Bone is a dynamic tissue that changes throughout life. This process is governed by osteocytes that exist in a lacuno-canalicular network (LCN), but is altered by several factors, including exercise, age, nutrition, and substance use. Artificial intelligence brought several enhancements to image segmentation for medical imaging. However, it has not been applied to study the LCN in human bone. This thesis implements novel deep learning methods on Synchrotron Radiation micro-Computed Tomography (SRµCT) datasets of human rib cortical bone microstructure to characterize osteoporosis-related features.

Ninety-seven human left sixth rib specimens (male:​ n = 60, female n = 37) were excised from cadavers with informed consent. The specimens were divided into age categories defined by decade. A 50-slice subset from six samples was segmented to train the U-Net++ deep learning model. It was compared to traditional and manual segmentation methods. Deep learning performed comparably to the traditional method, although it was more time-efficient. A follow-up model with the MA-Net architecture more accurately segmented the data. Comparing segmented microstructural parameters with opioid use, sex, and age revealed age as the most significant predictor of deteriorating bone health. The results did not provide strong evidence of drug-induced impacts on bone health as originally predicted, however, there are some indications hinting at a link between opioid use and bone health. A follow-up study implementing a rabbit model is underway to eliminate confounding factors present in a human population. However, this project successfully created a novel segmentation algorithm that performed more efficiently in SRµCT data segmentation.