• AI-driven segmentation tools for CT scans have rapidly evolved, enabling faster 3D reconstruction and automated segmentation of similar objects 1.

• These methods are highly effective in fields with repetitive sample types (e.g., biomedicine, geoscience), but less applicable to unique specimens in natural history collections.

• Most AI segmentation tools are embedded in proprietary software, limiting accessibility and compatibility with diverse file formats.

• We wanted to try to apply a partially automated, algorithm-based tool to streamline segmentation of complex and difficult to scan fossils.

• Problem: how to segment fossils with poor contrast between different structures? Each specimen has unique challenges and many segments need to be separated.

• Our solution: Biomedisa 2 smart interpolation tool 3. Using a random walk approach creates 3D volumes based on the data starting from segmentation of only a relative small number of slices, cutting user time. It is free, web based and works with non-proprietary software formats.

1. He Y et al. 2024 Organismal Biology , obae036. https://doi.org/10.1093/iob/obae036 

2. Lösel PD et al. 2020 Nature Communication 11, 5577 https://doi.og/10.1038/s41467-020-19303-w 

3. Lösel P, Heuveline V. 2016 Proceedings SPIE 9784, Medical Imaging 2016: Image Processing, 97842L. https://doi.org/10.1117/12.2216202 

4. Schnetz, L. et al. IN REVIEW. Palaeontology. 

5. Elliott DK, Petriello MA. 2011 Journal of Vertebrate Paleontology 31, 518–530.  https://doi.org/10.1080/02724634.2011.557113 

We thank Mark Wilson, John Bruner and Alison Murray (UALVP – University of Alberta, Edmonton, Canada) and the Field Museum, Chicago, USA (FMNH) for help in accessing the specimens in their care. We thank April Neander (University of Chicago) and Liz Martin-Silverstone (University of Bristol) for help with CT scanning of the specimens. 

This work is funded by the Leverhulme Trust RPG-2021-271 “Feeding without jaws: innovations in early vertebrates”. 

Background image is a Silurian-Devonian formation from Spitzbergen, Norway from Overduin et al. 2014. Geological Society London Special Publications, 388(1).

• • Over 20 fossil jawless armoured fish specimens from the Heterostraci group were CT scanned to extract 3D oral morphology.

  • We applied Biomedisa smart interpolation initially to aid the segmentation of  two specimens in rock matrix: the oral region of Athenaegis 4 (UALVP39923-Early Silurian) and the whole body of Poraspis 5 (FMNH-PF14474-Late Silurian/Early Devonian).

  • We tested Biomedisa smart interpolation: starting from manually segmented slices (1 in 5 for Athenaegis and 1 in 15-20 for Poraspis), it calculates the probability of each voxel belonging to a label and assigns it accordingly.

  • Athenaegis was CT scanned at the University of Bristol UK using a Nikon Metrology XTH 225ST with the following parameters: voltage: 115 kV, current: 94 μA, 3141 projections, 4 frames per projection, exposure: 1000 ms, no filtration, voxel resolution: 10.922 μm

  • Poraspis was CT scanned at University of Chicago USA using a Phoenix v|tome|x s 240 with the following parameters: voltage: 120 kV, current: 210 μA, 1400 projections, 3 frames per projection, exposure: 333 ms, 0.1 mm Cu filter, voxel resolution: 22.3950 μm

  • Manual segmentation was performed in Avizo 2021.2 using the brush tool after selecting an appropriate grey scale value for masking.

  • The label file was exported and uploaded to biomedisa.info along with the TIFF stack as a single file for the interpolation. All axes were selected in the settings for the algorithm to run correctly.

• • Biomedisa smart interpolation algorithm successfully cut processing time for segmentation of the specimens. 

  • It divided the volume into the desired segments while following the density values of the specimen extracting the fossil from the matrix.

  • The Biomedisa materials were imported in Avizo again to be cleaned and refined.

  • This process was applied to other 8 specimens with different levels of complexity.

  • Biomedisa smart interpolation tool can be used to quickly segment different objects even from poor quality scans. The web interface and easy to follow instructions do not require specific training. A new add-on to 3D Slicer makes it easy to sue also in free segmentation software and for medical-based research.