3D Reconstruction (Digital Reconstruction)

The conceptual foundations of 3D reconstruction in paleontology extend back over a century. In 1903, William Johnson Sollas published a method for investigating fossils by serial sectioning, in which specimens were progressively ground down and each exposed surface was drawn or photographed (Sollas, 1903, Philosophical Transactions of the Royal Society of London, Series B, 196: 259–265). These serial section images could then be used to understand the three-dimensional internal anatomy of fossils—a destructive but pioneering approach that laid the groundwork for all subsequent volumetric visualization of fossil material.

The true revolution in digital 3D reconstruction began with the application of X-ray computed tomography to paleontological specimens. In 1984, Glenn Conroy and Michael Vannier published a landmark paper demonstrating that high-resolution CT scanning could non-invasively generate three-dimensional images of matrix-filled fossil skulls (Conroy & Vannier, 1984, Science, 226: 456–458). This study established CT as a viable, non-destructive alternative to physical sectioning. Over the following decades, the technology spread rapidly, and by the 2010s, CT scanning had become the most widely employed digitization method in vertebrate paleontology.

Several distinct scanning technologies underpin the digitization step that precedes all 3D reconstruction. Each has characteristic advantages, limitations, and resolution ranges.

X-ray Computed Tomography (CT) and Micro-CT: Standard CT scanners use X-rays to penetrate an object and produce a tomographic dataset consisting of hundreds to thousands of cross-sectional slices. Micro-CT scanners, commonly found in research institutions, achieve resolutions on the order of 1–100 μm but are generally limited to specimens up to approximately 50 cm in diameter. Medical CT scanners offer lower resolution and energy but can accommodate much larger specimens and may be available at no additional cost. CT is currently the most commonly applied approach for digitizing fossil specimens because it captures both external and internal morphology non-destructively.

Synchrotron Radiation X-ray Tomographic Microscopy (SRXTM): Particle accelerators produce synchrotron radiation that enables ultra-high (sub-micrometer) resolution tomography. This technique has proven transformative for studying exceptionally preserved microfossils, embryonic stages, and fine anatomical details such as cellular-level preservation. However, the effective specimen size is typically restricted to a few centimeters, and access to synchrotron facilities is limited.

Laser Scanning: Surface-based laser scanners actively probe and characterize the external surface of a specimen. They can achieve sub-millimeter resolution and are particularly well suited for very large objects. Many laser scanners are portable, which makes them convenient for fieldwork or museum settings. However, they cannot capture internal structures.

Photogrammetry: This technique generates a digital 3D model by computationally processing multiple photographs taken from different angles around a specimen. It is highly cost-effective, easy to implement, and applicable to both very small and very large objects. Resolution is largely dependent on camera quality and the number of photographs. Photogrammetry has become increasingly popular for documenting and sharing fossil specimens, especially in educational and outreach contexts.

Following the important terminological distinction introduced by Lautenschlager (2016), digital restoration is the broader process of removing preservational artifacts to recover the morphology of a fossil as it was prior to fossilization. Digital reconstruction, in contrast, specifically refers to the creation of structures not directly preserved. In practice, the two terms are often used interchangeably in the broader literature, and the entire workflow is frequently referred to as '3D reconstruction' in common parlance.

The restoration process typically proceeds through a series of modular steps, each adding a degree of interpretation:

Removal of Breaks and Cracks: Small taphonomic imperfections (cracks, holes, surface irregularities) are removed through smoothing algorithms, region-growing operations, or manual interpolation during CT segmentation. Although this step introduces minimal interpretation, even minor surface artifacts can affect the outcomes of computational analyses such as FEA.

Reflection (Mirroring): Vertebrates exhibit bilateral symmetry, which can be exploited to replace missing elements on one side with a mirror image of the corresponding preserved element on the other side. This is one of the most commonly applied techniques in palaeoanthropological cranial restoration and has been widely adopted in vertebrate paleontology.

Superimposition: When corresponding elements on both sides of a specimen are incompletely preserved but retain different portions, they can be superimposed to create a more complete composite element.

Repositioning: Disarticulated fragments are digitally translated and rotated in three-dimensional space to reassemble them in their anatomically correct positions. This may be guided by articulation surfaces, break morphology, or landmark-based alignment algorithms.

Duplication: For serially repeated elements that lack bilateral counterparts (e.g., vertebrae), preserved elements can be duplicated and scaled to reconstruct missing portions of the series.

Retrodeformation: This is arguably the most challenging step, aimed at reversing post-mortem plastic deformation that has altered a fossil's shape without breaking it. Methods exploit bilateral symmetry using landmark-based geometric morphometric approaches: paired landmarks on either side of the midsagittal plane are used to warp the specimen back toward its original symmetrical configuration. DeVries et al. (2022) developed a reproducible workflow using armatures in the open-source software Blender that records and animates each retrodeformation step, greatly enhancing transparency and reproducibility.

Extrapolation: When missing portions cannot be replaced by reflection or duplication, they must be estimated. This can draw on information from other specimens of the same taxon, closely related taxa, or general anatomical constraints. Extrapolation introduces the highest degree of interpretation and is used cautiously.

