Comparative Digitization Techniques: Dire Wolf Fossil Case Study
Technologies such as three-dimensional (3D) digitization and 3D printing have revolutionized the world. Today, researchers and curators rely on 3D digitization and 3D printing to assist in specimen conservation. Digitizing offers a noninvasive technique, allowing data sharing, mass replication, and preservation of fragile materials (Allard et al. 2005; Niven et al. 2009; Wachowiak & Karas 2009; Henson 2015). The purpose of this case study was to digitally reconstruct and replicate one fossil using various 3D replication techniques.
Digitization techniques vary by equipment, and many require substantial space or funding, such as for CT or MRI machines. 3D scanners differ in means of replication, resolution, and ease of user interface (Lyons et al. 2000). When replicating fossils, accuracy of the digital data is important (Allard et al. 2005; Wachowiak & Karas 2009; Kuzminsky & Gardiner 2012; Henson 2015). Scanner resolution plays a large role in accurately capturing external morphology. Accurate replicas are not limited to specimen archival and display; replicas allow morphometric analysis and comparative research (Allard et al. 2005; Wachowiak & Karas 2009; Kuzminsky & Gardiner 2012; Henson 2015). Digitizing specimens allows online image manipulation and global sharing, which is not possible with a singular physical specimen.
In this study, we assess three common digitization methods used in paleontological research. We examined cost, ease of the user interface, and replica accuracy. To determine accuracy, we scanned one dire wolf (Canis dirus) fossil ulna with three varieties of scanners. The following scanners were used: Artec EvaTM portable hand-held 3D light scanner; NextEngineTM 3D laser scanner; GE Lightspeed VCT medical CT scanner.
Replicas were printed on a ZPrinter 250 Series 3D printer. To assess accuracy, digital scans and 3D printed replicas were required to reach less than 5 percent error based on various landmark measurements. Because no agreed percent error of replicated materials exists, a 5 percent error seems reasonable for large, non-detailed objects, based on the literature (Henson 2015). In addition to accuracy, post processing versatility, and time efficiency, user interface feasibility and costs were assessed.
Materials and Methods
Table 1: Artec Eva specifications based on www.artec3d.com.
First, we tested the Artec Eva hand-held 3D light scanner, EVA model, using a Hewlett Packard laptop running Artec Studios visualization software. The Artec Eva captures images by flashing a light and recording the rebounded image. This scanner offers the options of hand-held or mounted capabilities. Because of the relatively small size of this fossil, we elected to affix the Artec Eva to a tripod and place it in a light tent. The fossil was elevated on a turntable so that the scanner could differentiate between the fossil and the background. Lighting, field of view, and recorded frames per second were adjusted within the Artec Studios program to capture the fossil accurately.
Depth of field was set manually between 400 and 600mm with geometry and texture functions selected, limiting background artifacts. Frames per second (fps) was increased to the highest available, 16fps. The scanner did not always capture at 16fps, and instead ranged from 12-16fps depending on area complexity and turntable speed. The sensitivity tool was left on normal. The fossil underwent four 360 degree scans: proximal, distal, ventral, and dorsal (Fig 1). Set up and scanning took two hours to complete.
Figure 1: C. dirus ulna highlighting four 360 degree scans used to capture the image on the Artec Eva.
Artec Eva post processing also took approximately two hours. Fifteen individual scans created the 3D model. Post processing was completed using Artec Studios 9. The first step was to remove the turntable, clay, and tent from each individual scan. Once completed, the fine serial registration tool was used on each individual scan. Fine registration uses a specific algorithm, automatically aligning textures and geometry of the scan. Fine registration assigns a registration error numerical to each frame; higher numbers reveal more scan errors in that frame. Frames are a single-scanned image; if the scanner is recording at 16fps, it captured 16 frames in one second. One scan consists of hundreds of single frames. Fine registration separates failed and bad frames, allowing manual removal and thus increasing overall scan quality. Frames with errors above 0.2 were removed. The next tool was global registration, another automatic algorithm, which selects geometric points on each frame and matches the frames together, creating one single coordinate system. Global registration fails bad frames, allowing manual removal. After removing failed frames, the fusion tool translates the scan into a 3D model. Individual scans underwent fusion, then the scans were manually aligned. Fifteen single scans were manually aligned using three or more overlapping point placements to create one 3D model. This process can result in error if one point is offset during alignment, making this the most time consuming process. The edges tool fills microscopic holes in the model. Once editing is complete, the 3D model is exported and saved as a stereolithography (STL) file.
