Automatic Segmentation of Osteonal Microstructure in Human Cortical Bone Using Deep Learning: A Proof of Concept.

Area of Science:

• Forensic Anthropology
• Biological Anthropology
• Computer Vision

Background:

• Cortical bone microstructure analysis is vital for age-at-death estimation and animal-human differentiation in anthropology.
• Osteon frequency and metrics are key features, but current assessment is manual, time-consuming, and requires specialized training.

Purpose of the Study:

• To investigate the feasibility of using deep learning for the automatic analysis of human bone microstructure images.
• To develop a computer vision model capable of semantic segmentation of osteonal structures.

Main Methods:

• A U-Net deep learning architecture was employed for semantic segmentation of bone microstructure images.
• Images were classified into three categories: intact osteons, fragmentary osteons, and background.
• Data augmentation techniques were utilized to prevent model overfitting.

Main Results:

• The model achieved a Dice coefficient of 0.73 for intact osteons, 0.38 for fragmented osteons, and 0.81 for background.
• The average Dice coefficient across all classes was 0.64.
• A binary classification of osteon-background yielded a Dice coefficient of 0.82.

Conclusions:

• This study presents the first proof of concept for using deep learning and computer vision to differentiate intact and fragmentary osteons in human cortical bone.
• The automated approach shows potential to significantly enhance and broaden the application of histomorphological assessments in anthropology.
• Further model refinement and testing on larger datasets are recommended for future development.