Since the early 2000s, the expansion of digital anatomy tools has been aided by advances in computational power and accessibility of medical imaging such as Computed Tomography (CT). The greater accessibility to digital imaging of fossil material has allowed the reconstructions of inner cranial cavities (endocasts), sinuses cavities, and dental reconstructions of the enamel-dentine junctions (EDJ) of fossilized teeth. Despite great accessibility, the segmentation processes used to generate digital reconstructions of inner cavities remain time-consuming and require specific expertise in computer analysis, anatomy, digital imaging.
Profico et al. (2018) provide two fully-automated digital methods to minimize these time-consuming digital segmentation tasks. Both of these methods rely on point-of-views (POVs) to delineate a region-of-interest (ROI). In the CA-LSE method, POVs were located outside a ROI and all areas beyond are subtracted from the final reconstruction. In contrast, the AST-3D method relies on a ROI defined by POVs placed inside a cavity and all external areas, subtracted from the final reconstruction. While both methods are similar and can be used to generate reconstructions of the inner cavities, each method has slightly different benefits. Profico and colleagues conducted a comparison of both methods to determine strengths and weaknesses of each approach. While both of these methods are available through the Cran R network, two different R packages were tested: Morpho and Arothron.
Results indicated that in the Morpho package, CA-LSE had no restrictions on where POVs could be placed, but using AST-3D method in Morpho, POVs had to be manually placed inside the internal cavity for successful reconstruction. In the Arothron package, CA-LSE method allowed fully-automated placement of POVs outside the ROI surface, however, the AST-3D method a ROI must be defined by manually placed POVs within the inner cavity. In general, accuracy of the AST-3D and CA-LSE methods were determined by each method, with AST-3D more reliable generating reconstructions of inner cavities (such as endocasts), while the CA-LSE was more suited to reconstructions of outer structures (such as skulls).
Although, automatic approaches offer time-efficiency and often allow larger sample sizes to be more quickly processed, many fossilized skulls are highly fragmentary and automated methods remain limited when fossilized remains are partially or entirely matrix-filled with anatomical and digital expertise still requiring manual segmentation. In these complex scenarios, further fine-tuning of automated methods would be invaluable with inclusion of fully-automated, semi-automatic and manual options.