Table 2: NextEngine specifications based on www.nextengine.com
The second digitization method evaluated was the NextEngine (Ultra HD model) 3D laser scanner, using the same Hewlett Packard laptop running ScanStudio visualization software. Unlike the Artec Eva, the NextEngine scanner captures a photo of the image to record the texture, then places it over the digital data. The NextEngine scanner creates a digital image by shining a laser onto the fossil, causing the laser to bend while capturing the reflection using two cameras located on the scanner. The NextEngine scanner includes a small turntable which automatically rotates while the scanner records, known as the auto drive. The fossil was mounted on the turntable using the part gripper, allowing the platform to automatically rotate while holding the fossil in place.
The NextEngine field of view was placed on macro mode, 130 x 96.5mm, focusing on the fossil. The 360 degree scan setting along with the maximum amount of divisions per scans (16 dps) were used. The HD speed setting was used; this is the slower speed to capture the image, and this is the highest quality. The target tool was set to neutral. Set up and scanning took three hours to complete. The fossil underwent the same four 360 degree scans, focusing on the ventral, dorsal, proximal, and distal surfaces. Bracket scan setting was used to capture the trochlear notch; this records three angles at 16 scans per angle.
Post processing took place in ScanStudios. Like Artec Studios, the first tool used was the eraser. All background artifacts were manually erased. Artifacts consisted of the turntable and the part gripper. There was less background noise compared to the Artec Eva scan session. In total, 12 scans were used to complete the digital fossil. Manual alignment using a minimum of three overlapping point placements was used. There is an automatic alignment option, but it does not consistently work, so we suggest always manually aligning. Alignment points were harder to place than with the Artec Studios, partly because the point placement had issues with drag and drop. A further challenge was that point placement occurs while automatically zoomed into the image with texture turned on, making landmark differentiation difficult. Once placed, the point would zoom into the image, making it harder to align. Like the Artec Studios, alignment is the most time-consuming process and must be precise. Once aligned properly, the fusion tool merges the models to make one digital model. The remesh tool fills microscopic holes, sealing the model. Once editing is complete, the model is exported as an STL file. Post processing for the NextEngine model took one and a half hours.
Medical CT Scanning
Table 3: GE Lightspeed VCT CT scanner specifications based on www.gehealthcare.com.
CT scanning was completed using a GE Lightspeed VCT Medical CT scanner. The CT scanner was set to the “skeletal” setting when recording. CT recording was quick, taking only five minutes. The scan was saved as a DICOM (digital imaging and communications in medicine) file and placed on a CD. The initial set up and scanning took less than one hour. The file was imported and edited in 3DSlicer, a free visual editing software for post processing.
For post processing, the fossil was removed from the DICOM data using the region of interest (ROI) selection tool. After separating the fossil from the other data, the threshold selection tool was manually adjusted to select the density range of the fossil, removing the CT bed and any other background artifacts. The DICOM file was then exported as an STL file. Post processing took around one hour to complete.
All three scanner’s versions of the digital ulna fossil were printed on a ZPrinter 250 series 3D printer. The digital STL files were measured and compared to the original fossil for accuracy. The 3D printed fossils were also measured and compared to the original fossil for accuracy. The digital fossils were measured in the Artec Studios program to ensure no program measurement bias. The 3D printed fossils were measured using Pro-Max Frowler NSK electronic digital calipers. Fifteen-point placement measurements were recorded (Fig 2) and each was measured five times to account for measurement variability.
Table 4: Measurement locations of the C. dirus ulna fossil and replicas.
