Palaeontologia Electronica http://palaeo-electronica.org

METHODOLOGIES FOR THE VISUALIZATION AND RECONSTRUCTION OF THREE-DIMENSIONAL FOSSILS FROM THE HEREFORDSHIRE LAGERSTÄTTE

Mark D. Sutton, Derek E.G. Briggs, David J. Siveter, and Derek J. Siveter Mark D. Sutton. Department of Earth Sciences, University of Oxford, Parks Road, Oxford, OX1 3PR, U.K. Derek E.G. Briggs. Department of Earth Sciences, Wills Memorial Building, Queen's Road, Bristol, BS8 1RJ, U.K. David J. Siveter. Department of Geology, Bennett Building, University of Leicester, University Road, Leicester, LE1 7RH, U.K. Derek J. Siveter. University Museum of Natural History & Department of Earth Sciences, University of Oxford, Parks Road, Oxford, OX1 3PR, U.K.

ABSTRACT

The dominantly soft-bodied fauna of the Silurian (Wenlock) Herefordshire Lager- stätte of England is preserved as three-dimensional calcitic fossils within calcareous nodules. These fossils are difficult to interpret from study of sections produced by split- ting, and are difficult also to isolate from their matrix. CT scanning of nodules is feasi- ble, but is expensive and unlikely to be able to image the fossils at a sufficiently high resolution. We describe a technique based on serial grinding and digital photography which enables the capture of three-dimensional morphological information from the fossils. Computer video files are used to package the resulting datasets for inspection . Three-dimensional reconstructions can be created by a method involving the manual tracing of outlines on each image, but we consider this approach inappropriate in this instance. Datasets from grinding are treated instead as volumes, which are rendered either directly to produce virtual X-ray images, or by the computation of isosurfaces which are then visualized by ray-tracing. Rendering is performed from sequential angles to produce video files of rotating three-dimensional models. The method is illus- trated with reconstructions of a trilobite and graptolite associated with the Hereford- shire soft-bodied fauna, and of the worm-like mollusc Acaenoplax hayae. Reconstructions produced in this way can be retouched easily and manipulated to remove noise and preservational artifacts, to dissect out structures for detailed study, or to combine part and counterpart into a single reconstruction. The approach will pro- vide the primary means of study of the Herefordshire fauna, and has the potential to provide a method of obtaining and/or communicating morphological information in a variety of palaeontological applications. KEY WORDS: isosurface, reconstruction, Silurian, soft-bodied, three-dimensional. Copyright: Palaeontological Association, 22 June 2001 Submission: 23 March 2000 Acceptance: 12 March 2001

Sutton, Mark D., Briggs, Derek E. G., Siveter, David J., and Derek J. Siveter, 2001. Methodologies for the Visualization and Reconstruction of Three-dimensional Fossils from the Silurian Herefordshire Lagerstätte. Palaeontologia Electronica, vol. 4, issue 1, art. 1: 17pp., 1MB. http://palaeo-electronica.org/2001_1/s2/issue1_01.htm MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

Figure 1. Detail of fine spines preserved on margin of Acaenoplax hayae Sutton et al. (2001), from the Herefordshire Lagerstätte; OUM C.29503b.

INTRODUCTION shapes within a lighter-coloured matrix (Briggs et al. 1996; see also Orr et al. The non-biomineralized fauna of the 2000). Few unambiguous internal features Herefordshire Lagerstätte (Briggs et al . of the organisms are known, but external 1996) represents a rare opportunity to details are commonly reproduced with a sample the true taxonomic composition degree of fidelity sufficient to preserve fine and palaeoecology of a Silurian benthic spines 30 µm or less in diameter (Figure community. Fossils occur in spherical to 1). subelliptical calcareous nodules found in a All taxonomic and palaeoecological locally thick bentonite deposit in the Coal- study of the Herefordshire biota must be brookdale Formation, Wenlock Series . underpinned by careful resolution of the They vary in size from under a millimetre morphologies of the fossils preserved. to several centimetres, with worms, However, while three-dimensional preser- sponges, and non-biomineralized arthro- vation is clearly richer in morphological pods dominating the assemblage. All taxa information than the two-dimensional or are preserved in three dimensions, with no near two-dimensional modes of most other indication of post-mortem disturbance or Konservat-Lagerstätten (e.g., the Burgess compression. While some fossils preserve Shale, see Allison and Briggs 1991), it organic material (e.g., the graptolites) and introduces special problems. Surfaces others siliceous skeletons (e.g., the radi- exposed by splitting nodules are essen- olarians), for the most part specimens tially randomly oriented with respect to the consist of dark-coloured sparry calcite organism and show only those structures