Beyond the restoration of hard-tissue morphology, 3D reconstruction techniques have opened avenues for investigating soft-tissue anatomy in extinct organisms. Osteological correlates—features on bone surfaces such as muscle scars, attachment ridges, and foramina—provide evidence for the size, position, and orientation of muscles, blood vessels, nerves, and other soft tissues. Using CT-derived 3D models as scaffolds, researchers digitally reconstruct musculature and other soft tissues in 3D modeling software. Lautenschlager (2013, Journal of Anatomy, 222: 260–272) demonstrated a novel approach for digital muscle reconstruction in the therizinosaur Erlikosaurus andrewsi, using CT data to model jaw adductor musculature and estimate bite force. Similarly, Witmer and colleagues (2008) used CT to visualize and reconstruct brain endocasts, endosseous labyrinths, and neurovascular structures in dinosaurs, crocodilians, and birds, providing insights into sensory capabilities and cognitive function.

The software landscape for digital 3D reconstruction in paleontology is diverse, spanning commercial and open-source options. For CT segmentation, commercial packages such as Amira/Avizo (Thermo Fisher Scientific), VG Studio Max (Volume Graphics), and Mimics (Materialise) are widely used and offer powerful automated and manual segmentation tools. Free alternatives include SPIERS (developed specifically for virtual paleontology), 3D Slicer, and Dragonfly. For mesh editing, retrodeformation, and reconstruction, Blender (Stichting Blender Foundation) has emerged as the dominant open-source platform in paleontology, offering sculpting, rigging, animation, and rendering capabilities. MeshLab (Visual Computing Lab, ISTI-CNR) provides a free, lightweight tool for mesh processing. The R package Morpho and the software Landmark are used for geometric morphometric-based retrodeformation.

3D reconstructions serve as the substrate for a broad range of quantitative analyses:

Geometric Morphometrics (GMM): Landmark-based and surface-based morphometric methods quantify shape variation and covariation among specimens. Accurate 3D reconstruction is essential because taphonomic distortion can severely bias shape analyses.

Finite Element Analysis (FEA): Engineering-derived simulation technique that models the distribution of stress and strain in a structure under applied loads. In paleontology, FEA has been used to investigate cranial biomechanics, feeding function, and locomotor adaptations across a wide range of taxa.

Computational Fluid Dynamics (CFD): Simulates the flow of fluids around or through 3D models to test functional and ecological hypotheses. For example, CFD has been applied to Cambrian echinoderms and ammonoid shells to investigate feeding currents and swimming hydrodynamics.

Multi-Body Dynamic Analysis (MDA): Models the interactions of multiple rigid or deformable bodies connected by joints to simulate movement. MDA has been used to estimate bite forces in Tyrannosaurus rex and neck function in Allosaurus.

3D Printing (Rapid Prototyping): Digital models can be printed as physical objects for research, education, and museum exhibits. This is particularly valuable for producing scaled replicas, creating composite skeletons from fragmentary material, and making rare specimens accessible to broader audiences.

The digital nature of 3D reconstructions has catalyzed a culture of open data sharing in paleontology. MorphoSource (morphosource.org), launched at Duke University in 2013, is the largest web archive for 3D museum specimen data, hosting CT scan datasets and 3D mesh files that can be downloaded for research. Other repositories include Dryad (datadryad.org), Figshare (figshare.com), and Zenodo (zenodo.org). Platforms such as Sketchfab enable interactive online viewing and even augmented/virtual reality experiences with 3D fossil models. The open-access publication and deposit of both raw scan data and edited 3D models has become a best-practice standard, as advocated by Davies et al. (2017, Proceedings of the Royal Society B, 284: 20170194).

Despite its transformative power, 3D reconstruction in paleontology faces several important challenges. Each step in the restoration process introduces a degree of interpretation and subjectivity, and the cumulative effect can be significant, particularly for heavily deformed or fragmentary specimens. Manual segmentation remains extremely time-consuming—a single vertebrate skull can require weeks or months of work. Reproducibility is a concern because different researchers may produce different restorations from the same raw data. DeVries et al. (2022) explicitly addressed this by developing the armature-based Blender workflow that documents every manipulation, but widespread adoption of such rigorous documentation standards is still in progress.

For retrodeformation specifically, the assumption of bilateral symmetry may not always hold due to functional lateralization, fluctuating asymmetry, or symmetric deformation (e.g., uniform dorsoventral compression), which landmark-based methods cannot readily detect or correct. Fossil taxa known from single, incomplete specimens—a common situation—pose particular difficulties because reference specimens for automated restoration are unavailable.

Recent years have seen increasing application of deep learning and machine learning to accelerate aspects of the 3D reconstruction pipeline. Deep neural networks have been trained to perform automated CT segmentation of fossil specimens, dramatically reducing the time required to extract 3D models from raw scan data. For example, a 2022 study published in Frontiers in Earth Science demonstrated the application of deep learning to CT segmentation of dinosaur fossils, achieving results comparable to expert manual segmentation in a fraction of the time. AI-based approaches are also being developed for automated fossil identification, classification, and morphological analysis. Although these methods are still in early stages of adoption and require extensive training datasets, they represent a significant frontier for improving the speed, scalability, and consistency of 3D reconstruction workflows.

3D reconstruction has fundamentally transformed paleontological practice. It enables researchers to extract morphological information from fossils that would be inaccessible through traditional preparation methods, to perform rigorous quantitative analyses of form and function, and to share data openly with the global scientific community. The term 'virtual paleontology'—defined by Sutton et al. (2014, Techniques for Virtual Palaeontology, Wiley) as the study of fossils through three-dimensional digital visualizations—has become a recognized subdiscipline. The convergence of improved scanning technology, open-source software, powerful computational methods, AI automation, and online data repositories ensures that 3D reconstruction will continue to be a central pillar of paleontological research for the foreseeable future.