Figure 2: Point placement measurements of the C. dirus ulna, digital files, and 3D printed replicas.
Results and Discussion
Limitations faced all three digitization methods. The Artec Eva and NextEngine had difficulties recording the distal fracture, the sharp edges of the ulnar diaphysis, and the trochlear notch, despite multiple angled scans used to capture the difficult areas. The Artec Eva and the NextEngine only captured the external structure, losing internal morphology.
CT results were less limited, capturing internal morphology and sharp edges. Fine details were lost due to the CT scanner layer limits, such as the paper identification tag on the dorsal diaphysis, which was too thin for the CT to record. The NextEngine and CT models more closely resembled the texture and size of the fossil. The Artec Eva replica rounded and smoothed the fossil, along with extending the size of the break and shaft thickness.
Table 5: Percent error with minimum and maximum error of digital replica and 3D printed replica.
Based on fifteen point measurements, we calculated percent error and mean percent error (entire percent error/n) of the digital models and 3D printed models. All three digital models measured above 5 percent error at the widths of the distal break, trochlear notch, lateral coronoid process, posterior ulna at trochlear notch, and length of the medullary cone. The 3D printed model error differed from the digital models showing measurement discrepancies in the Artec Studios program. These errors could be due to the inability to identify the exact location of the end of the bone, allowing point placement to be larger or smaller than the actual image. The maximum error was the width of the distal ulnar break, small with sharp edges. The NextEngine 3D print maximum error was the width of the posterior ulna at the trochlear notch. Maximum error of the CT model was also the width of the break. Visually, there are differences between the original fossil and the 3D printed replicas, with the CT 3D model more closely resembling the original fossil, followed by the NextEngine, and lastly, the Artec Eva (Fig 3).
Figure 3: Left: Original fossil beside 3D printed replicas. Artec 3D model: Blue boxes indicate area above 5 percent error based on point measurements. Medical CT 3D model: Orange box indicates area above 5 percent error based on point measurements. NextEngine 3D model: Green boxes indicate area above 5 percent error based on point measurements.
Further analysis of the point measurements comparing the replicas and the original fossil were placed in a regression model showing relatedness between the datasets. Original fossil measurements were set as the predicted value, 0. Artec Eva replicas were set as the observed values and compared to the predictor (Y=a+bX). Residual results of the Artec 3D model did not stray far from the predictor (0), with a residual variance of 0.38 (sd= 0.64). The Artec Eva digital model showed a higher residual variance of 2.21 (sd= 1.54). The same regression model ran on the NextEngine replicas using the fossil as the predictor (0). Residual variance results of the NextEngine 3D printed model was 2.78 (sd= 1.73). The NextEngine digital model residual variance was 1.13 (sd= 1.19). The same regression model ran on the CT data used the fossil as the predictor (0). Residual variance of the CT 3D printed replica was 1.66 (sd= 1.34). The Digital CT replica residual variance was 1.01 (sd= 1.04) showing a closer relation to the original fossil than the 3D printed model. Residual means of the datasets were collected, including confidence intervals (Table 6). Artec Eva 3D was consistently higher than the other replications. The CT replica had the lowest variance and smallest confidence intervals, indicating these replicas resembled the original fossil more accurately than the light scanner and laser scanner (Fig 4).
Table 6: Residual data of all of the measurements.
Figure 4: Residual means of point-placement measurements with focus on confidence intervals for replicas in relation to the control. * indicates digital model measurements.
Due to the small sample size, a bootstrap analysis was performed on the residual data to better assess statistical variability and resample residuals. Running a bootstrap creates a larger, random data set, adding more statistical variability. Each residual dataset has a 500 bootstrap sample, creating hypothetical variances. Means, variances, and confidence intervals were collected and compared to the original fossil. Variances show the range of the data compared to the original fossil material. A smaller variance indicates a more accurate replica (Table 7). The bootstrap algorithm still places the ideal means of replication as the CT scanner, followed by the NextEngine. The NextEngine and Artec Eva means and confidence intervals are very close, mean difference being 0.34. These differences do not make one significantly preferable to the other (Fig 5). In this instance, the NextEngine was able to replicate the fossil better than the Artec Eva.