2 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS that happen to be intersected by the split. many cross-sections are obtained from Structures of fossils are thus easily over- parallel planes intersecting the . Data looked or misinterpreted. Furthermore, it is of this form can be collected in several dif- frequently difficult to distinguish the genu- ferent ways, but all approaches produce ine termination of a structure, such as the datasets that can be reconstructed into body of a worm, from the situation where it three dimensions with the aid of a com- passes out of the plane of section. These puter. factors make attempts to reconstruct mor- phology from such ‘natural’ splits difficult APPROACHES TO SLICE DATA ACQUISITION and feasible only when a large number of Non-destructive approaches specimens are known or when the mor- phology is relatively simple. If essential Two commonly used medical scan- morphological data are to be obtained for ning technologies, Computed Tomography the majority of taxa, to which neither of (CT) and Magnetic Resonance Imaging these conditions apply, then as much of (MRI), use non-invasive techniques to pro- the three-dimensional form as possible duce ‘slice-based’ datasets representing must be extracted from a single specimen. respectively, the X-ray absorption, and The simplest and most common degree of magnetically induced nuclear approach to determining the three-dimen- resonance of materials in the sample. The sional form of a fossil is to separate it from latter technique is not well suited to solid the matrix, either chemically or mechani- geological materials, but CT has been cally. Chemical dissolution techniques, used successfully to provide three-dimen- though widely used in many branches of sional data on fossils (Hamada et al. 1991; palaeontology (including the recovery of Torres 1999). Although our preliminary Konservat-Lagerstätte fossils such as the experiments have shown that the Here- ‘Orsten’ faunas; see Müller et fordshire fossils within nodules can be al. 1995) require fossils and matrix to be detected by CT scanners, generally avail- chemically dissimilar. Although dissolution able equipment is capable at best of slice has been used to extract microfossils such spacings and pixel sizes of around 0.3 mm as radiolarians from the Herefordshire to 0.5 mm. Specialised CT equipment nodules (Briggs et al. 1996), the predomi- (e.g., the University of Texas High Resolu- nantly calcitic nature of both matrix and tion X-ray Computed Tomography Facility) macrofossils precludes its use in the study can, under ideal conditions, achieve reso- of most of the faunal elements. Mechani- lutions over an order of magnitude finer cal excavation of the Herefordshire fossils than this, and might thus in principle be a is possible and has some utility for the viable tool for studying the fauna. How- study of large and robust structures (e.g., ever, time on such equipment is expensive Briggs et al. 1996, fig. 1c, e). However, it is and its ability to consistently distinguish both extremely time consuming and inap- fossil from matrix in the Herefordshire propriate for resolving small or fragile material at high resolutions is as yet structures, and thus is not a satisfactory unproven. While CT may provide a valu- approach for the study of the fauna as a able tool for investigating large or whole. In view of these difficulties, our extremely rare taxa, we do not consider it study has concentrated not on the physi- to be a viable approach to the routine cal isolation of the fossils but on tech- study of this fauna. niques based on serial slices, in which

3 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

Serial slicing Baker 1978; Hammer 1999). Grinding has been carried out by other researchers with Serial sectioning of fossils is the nor- a variety of pieces of dedicated equip- mal approach to the collection of ‘slice- ment, the most popular of which have based’ datasets. However, the term ‘serial been variants of the ‘Croft Grinder’ (Croft sectioning’ is commonly used to encom- 1950; see also Ager 1965; Sandy 1989). pass both serial slicing by saw or blade The principle differences between our data and serial grinding (Ager 1965, p. 213). acquisition technique and those used by These are substantially different tech- most previous authors are the relative sim- niques, and we prefer to use the separate plicity of the grinding device, and the terms to distinguish between them. method of data capture—photography at Serial slicing is less commonly used regular intervals rather than manually pre- on fossils than serial grinding, although it pared drawings often at irregular intervals. does allow much of the original material to be retained whilst providing the required Considerations for Data Acquisition by Serial multiplanar views. The method is best Grinding Photography. suited to larger specimens than the Here- The destructive nature of serial grind- fordshire fossils because intact slices sub- ing places special demands on the pho- stantially thinner than 1 mm are extremely tography of each freshly exposed surface. difficult to produce by sawing. In contrast, The photographic technique used must biological microtome equipment can gen- not only capture as much information as erate exceptionally fine slices (as small as possible, but it must have a very low fail- 1 µm). However, microtomy requires a ure rate, because images cannot be re- non-brittle sample capable of forming a taken after grinding has progressed. Digi- thin cohesive slice—properties not associ- tal rather than conventional photography is ated with crystalline materials. Almost cer- therefore a key element of our method tainly for these reasons, experiments on since digital images can be downloaded the use of microtomy on Herefordshire instantly to a computer where they can be fossils were unsuccessful, despite their checked for quality before grinding contin- vacuum impregnation with resin. ues. The use of a digital camera also Serial grinding offers significant economic savings as well as removing the need for images to be Serial grinding involves the sequential scanned into a computer for reconstruc- removal by abrasion of small thicknesses tion. of material from a single planar surface Registration. Although the process- which is photographed at each stage. ing of images from serial grinding is dis- Although this approach is destructive, it cussed below, one aspect must be suffers from none of the problems of a lack considered here as it impacts on the early of cohesiveness associated with slicing, stages of the technique. Before any recon- and inherently produces very flat polished struction can be attempted images must surfaces which are ideal for photography. be ‘registered’ (or aligned) so that the For these reasons we consider it the only arrays of pixels line up correctly. This can viable approach to obtaining high-quality be accomplished in one of two ways: 1) serial images from the Herefordshire fos- the sample can be placed in exactly the sils. Serial grinding is not a new technique same position relative to the field of view and has been used extensively on fossils of the camera for each image capture; or of widely differing sizes (e.g., Sollas and 2) the sample can be processed so that it Sollas 1913; Stensiø 1927; Ager 1956-67;

