Comparing semi-automatic and deep learning-driven segmentation of an archaeological human mandible from Isola Sacra (Fiumicino, Italy, 1st - 3rd century CE)

The use of lab-based microtomography to study mineralized human and animal tissues, qualitatively and quantitatively, requires precise and consistent segmentation of the three-dimensional digital imaging. Particularly, the analysis of dental mineralized tissues is a critical step for pushing the boundaries of our knowledge about human evolutive trajectories and lifestyles. As a matter of fact, manual segmentation is a complex, time-consuming and difficult-to-automatize process. Traditionally, the segmentation of mineralized tissues relied on semi automatic methods (e.g., histogram-based thresholding), which often struggle to accurately delineate the structures with similar radiopacity, as dentine and surrounding bone. This study explores the efficacy of deep learning convolutional neural networks (CNNs) as a new approach for the automatic segmentation of the phases of mineralised dental tissues and the as-sociated bone matrix. Results indicate the effectiveness of Artificial Intelligence-driven technologies for the automatic segmentation of bone and tooth samples from archaeological finds, particularly when compared to traditional semi-automatic segmentation.