Individual component-based parameter-adaptive segmentation approach for improved segmentation of synchrotron µCT data of osteocyte lacunae in bone tissue

Bone tissue is highly complex and dynamic, capable of adapting to mechanical demands and repairing itself through remodeling processes. This remodeling results in a heterogeneous mineral distribution, with lower mineralization in younger bone regions and higher mineralization in older ones. Osteocytes - bone cells residing in small lacunae within the mineralized bone matrix - orchestrate this remodeling. Additionally, osteocytes actively modify their peri-lacunar mineralized tissue. These characteristics, combined with the high osteocyte density of several tens of thousands per mm³ , make the distribution, size, and shape of osteocyte lacunae highly relevant characteristics of bone tissue. To study osteocyte lacunar properties, synchrotron-based computed tomography (µCT) has become increasingly popular over the past decade due to its combination of high spatial resolution, sensitivity to mineral density variations, and rapid data acquisition. However, segmenting lacunae and quantifying their properties remains challenging. Osteocyte lacunae exhibit diverse shapes and sizes, and their surrounding mineral density can vary significantly between lacunae, even within the same tissue sample. Consequently, no global gray value threshold can provide an equally accurate segmentation across different tissue regions within the same sample. More advanced segmentation techniques, such as those based on top-hat transformations, require the definition of a structuring element whose size must be tailored to the feature size, in this case, the lacunae. In this study, we propose a novel approach to segmentation that adjusts the threshold value and the size of the structuring element for each lacuna individually. This method, referred to as the Kangaroo Segmentation Approach, involves an initial rough segmentation, followed by connected-component analysis and refinement steps applied to each component. The results of this Kangaroo Segmentation Approach are compared with conventional Otsu thresholding and thresholding methods based on top-hat transformations. Our findings demonstrate a significant improvement in segmentation accuracy with the proposed method.