4 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS contains physical ‘fiduciary’ (or alignment) markings of some sort, that can be used subsequently to align the images. The second approach is simpler in practice and is preferred in our work. Fiduciary markings used in similar work by previous authors are, typically, straight holes drilled perpendicular to the grinding surface so that they appear in the corners of the image (e.g., Herbert 1999, figure 3.2; Baker 1978). However, holes less than 0.5 mm in diameter cannot be produced reli- ably with current drilling technology and larger holes would be unacceptable in the context of this study. Consequently, we use two perpendicular edges (see below) as fiduciary structures, which are roughly aligned in the field of view when images are first captured. These edges are used to guide precise digital alignment during the registration phase of post-processing Figure 2. Grinding process. (see Appendix 1, section 1) .

METHODOLOGY FOR GRINDING ground away before the ceramic comes AND DATA CAPTURE into contact with the grinding plate and prevents any further removal of material. Specimens selected for serial grinding Specimens are ground manually on a (Figure 2.1) are first sawed to produce a glass plate (Figure 2.5) with 600 grade small block with two faces perpendicular carborundum powder and water to remove to each other and to the direction of grind- thicknesses of approximately 30-50 µm at ing, and as close to the specimen as pos- a time. After each thickness is removed, sible. These side faces are then polished the slide holder is inverted and placed on to obtain the sharp edges at their junction its base. The surface is then washed and with the ground face (Figure 2.2). The polished for a few seconds with a small flat block is set within a resin cuboid or cylin- plate coated in fine-grade aluminium oxide der (Figure 2.3), that is then mounted with powder and water (Figure 2.6). This final wax onto a Wirtz Buehler Slide Holder polishing improves image quality without (Figure 2.4). Note that while Figure 2 illus- removing any appreciable thickness of trates a grinding run performed on a single sample. The surface is then washed fossil, multiple specimens are normally again, and the slide-holder placed under a mounted within one resin block for greater Leica MZ8 binocular microscope. An efficiency. The slide-holder consists of an image of each specimen and alignment inner metal cylinder that, by means of a edges (Figure 2.7) is then captured by screw thread, can be positioned accu- means of a digital camera attached to the rately to protrude any required distance microscope operating at 1000 by 800 pixel beyond the plane of an outer hard ceramic resolution in 24-bit colour. The image is disc. This enables a fixed thickness to be checked for quality prior to filing as part of

5 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS a sequential set. All images for a grinding a three-dimensional reconstruction in the run are taken at the same magnification, accepted sense and are best used to com- with the same camera settings, and under plement rather than substitute for such consistent lighting conditions; only the reconstructions. focus is re-adjusted for each image. The Construction of three-dimensional models grinding process is then repeated for as many iterations as required. The time The ‘surfaces’ method. Two method- taken per image depends on how many ologies exist for the construction of three- specimens are mounted at once for a dimensional models from data based on grinding run, but on average it takes six slices (Brown and Herbert 1996; Herbert minutes to grind a 30 µm interval and one 1999). The first, the ‘surfaces’ approach minute to photograph each specimen. (Herbert 1999; see also Chapman 1989), After completion of a grinding run each begins with a manual interpretation step, dataset is prepared for use by registration tracing all structures of interest from each of the images (see Appendix 1, section 1). slice image (Figure 4.1-4.3). Traced out- lines are digitized into line segments form- VISUALIZATION AND RECONSTRUCTION ing closed loops (Figure 4.4). The reconstruction algorithm assembles these Slice videos stacked two-dimensional loops into a Computer video files can be assem- three-dimensional geometry, typically a bled directly from the datasets resulting mesh of triangular facets (Figure 4.5), from grinding (see Appendix 1, section 2), which can be rendered into a two-dimen- providing a simple way of leafing back- sional view by a number of standard com- wards and forwards through the slice puter graphics techniques. The two key images (Figure 3). These files form an advantages of this method are that it, (1) invaluable tool for the exploration of does not require photographic images of datasets and, although less striking than full three-dimensional reconstructions, are often more useful for tracing the origins and termination of small or obscure struc- tures. Nonetheless, they do not constitute

Figure 3. Slice video showing raw dataset (distal por- Figure 4. Vector-based reconstruction process of Her- tion of monograptid graptolite; OUM C.29526a). bert (1999).

