Amphibian

High-energy computed tomography of a stone block measuring 370 × 200 × 160 mm³ with a fragment of a skull roof that is only partially exposed on one side provides a unique gateway to previously concealed structures.

Fossils are generally very challenging objects for computed tomography: on the one hand, highly absorbing materials (for example sandstone) must be penetrated, and at the same time the delicate structures of mineralized bones with low material contrast must be depicted at high resolution and without artifacts. The solution to this challenge lies in modern high-energy imaging using a 9 MeV linear accelerator as the X-ray source.

The helical CT results already reveal, in the individual CT slices, the thin bone structures as well as the surrounding stone matrix. Even at this stage, the preparation provides initial insights into the position, extent, completeness, and dimensions of the bones inside, equipping paleontologists with essential prior knowledge for any subsequent manual preparation.

The CT dataset of the fossil is the starting point for downstream digital data processing—another core competency of the Development Center X-ray Technology at Fraunhofer IIS:

Segmentation—that is, the virtual extraction of the upper jaw bone from the rock—posed a particular challenge because there was only minimal contrast between bone and stone. Furthermore, spherical inclusions and the casting compound resembled the bone material so closely in CT imaging that conventional thresholding methods were not applicable. The solution employed was an in-house development by the Fraunhofer IIS Research Group for Knowledge-Based Image Processing and Visualization: ASVoxCraft. This software enables users to train a machine learning model for object segmentation via an interactive click procedure, without the need for laboriously created ground-truth data. In particular, the model can be improved incrementally to solve even difficult segmentation problems in large 3D CT datasets. Here, click annotations distinguishing between “amphibian” and “non-amphibian” enabled the bone to be digitally freed from the rock within a few hours of work.

The segmentation results can be processed into striking visualizations of the fossil bones or video animations, and they provide paleontologists with a dimensionally accurate digital reconstruction as a basis for scientific work and further analyses, for example to achieve precise species identification.