Histological bone research, which involves measuring the characteristics of a bone (in this paper, we consider exclusively the boundary perimeter (bone perimeter, B.Pm), cross-sectional area (bone area, B.Ar), width (cortical width, Ct.Wi)), has been greatly enhanced by advanced medical imaging like the computed tomography (CT) scans, which has made it possible to get cross-sectional images (or slices) of different areas of a bone. The long bone, which is the major focus of our research, is divided into three parts according to its structure: diaphysis, metaphysis, epiphysis. Efficiently measuring the bone metrics mentioned above relies on being able to differentiate these areas from one another, since the different regions require unique approaches. For example, the diaphyseal region consists completely of the cortical bone and therefore obtaining the bone metrics is not perplexing. On the other hand, when it comes to the area of the metaphysis, and especially epiphysis, it becomes more difficult to distinguish the cortical bone from the spongy bone, and so the task of extracting the bone metrics is hampered. In this paper, we propose a fully automated solution for histological research of diaphysis. For this we designed and trained a convolutional neural network-based model to classify the cross-sectional images. Furthermore, we created computational methods to extract the bone measurements and applied them to bones classified as "diaphysis".
Article ID: 2021S22
Publisher: Canadian Artificial Intelligence Association