6 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS the slices and, (2) is capable of working value of some property of the object at that with low-quality, sparsely-spaced data. point. The sequential slice images gener- Neither of these is relevant to the present ated by the grinding technique (after minor study. This method also has several disad- post-processing — see Appendix 1, sec- vantages, especially in the context of the tion 3), represent a volume of this sort in Herefordshire fauna. The initial interpreta- which the values of the voxels represent tion stage is extremely time-consuming the reflectivity of the surface as captured when a large number of slice images are by the digital camera. Volume-based involved, and also threatens the objectivity three-dimensional reconstructions, largely of the reconstruction as structures must be utilizing datasets from MR or CT scanners, delineated before the three-dimensional are routinely used in medical applications form becomes apparent. When the mar- and have been used to produce volume- gins of the fossil structures are unclear (as based reconstructions of fossils (e.g., is often the case with the Herefordshire Hamada et al. 1991; Torres 1999). These fossils), there is a danger that subjective methods have also been applied to opti- interpretations will influence the recon- cally captured volumes. Brown and Her- structed morphology. In addition, the algo- bert (1996) attempted to produce volume- rithmic reconstruction of stacked two- based models from serial section data, dimensional loops into three-dimensional although the low resolution of the grapto- surfaces is subject to the ‘correspondence lite sections on which that work was based problem’–that is, the difficulty of determin- precluded a successful reconstruction. ing the manner in which loops in subse- More successfully, Hammer (1999) recon- quent slices connect to each other. structed a halysitid coral colony from vol- Herbert (1999) approached this problem ume data obtained by serial grinding and with algorithms that made a priori assump- captured with a flatbed scanner. tions about the morphology of the object to Volume rendering. There are various be reconstructed, an approach that is techniques, collectively known as volume clearly not viable for the reconstruction of rendering, for the direct visualization of poorly understood morphologies such as volume data . Although they vary in detail, those of many Herefordshire fossils. all involve forms of ‘ray casting’, where the Other available approaches require the computer calculates an array of virtual user to determine correspondence, a pro- light rays ‘shone’ into the volume from a cedure that is both time consuming and specified angle, determines their fate represents another potential source of according to mathematical rules based on subjectivity. the values of voxels in the volume, and Volume-based methods. The second builds a two-dimensional image from the reconstruction methodology, that of vol- results of these calculations. We have ume-based reconstruction, is preferred in used mean-intensity volume rendering, the present study. Volume representation where the final intensity of the virtual ray is is a three-dimensional extension of ‘raster’ determined by the mean values of the vox- or ‘bitmap’ representations of two-dimen- els it encounters. This algorithm produces sional images, in which an image is mod- a virtual ‘X-ray’ image of the volume from elled as a rectangular array of regularly any angle the user chooses (Figure 5). spaced pixels. A volume dataset thus con- Like true X-radiographs, the images are a sists of a three-dimensional array of regu- relatively poor aid to visualizing the outer larly spaced volume elements or ‘voxels’, surface of objects, but are effective for each stored as a number representing the imaging of large or strongly absorbent

7 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

Figure 6. Ray-traced isosurface visualization of anterior portion of a cranidium of the trilobite Tapinocalymene Figure 5. Volume-rendered ‘X-ray’ visualization of distal vulpecula Siveter 1980; OUM C.29527a. 172 slices at portion of monograptid graptolite; OUM C.29526a. 165 50 µm interval. Specimen is approximately 16 mm slices at 30 µm interval. Specimen is approximately 5 mm wide.Top is anterior. long.

(=dark) structures that otherwise may be Animation. Isosurfaces have been obscured by less substantial features. used previously to produce static images Isosurfaces. While volume data can of fossils (e.g., Torres 1999; Hammer thus be rendered directly, the preferred 1999; Brown and Herbert 1996). While approach for medical visualization from such images have some illustrative value, CT or MR data is usually to generate poly- a true appreciation of three-dimensional gon mesh surfaces known as isosurfaces. morphology is best achieved with an inter- These are generated by algorithms that active model. We produce these (for both connect points of a constant (user-deter- isosurface and volume rendering tech- mined) intensity within the volume. If the niques) from a batch of images rendered threshold level is chosen correctly, an iso- from sequential angles and assembled surface generated from a dataset such as into a computer video file that can be pre- shown in Figure 3 should correspond to sented as a reconstruction of the speci- the original external surface of the organ- men (Figures 7-11). As with video files assembled directly from slice images, ism. Isosurfaces can be rendered with any of several computer graphics tech- these files can be played backwards or niques available for the visualization of forwards and provide a simple but effec- geometrically defined surfaces. Ray-trac- tive means of viewing the reconstruction ing, though the most computationally from a variety of angles. They also enable expensive of these techniques, provides the reconstruction to be distributed easily the clearest and most natural visualiza- for viewing on any modern computer with- tions (Figure 6). Although volume-ren- out the need for dedicated software. dered ‘X-ray’ reconstructions have some Pitfalls for three-dimensional reconstructions application to the Herefordshire fossils, ray-traced isosurfaces are our preferred Transparency of calcite. The nature means of three-dimensional visualization. of the Herefordshire material necessitates Reconstructions produced in this way guarding against two potential pitfalls. The retain the objectivity of volume data with- first concerns the assumption that each out compromising the clarity and sharp- image represents the intersection of the ness of the rendered images. fossil with a single plane. The calcite that comprises the fossils is normally dark and

8 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

Figure 9. AVI format video file of ray-traced isosurface reconstruction of anterior portion of a cranidium of the trilobite Tapinocalymene vulpecula Siveter 1980; OUM C.29527a. 172 slices at 50 µm interval. Specimen is approximately 16 mm wide.

Figure 7. AVI forat video file of ray-traced isosurface reconstruction of distal portion of monograptid graptolite; OUM C.29526a. 165 slices at 30 µm interval. Specimen is approximately 5 mm long.