Table 7: Bootstrap variance data of all of the measurements.
Figure 5: Bootstrap analysis of the data sets. Error bars show 95 percent confidence intervals of variance means. * indicates digital model measurements.
Startup and Labor
Startup costs are important when determining which digitization method with which to work. Startup cost of the Artec Eva is $22,000.00, the NextEngine is $2,995.00, and the GE Lightspeed VCT Medical CT scanner costs $424,200.00. The NextEngine is significantly cheaper and more affordable on most research budgets, whereas the other two machines could be substantially over budget. Processing time and user interface are important when working with digital replications, and impact cost efficiency. The NextEngine required four and a half hours of scanning and post processing, while the Artec Eva took four hours. User interfaces are similar, as both have specific in-house software associated with the scanners. Both interfaces allow digital manipulation, fusion, and alignment. Looking at ease of user interface, Artec Studios was easier to manipulate and learn. CT scanning and post processing took less than two hours. The included editing software was unable to create digital STL files, but the free 3DSlicer program was easy to manipulate after viewing an internet tutorial. 3DSlicer user interface was not easier to manipulate than Artec Studios, but 3DSlicer was free whereas Artec Studios costs $1,200.00. All programs required a computer with a 64-bit processor to successfully operate the 3D manipulation programs. Thus computer hardware costs are comparable among scanners.
Due to the small sample size of one fossil, this investigation cannot confirm which scanning methodology is superior, but instead shows different digitization methods available and their ability to digitize one specimen. All three machines were able to successfully replicate the fossil. When compared side-by-side with the original fossil, all three appeared relatively accurate, with the Artec Eva replica being the least accurate. When looking at accuracy, only the CT data shows internal morphology. Based on percent error and variances, CT data shows the least amount of variation from the original fossil, with the next accurate being the NextEngine laser scanner. The least accurate of the three methods was the Artec Eva. For this particular bone, the price of the scanner does not predict accuracy when creating a museum-grade replication.
In conclusion, digital copies and 3D printed replicas using the techniques outlined in this study show that fossil replication is ideal when preserving the original data, while still allowing research to be conducted without risking harm to the original material. Based on these three forms of replication, the more expensive scanner does not always guarantee more accurate replication.
I would like to acknowledge Cabell Huntington Hospital and John Napier for use of their Medical CT scanner. I would also like to thank Dr. Paul Constantino, Dr. Greg ‘G-Force’ Popovich, Dr. Suzanne Strait and Joseph Hamden for their advice, use of their equipment, and professional assistance during this project.
Allard T, Sitchon M, Sawatzky R, Hoppa R. 2005. Use of Hand-held Laser Scanning and 3D Printing for Creation of a Museum Exhibit. In: Mudge M, Ryan N, Scopigno R (Editors). Proceedings of 6th International Symposium on Virtual Reality, Archaeology and Cultural Heritage: Short and Project Papers. p 97-101.
Lyons P, Rioux M, Patterson R. 2000. Application of a Three-dimensional color laser scanner to paleontology: An interactive model of a juvenile tylosaurus sp. Basispenoid-basioccipital. palaeo-electronica.org/2000_2/mosasaur/issue2_00.htm (May, 2016).
Henson K. 2015. Stories in bones still told: digitization and replication of the Clover Site, Fort Ancient Human remains. Thesis.
Kuzminsky S, Gardiner M. 2012. Three-dimensional laser scanning: potential uses for museum
conservation and scientific research. Journal of Archaeological Science. 39: 2744-2751.
Niven L, Steele T, Finke H, Gernat T, Hublin J. 2009. Virtual skeletons: using a structured light scanner to create a 3D faunal comparative collection. Journal of Archaeological Science. 36: 2018-2023.
Wachowiak M, Karas B. 2009. 3D scanning and replication for museum and cultural heritage applications. Journal of the American Institute for Conservation. 48: 141-158.