FEXHIS dataset for automatic segmentation of cortical bone microstructure

Summary This dataset comprises light microscopy images (objective: 10x/0.45) of human bone at three anatomical locations: Lateral aspect of the proximal part of the femoral diaphysis Superior segment of the femoral neck Inferior segment of the femoral neck Additionally, the dataset includes annotations from manual segmentations for the following classes: Haversian canal Osteocyte lacuna Cement line Background Further, results from an automatic segmentation algorithm for cortical bone microstructure are available. Sample Information The 94 femora presented in this dataset originate from 47 donors (25 females, 19 males and 3 unknown) aged 57 to 96 years at time of death. For each sample, information about the laterality (left/right bone), age and sex of the donor is available. All specimens were obtained by the Division of Anatomy of the Medical University of Vienna, Austria, with the informed consent of the donors. The file SampleList.csv contains the following information. Donor Donor ID used in the present project Internal ID Internal donor ID, which allows the data in this repository to be linked to other projects using the same femora. For further information, please refer to the projects listed under related works. Age (year) Donor age in years at time of death Sex (-) Donor sex, F: female, M: male Missing values are denoted with NA. Methods A detailed description of the methods related to the presented data can be found in the associated publication:Simon, Mathieu, et al. "Automatic segmentation of cortical bone microstructure: Application and analysis of three proximal femur sites." Bone 193 (2025): 117404. The code for the automatic segmentation algorithm is publicly available under https://github.com/artorg-unibe-ch/FEXHIP-Histology. Data Light Microscopy Images Images were acquired with a pixel size of about 1 μm using a light microscope (AxioImager M2 with Axiocam MRc, Carl Zeiss AG, Germany) with a 10×/0.45 objective and the ZEN 2.6 Pro software. For each donor and anatomical location, the raw image and/or an overview and zoom image in .jpg format are available. Manual Segmentation 25 random regions of interests (ROIs) of 0.5 mm side length representing histological features of interest were selected from right and left femora of the three donors with unknown sex and age and were segmented by 12 operators. Each operator received identical instructions to segment four classes (Haversian canal, osteocyte lacuna, cement line, and background) using Computer Vision Annotation Tool (CVAT). Three ROIs were segmented per operator and one of these three was common to every operator, allowing inter-operator variability assessment. Automatic Segmentation This folder contains the ROIs selected for model training and automated segmentation results associated with the publication Simon, Mathieu, et al. "Automatic segmentation of cortical bone microstructure: Application and analysis of three proximal femur sites." Bone 193 (2025): 117404. The associated code for the automatic segmentation algorithm is publicly available under https://github.com/artorg-unibe-ch/FEXHIP-Histology. Data Access Access to this repository is restricted for data protection and privacy reasons. To request access to this dataset, please submit a request through Zenodo and provide a brief description of the intended use of the data. Access will be granted only after completion and signing of a Data Transfer and Use Agreement. Contact Philippe Zysset ([e-mail]) for further questions. Related Works Associated Publication Simon, Mathieu, et al. "Automatic segmentation of cortical bone microstructure: Application and analysis of three proximal femur sites." Bone 193 (2025): 117404 GitHub Repository Automatic Segmentation Algorithm https://github.com/artorg-unibe-ch/FEXHIP-Histology Other Projects Using the Same Femur Samples Bracher, S., Kochetkova, T., Gerber, G.& Zysset, P. (2026). Tensile properties of bone collagen in the human proximal femur (FEMCOL) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21804708 Voumard, B., Dudle, A., Gugler, Y., Gerber, G.& Zysset, P. (2026). Multiscale imaging and mechanical testing dataset of human proximal femora (FEMHALS) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21494458