Figure 10. AVI format video file of ray-traced isosurface reconstruction of anterior fragment of the mollusc Acae- noplax hayae Sutton et al. 2001; OUM C.29532. 190 slices at 30 µm interval. Specimen is approximately 5 Figure 8. AVI format video file of Volume-rendered ‘X-ray’ mm long. reconstruction of distal portion of monograptid graptolite; OUM C.29526a. 165 slices at 30 µm interval. Specimen is approximately 5 mm long. occur around specimens of certain taxa (Figure 12). Fossil material within these opaque, but translucent regions can haloes is often no darker than the matrix reveal features beneath the surface plane outside them, making the selection of a or be relatively light in shade when matrix single isosurface threshold level for the lies only a short distance below the plane. whole fossil impossible. Reconstructions Such occurrences, which have the poten- may therefore fail to include extremities tial to invalidate reconstructions partially, within these regions that are nonetheless are detected by inspection of the slice clearly visible on the slice images. While images and corrected by manual retouch- this problem can be overcome by careful ing (see Appendix 1, Section 3). manual darkening of these regions, a sim- ‘Halo’ effects. Light-coloured regions, pler alternative is often to prepare a dis- presumed to be reduction haloes, often section (see below) of the inside of a light

9 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

fossils may be obtained by gluing part and counterpart together before sawing to size so that the entire fossil can then be ground through as one unit. However, this approach can produce artifacts from imperfect alignment of the glued surfaces or from the glue itself. Alternatively and more satisfactorily, part and counterpart can be reconstructed separately and reas- Figure 11. AVI format video file of ray-traced isosurface sembled digitally before rendering. Figure reconstruction of antero-median fragment of the mol- 13 is an example of a whole fossil recon- lusc Acaenoplax hayae Sutton et al. 2001; OUM structed in this way. C.29529. 243 slices at 30 µm interval. Specimen is approximately 7 mm long. Dissection of models Reconstructions of portions of a fossil, region, and to reconstruct this section sep- or of a subset of the structures present, arately. are often of great utility. A strength of the EXTENSIONS OF RECONSTRUCTION volume-based approach described here is the ease with which such ‘dissected’ TECHNIQUES reconstructions can be prepared. Copies Whole fossil reconstructions of the appropriate slice images are simply manipulated using image editing software Reconstructions from a single grinding to remove structures or portions of the fos- run typically comprise just part of a fossil, sil that are not required (see Appendix 1, as grinding normally commences from the section 3), and are then rendered in the plane along which the nodule was origi- normal way. Slice videos and pre-dissec- nally fractured. Reconstructions of entire tion three-dimensional reconstructions

Figure 12. Reduction haloes surrounding an unnamed annelid; OUM C.29525a.

10 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

ogy. Virtual models of fossils are useful tools for the dissemination of morphologi- cal information, as they contain more infor- mation than static images, yet can be copied, distributed, and viewed quickly and at minimal cost. Applications in edu- cation can be envisaged as a means of familiarizing students with morphology of fossils, particularly where material is too large, small, or rare to be physically manipulated. Computer-based models Figure 13. AVI format video file of ray-traced isosur- may also prove to be an efficient way of face reconstruction integrating part and counterpart of conveying morphology to professional a cranidium of the trilobite Tapinocalymene vulpecula palaeontologists, particularly those less Siveter 1980; OUM C.29527a-b. 29527a (anterior): 172 slices at 50¨µm interval. 29527b (posterior): 197 slices familiar with the group in question. These at 50 µm interval. Specimen is approximately 13.5 mm techniques might also provide a methodol- long medially. ogy for capturing morphological informa- tion as a prelude to further computer- provide the ideal tools to guide such dis- based modelling work. Such studies might section work, and ensure that unwanted include the reconstruction of skeletal structures are removed accurately. structures (e.g., conodont feeding appara- tus) from isolated material, the illustration APPLICATIONS OF RECONSTRUCTION of theoretical models of unpreserved soft TECHNIQUES parts (e.g., brachiopod musculature), and studies of functional morphology that The techniques described herein, require digital models (e.g., finite element although largely illustrated with well- stress analysis). understood biomineralized taxa, have been developed to elucidate the morphol- ACKNOWLEDGEMENTS ogy of the essentially soft-bodied organ- isms that dominate the Herefordshire We thank Jenny Corrigan, Juliet Hay, fauna (such as Acaenoplax hayae, Fig- Jeremy Hyde, and Philip Jackson for ures 10-11, see also Sutton et al. 2001), assistance with various aspects of speci- and provide the primary means for palae- men processing. This work is funded by ontological studies. However, they are NERC Grant GR3/12053. easily adapted to other situations and essentially require only that the fossil pre- REFERENCES serves three-dimensional form and is visu- ally distinguishable from matrix. Despite Allison, P. A., and Briggs, D. E. G. 1991. p. 120-140. The the destructive nature of grinding we thus taphonomy of soft-bodied . In Donovan, S. K. (ed.), Fossilization: the process of taphonomy. expect that similar applications, whose Belhaven Press, London. main goal is simply to determine morphol- Ager, D. V. 1956-67. A Monograph of the British Liassic ogy, will arise in other situations where iso- Rhynchonellidae. Palaeontographical Society, Monographs, p. 1-172. lation of the fossils in question is Ager, D. V. 1965. Serial grinding techniques. p. 212-224. impractical. In Kummel, B. and Raup, D. (eds.), Handbook of These techniques also have broader Paleontological Techniques. W. H. Freeman, New potential applications within palaeontol- York.

