ence and technological development these will be overcome and the process holds promise as an alternative to stereolithography for the routine production of models for clinical use.

ACKNOWLEDGEMENTS

The authors are grateful to Andrew Marsden, the CARP project team andjames Pakianathan at the University of Leeds; Bernhard Geiger the author of Nuages and Chris Davies at CNSoftware.

REFERENCES

Udupa, J. IL., Hung, H. M. and Chudng, K S., Surface and volume rendering in three-dimensional imaging: a comparison. j. Digit Imaging, 1991, 4. 159-168.

Arridge, S. R., Grindrod, S. R., I.inney, A. D., Tofts, P. S. and Wicks, D., Using greyscale voxel databases for improved shading and segmentation. Mdirnl Jn&rmatic.s,

1989, 14, 157-171. Rosenman, J. G., Chaney, E. I,., Cullip, T. J., Symon, J. R.. Chi, V. I,., Fuchs, H. and Stevenson, D. S.. VISTAnet:

Interactive real-time calculation and display of three- dimensional radiation dose: an application of gigabit net-

95

Medical applications of rapid prototyping: E. Berry et al.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

working. Int. J. Radiation Oncol. Biol. Phys., 1993, 25, 123-129. Vannier, M. W., Hildebolt, C. F., Marsh, J. L., Pilgram, T. R, McAlister, W. H., Shackelford, G. D., Offutt, C. J. and Knapp, R. H., Craniosynostosis: diagnostic value of threedimensional CT reconstruction. Radiology, 1989, 173, 669-673. Mankovich, N. J., Samson, D., Pratt, W., Lew, D. and Beumer, J., Surgical planning using three-dimensional imaging and computer modeling. Otolayngologic Clinics of Narth America, 1994, 27, 875-889.

Vannier, M. W., Marsh, J. L. and Warren, J. O., Three dimensional CT reconstruction images for craniofacial surgical planning and evaluation. Radiology, 1984, 150, 179-184. Moss, J. P., Grindrod, S. R., Linney, A. D., Arridge, S. R. and James, D., A computer system for the interactive planning and prediction of maxillofacial surgery. Amer-

ican Journal of Orthodontics and Dentofacial Orthopaedics,

1988, 94, 469475. Gillespie, J. E., Isherwood, I., Barker, G. R. and Quayle, A., Threedimensional reformations of computed tom- ography: The assessment of facial trauma. Clinical Badi-

olqy, 1987, 38, 523-526. Mankovich, N. J., Cheeseman, A. M. and Stoker, N. G., Three-dimensional anatomy with stereolithography mod- els. Journal of Digitital Imaging, 1990, 3, 200-203. Arvier, J. F., Barker, T. M., Yao, Y. Y., D’Urso, P. S., Atkin- son, R. L. and McDermant, G. R., Maxillofacial biomodel- ling. Br. J Oral Maxillofac. Surg., 1994, 32, 276283. Bill, J., Wittenberg, G., Reuther, J. F. and Michel, C., Ster- eolithography-based operation planning in maxillofacial reconstructive surgery. European Radiology, 1993, 3(suppl), 250. Ashley, S., Rapid prototyping systems. Mechanical Engin-

eering, 1991, 113, 34-43. Agarwala, M. R, Bourell, D. L., Beaman, J. J., Wu, B. and Marcus, H. L., Structurally sound metal parts by selective laser sintering. In Proceedings of EPD Congress, ed. G. W. Warren. 1994, pp. 833-851. Lee, G. and Barlow, J. W., Selective laser sintering of cal- cium phosphate powders. In Proceedings of the Solid Free- form Fabrication Symposium, eds H. L. Marcus, J. J. Beau- man, J. W. Barlow, D. L. Bourell and R. H. Crawford. 1994, pp. 191-197. Levy, R. A., Guduri, S. and Crawford, R. H., Preliminary experience with selective laser sintering models of the human temporal bone. American Journal of Neuroradiology