11 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

Baker, P. G. 1978. A technique for the accurate recon- elling and analysis of fossils and their distribu- struction of internal structures of micromorphic fos- tions. John Wiley and Sons, Chichester. sils. Palaeontology, 19:565-584. Müller, K. J., Walossek, D., and Zakharov, A. 1995. Briggs, D. E. G., Siveter, David J., and Siveter, Derek J. ‘Orsten’ type phosphatized soft-integument preserva- 1996. Soft-bodied fossils from a Silurian volcaniclas- tion and a new record from the Middle Cambrian tic deposit. Nature, 382:248-250. Kuonamka Formation in Siberia. Neues Jahrbuch Brown, I. M., and Herbert, M. J. 1996. Knowledge-based für Geologie und Paläontologie Abhandludgen, geological visualisation using AVS. Electronic Geol- 197:101-118. ogy [online], 1 (Special volume on the proceedings Orr, P. J., Briggs, D. E. G., Siveter, David J., and Siveter, of the Geoscience Information Group on Geological Derek J. 2000. Three-dimensional preservation of a visualisation; the intelligent picture?):37-47. Available non-biomineralised arthropod in concretions in Sil- from Internet: http://www.electronicjournals.co.uk/eg/ urian volcaniclastic rocks from Herefordshire, pdf/EG1-5.pdf England. Journal of the Geological Society, Lon- Chapman, R. E. 1989. Computer assembly of serial sec- don, 157:173-186. tions. p. 157-164. In Feldmann, R. M., Chapman, R. Sandy, M. R. 1989. Preparation of serial sections. p. E., and Hannibal, J. T. (eds.), Paleotechniques. 143-150. In Feldmann, R. M., Chapman, R. E., and Paleontological Society Special Publication 4. Hannibal, J. T. (eds.), Paleotechniques. Paleonto- Croft, W. N. 1950. A parallel grinding instrument for the logical Society Special Publication 4. investigation of fossils by serial sections. Journal of Sollas, I. B. J., and Sollas, W. J. 1913. A study of the Paleontology, 24:693-698. skull of a Dicynodon by means of serial sections. Hamada, T., Tateno, S., and Suzuki, N. 1991. Three Philosophical Transactions of the Royal Society dimensional reconstruction of fossils with X-ray and of London, 204: 201-225. computer graphics. Scientific Papers of the Col- Stensio, E. A. 1927. The Downtonian and ver- lege of Arts and Sciences, University of Tokyo, tebrates of Spitzbergen. Skrift, Svalbard Nordis- 41:107-118. havet, 12: 1-31. Hammer, Ø. Computer-aided study of growth patterns in Sutton, M. D., Briggs, D. E. G., Siveter, David J., and tabulate corals, exemplified by Catenipora heintzi Siveter, Derek J. An exceptionally preserved vermi- from Ringerike, Oslo Region. Norsk Geologisk form mollusc from the Silurian of England. Nature, Tidsskrift, 79:219-226. 410: 461-463. Herbert, M. J. 1999. Computer-based Serial Section Torres, A. M. 1999. A three-dimensional CT (CAT) scan Reconstruction. p. 93-126. In Harper, D. A. T. (ed.), through a rock with alga Ivanovia tebagaen- Numerical Palaeobiology: Computer-based mod- sis. Journal of Paleontology, 73:154-158.

12 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

APPENDIX 1: DETAILS OF COMPUTER-BASED PROCESSING

This represents a step-by-step docu- image; these ensure that the object is mentation of the computer-based ele- exactly the same size as the image, which ments of the reconstruction process used assists with placement in the next step. to produce slice videos and isosurface Once the alignment object is complete it is models. For each step we provide both a placed into the clipboard with a copy com- ‘package-independent’ overview (in bold), mand and the file closed. followed by ‘package-specific’ details of The registration process now com- our method. The latter are intended only mences, and the following procedure is as a description of our procedures, not as performed for each image. The image to a set of instructions for those unfamiliar be aligned is loaded, and the alignment with the software involved, nor indeed as object is inserted with the ‘Paste As New an exclusive recommendation of any spe- Object’ command—this alignment object cific software package—equivalents of all is not moved subsequently, but used as a major packages used are available for guide for aligning the image. The image most platforms. Details of software pack- itself is converted to an object with the ages used in the study are given in ‘Create From Background’ command. This Appendix 2, and terminology specific to image object is then shifted manually with these packages is placed within inverted the ‘Object Picker’ tool selected, by use of commas on first usage. the cursor keys, so that the point where the two alignment edges meet is correctly 1) Registration and cropping positioned. The rotational centre of the Images are first registered object is then dragged to that point, and (aligned), using the alignment edges in the correct rotation of the image applied by the images as a guide. An image- means of trial and error entries of angles editing software package is used to in the ‘Tool Settings Roll-up’. When and shift and rotate the images manually, only when the correct rotation is deter- so that the edges are in exactly the mined, it is applied with the ‘Transform’ same position. Once registered, the button on the roll-up, and any final single- images are cropped to remove extrane- pixel shifts required for perfect alignment ous background, including registration are performed with the cursor keys. edges. All images must be cropped Finally, the alignment object is removed using exactly the same parameters so from the image which is saved in its origi- that alignment is preserved. nal format. This process must be carried One image, normally the first, is cho- out for each image in the dataset and is sen as the datum for alignment. This thus fairly time-consuming. Much of the image is loaded into PhotoPaint, and a work can be automated, however, by new ‘object’ is created to lie above the recording actions as ‘scripts’. We normally background. Straight lines of a single pixel make use of one script to create the nec- in width are drawn on this object to corre- essary objects, and another to remove the spond to the alignment edges of the image alignment object after alignment. In prac- beneath. The ‘Eraser’ tool is then used to tice, we find that it is possible to register remove portions of these lines, so they up to 60 images per hour. appear as coarse dashes. Oblique lines After registration, the images are are also drawn on the object from edge to cropped to size with the ‘Deskew Crop’ edge near each of the four corners of the tool. Care must be taken that the crop

13 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS area set includes all of the fossil on every shire fossils are usually more clearly image. Once the correct crop size is deter- distinguished from the matrix if the mined, the action is recorded as a script, conversion to monochrome is per- and then applied to all images with the formed using the red channel only. ‘Batch Playback’ command. These operations, which must be iden- tically performed on all images in the 2) Post-processing for preparation of slice dataset, are carried out with an image videos editing software package. Slice videos, if required, can be pre- Any required retouching of images pared directly from the registered and should be carried out at this step, cropped image set using a suitable although in most cases we prefer to software package. Typically such vid- reconstruct an un-retouched or mini- eos are prepared from a second copy mally retouched set of images first, of this dataset, the images of which are then return to this point in the process colour balanced and/or reduced in size to perform any required ‘cleaning’ on a prior to video production. This pro- second copy of the dataset. Retouch- cessing is carried out with image-edit- ing is carried out manually using an ing software. Size reduction is image-editing software package. In necessary only if the resulting video most cases it comprises the deletion of would otherwise be too large to store objects in the matrix that appear in the or play on any given computer. reconstruction but are clearly not part To prepare for video production, a of the main fossil, or of structures visi- copy of the registered and cropped image ble through translucent regions of cal- set is made, and the following PhotoPaint cite. Retouching can also include the operations are applied first to one image, brightening of structures that are recorded as a script, and then applied to clearly part of the main fossil, but are each image by means of the Batch Play- too dark to appear in the reconstruc- back command. tion (i.e., are falling below the chosen a. Colour is balanced to taste with the isosurface threshold, usually because ‘Level Equalization’ command. of transparency problems). We find it b. Image size is reduced, if necessary, helpful to choose an isosurface thresh- with the ‘Resample’ command. old during the retouching step, so that Processed images are opened as a these problems can be assessed within file sequence video in Ray Dream, and the image editing software package. then saved as an AVI video file in Cinepak Post-processing operations are per- format. formed on each image using PhotoPaint. As in Step 2 above, they are first per- 3) Post-processing for isosurface formed on one image, the actions reconstruction recorded as a script, and then applied to To produce a dataset suitable for each image by means of the Batch Play- the isosurface algorithm, the registered back command. and cropped dataset must be con- a. The image is converted to mono- verted from colour to monochrome, chrome. This can be accomplished with inverted (so that fossil is light and the ‘Convert to Greyscale (8-bit)’ com- matrix dark), and reduced in size if the mand. If only the red channel is required restrictions of the isosurface package (see above), the Level Equalization com- so demand. We find that the Hereford- mand is used to remove the green and