1994, 15, 473-477. Keppel, E., Approximating complex surfaces by triangu- lation of contour lines. IBM Journal of Research and Devel-

opment, 1975, 19, 2-11. Bloor, M. S. and Owen, J., A History of Product Data Exchange Standards. UCL Press, London, 1994. Craven, C. M., Naik, K. S., Blanshard, K. S., Batchelor, A. G. G. and Spencer, J. A., Multispiral three-dimensional computed tomography in the investigation of craniosy- nostosis: technique optimization. British Journal of Badi-

ology, 1995, 68, 724730.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

Robb, R. A., A software system for interactive analysis of biomedical images. In 3D imaging in Medicine, eds K H. Hohne, H. Fuchs and S. M. Pizer. NATO AS1 Series F60, 1990, pp. 333-361. Lorensen, W. E. and Cline, H. E., Marching cubes: a high resolution 3D surface reconstruction algorithm. Computer Graphics, 1987, 21, 163-169. Childs, T. H. C. and Juster, N. P., Linear and geometric accuracies from layer manufacturing. Annals of the CZRP, 1994, 43, 163-166. Kalender, W. A., Polacin, A. and Suss, C., A comparison of conventional and spiral CT: an experimental study on the detection of spherical lesions. Journal of Computer Assisted Tomography, 1994, 18, 167-176. Diamond, R., Two contouring algorithms. In Compu- tational Crystallography, ed. D. Sayre. Oxford University Press, Oxford, 1982, pp. 266-272. Udupa, J. K., 3D visualization of images. In Digital Zmag- ing, eds W. R. Hendee and J. H. Trueblood. Medical Physics Publishing, Madison, WI, 1993, pp. 325-402. Heffernan, P. B. and Robb, R. A., A new method for shaded surface display of biological and medical images. IEEE Transactions on Medical Imaging, 1985, M14, 26-38. Hillman, G. R., Elder, K and Sheppard, L. C., Building solid models from serial contour images. SPIE Image and Video Processing, 1995, 2421, 152-160. Raya, S. P. and Udupa, J. K, Shape-based interpolation of multidimensional objects. IEEE Transactions on Medical

Imaging, 1990, MI9, 3242. Mole, C., Gerard, H., Mallet, J. L., Chassagne, J. F. and Miller, N., A new three-dimensional treatment algorithm for complex surfaces: applications in surgery.J. Oral Max- illofac. Surg, 1995, 53, 158-162. Odesanya, 0. S., Waggenspack, W. N. and Thompson, D. E., Construction of biological surface models from cross sections. IEEE Transactions on Biomedical Engineering, 1993, 40, 329-334. Kalvin, A. D. and Taylor, R. H., Superfaces: polyhedral approximation with bounded error. SPIE Image Capture, Formatting and Display, 1994, 2164, 2-13. Bookstein, F. L., Prinicipal warps: thin-plate splines and the decomposition of deformations. IEEE Transactions on

Pattern Analysis and Machine Intelligence, 1989, PAMI-11, 567-585. Ripamonti, U., Calvarial reconstruction in baboons with porous hydroxyapatite. J. Craniofac. Surg., 1992, 3, 149- 159. Nishiyama, T., Nakajima, T., Yoshimuta, Y. and Nakani- shi, Y., Utilizing solid models for preoperative shaping of HAP-TCP ceramic bone substitute-application for craniomaxillofacial surgery. European Journal of Plastic Sur-

gely, 1994, 17, 173-177. Yao, Y. Y., Arvier, J. F. and Barker, T. M., Technical note: maxillofacial biomodelling-preliminary result. British

Journal of Radiology, 1995, 68, 519-523. 35. zur Nedden, D., Knapp, R., Wicke, K., Judmaier, W.,

Murphy, W. A., Seidler, H. and Platzer, W., Skull of a 5,300-year-old mummy: reproduction and investigation with CT-guided stereolithography. Radiology, 1994, 193, 269-272.