14 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS blue signals, the Convert to Greyscale (8- gram. This requires the images be in Win- bit) command is then applied, and finally dows BMP format (uncompressed), be the ‘Brightness-Contrast-Intensity’ com- consecutively numbered with three digits mand is used to enhance the visual clarity (i.e., file names of the form of the image (which will be too dark). Name001.bmp, Name002.bmp), and b. The image is inverted using the placed in a single directory. The user ‘Invert’ command. selects the first file, clicks the process but- c. If necessary, the image is resam- ton, and when requested enters the num- pled to a smaller size. AVS (see below) ber of files in the dataset. On completion, can only accept volumes up to 255 x 255 x the program places the AVS format data 255 voxels in size, so the image size must file in the same directory, with the name be reduced to 255 x 255 pixels or smaller ‘output.dat’. before a reconstruction can be attempted. An AVS ‘network’ is created, as fol- Resampling ideally is carried out with an lows. The data file produced in Step 3 is integral divisor (e.g., at 50% or 33.33%); read with the ‘read volume’ module, which arbitrary values can introduce an element feeds the ‘average down’ module. This of blurring, although this is rarely of critical should be set with relatively high x, y, and importance. The final image width must z coefficients (2 or more) at this step. The also be a multiple of 4 to satisfy output from the ‘average down’ module is AVSCONV. This can be accomplished fed to the right hand input port of the ‘isos- either by padding with black space using urface’ module, the output of which is sent the ‘Paper Size’ command, or by further to the right hand port of the ‘geometry careful cropping. viewer’ module. The isosurface level Retouching, if required, is also carried should now be set to produce the best out in PhotoPaint, mainly by use of the possible reconstruction; as noted above ‘Eraser’ tool to delete objects, and the we normally find it helpful to choose this ‘Effect’ tool to brighten others. The level during retouching. When the isosur- ‘Threshold’ command can be used to face appears correct, the coefficients of apply the chosen isosurface threshold the ‘average down’ module should be level and determine what will and what will reduced to 1, and the output of the ‘isosur- not appear in the reconstruction. The face’ module connected to the ‘write threshold effect can then simply be geom’ module, which should be set to pro- removed with the ‘Undo’ command. duce a ‘DXF’ file. As the processing involved in the network can be time con- 4) Isosurface production suming, we find it helpful to disable the The post-processed image set must flow executive whilst making changes to first be converted into whatever volume parameters of these modules. format the isosurface software DXF files produced by the ‘write geom’ requires, using appropriate software. module can be very large. We find that Once informed of the appropriate files much over 100MB in size are imprac- threshold level the isosurface software tical to render in Step 5 below. In some can then generate the surface. The sur- cases therefore, particularly with datasets face may then need to be converted to containing a large number of images, the a format that can be read by the ray- final isosurface is produced with x and y tracing software employed. settings of 2 rather than 1 in the ‘average We convert the dataset to AVS volume down’ module. Although we normally pre- format using the custom AVSCONV pro- fer not to average down in the z axis, if

15 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS alignment errors are producing strong arti- ‘1 in.’. Once imported, the isosurface is facts (lineation parallel with the plane of stretched in the z axis to correct for slice grinding), a z setting of 2 can help to spacing—the z scale property of the isos- smooth these out, as well as to reduce urface object is set to 100% x (Pixels/mm output file size. in resampled images) / (slices/mm). Any other objects are imported and treated in 5) Rendering the same way, and carefully moved into Geometry files describing isosur- position relative to each other. When faces are read into a ray-tracing soft- bringing two objects together, we prefer ware package, or any other three- first to bring a single corresponding point dimensional visualization package. Iso- into contact, move the ‘hotpoint of rotation’ surfaces are stretched to correct for to that point, and then rotate the object differences in voxel spacing within into position using the ‘virtual trackball’. A slices and between slices (this may be ‘shader’ is then applied to the object(s), done during Step 4 instead, if the soft- lights placed in the scene, and an anima- ware allows). If multiple isosurfaces are tion set up to rotate the object in the to be joined (e.g., Figure 11), they are desired fashion over a time period of sev- both imported into one scene, and then eral seconds. The ‘production frame’ is manually moved into a contiguous carefully positioned and sized to the mini- position. The visualization package is mum size necessary to contain the object then used to generate a video file of the at all steps of the animation. The anima- specimen rotating in the desired fash- tion is then rendered within Ray Dream, ion. normally with an image width of 300 to400 DXF files produced by AVS are read pixels and a frame rate of 15 or 18 frames into Ray Dream with the ‘Import’ com- per second. It is saved as an AVI video file mand. All the default import options are in Cinepak format, which forms the final left unaltered, except ‘Scaling Conversion reconstruction. Factor’ which is set to ‘0.5 cm’ rather than

16 MARK D. SUTTON, DEREK E. G. BRIGGS, DAVID J. SIVETER, & DEREK J. SIVETER: 3D VISUALIZATION OF FOSSILS

APPENDIX 2: SOFTWARE AND EQUIPMENT

1) Commercial Software AVSCONV version 1.1 (for Image editing: ) Author: Mark Sutton PhotoPaint version 8 (for Microsoft Windows) URL for download: Corel Corporation palaeo-eletronica/2001_1/s2/program Corporate Headquarters Export from AVS geometry format to 1600 Carling Avenue DXF for import into RayDream: Ottawa, Ontario ‘write geom’ AVS module [Must be K1Z 8R7, Canada installed into AVS system] www.corel.com Author: Georg Thallinger Rendering and production of slice vid- URL for download: eos: http://www.iavsc.org/repository/avs5/ Ray Dream 3D version 1.02 (for source/data_output/write_geom/ Microsoft Windows) 3) Equipment used MetaCreations Corporation 6303 Carpinteria Ave. SV Micro Digital Microscope Camera Carpinteria, CA 93013, USA Sound Vision, Inc. www.metacreations.com 492 Old Connecticut Path Framingham, MA 01701 Isosurface production: http://www.soundvisioninc.com/ AVS version 5 (for Solaris) Advanced Visual Systems Inc. Wirtz-Buehler ‘Slide Holder’ (hand- 300 Fifth Avenue Waltham, held grinder) MA 0245, USA Buehler Krautkramer UK www.avs.com Milburn Hill Road University of Warwick Science Park 2) Coventry CV4 7HS United Kingdom Conversion of image files to AVS vol- http://www.buehlerltd.com/ ume format:

17