下载链接：

http://datadryad.org/dataset/doi%253A10.5061%252Fdryad.b2rbnzsq4

下载链接

链接失效反馈

官方服务：

资源简介：

Computed tomography (CT) enables rapid imaging of large-scale studies of bone, but those datasets typically require manual segmentation, which is time-consuming and prone to error. Convolutional neural networks (CNNs) offer an automated solution, achieving superior performance on image data. Here, we used CNNs to train segmentation models from scratch on 2D and 3D patches from micro-CT scans of otter long bones. These new models, collectively called BONe (Bone One-shot Network), aimed to be fast and accurate, and we expected enhanced results from 3D training due to better spatial context. Our results showed that 3D training required substantially more memory. Contrary to expectations, 2D models performed slightly better than 3D models in labeling details such as thin trabecular bone. Although lacking in some detail, 3D models appeared to generalize better and predict smoother internal surfaces than 2D models. However, the massive computational costs of 3D models limit their scalability and practicality, leading us to recommend 2D models for bone segmentation. BONe models showed potential for broader applications with variation in performance across species and scan quality. Notably, BONe models demonstrated promising results on skull segmentation, suggesting their potential utility beyond long bones with further refinement and fine-tuning. Methods Materials Limb bones from the North American river otter (Lontra canadensis) were borrowed from four museums — OMNH (SNM): Sam Noble Oklahoma Museum of Natural History (Norman Oklahoma); UAM: University of Alaska Museum of the North (Fairbanks, Alaska); and UF: Florida Museum of Natural History (Gainesville, Florida); UWBM (BM): Burke Museum of Natural History and Culture (Seattle, Washington). In total, the sample consisted of 38 elements (humerus, radius, ulna, femur, tibia, and fibula) from nine individuals. Specimen kV µA Filter Res. (µm) Provenance Sex Side Element Group OMNH 44262 160 312 Copper 49.99 Tennessee F R Humerus Fitting L Radius Fitting L Ulna Fitting OMNH 53994 160 312 Copper 49.99 Tennessee M L Femur Fitting L Fibula Fitting L Humerus Testing R Radius Testing L Tibia Fitting R Ulna Testing UAM 24789 160 312 Copper 49.99 Alaska M R Femur Testing R Fibula Testing R Tibia Testing UAM 67696 90 556 None 49.99 Alaska F L Femur Fitting L Humerus Fitting 160 312 Copper 49.99 L Fibula Fitting 160 312 Copper L Radius Fitting 160 312 Copper L Tibia Fitting 160 312 Copper L Ulna Fitting UF 23593 79 633 None 49.99 Florida M L Femur Fitting L Humerus Fitting UF 24550 79 633 None 49.99 Florida F L Femur Fitting L Humerus Fitting 160 312 Copper 49.99 L Fibula Fitting L Radius Fitting L Tibia Fitting L Ulna Fitting UF 31151 160 312 Copper 49.99 Florida M R Humerus Fitting R Radius Fitting R Ulna Fitting UWBM 78743 160 312 Copper 49.99 Washington F R Femur Fitting L Fibula Fitting L Tibia Fitting UWBM 81969 160 312 Copper 49.99 Washington M R Femur Testing L Fibula Testing L Humerus Testing R Radius Testing L Tibia Testing R Ulna Testing Imaging Bones were scanned at the University of Arkansas MICRO lab using a Nikon XTH 225 ST. Scans were performed in sets of 2-4 bones. Scan settings are included in the preceding table. Of the 12 sets, eight were used for training/validation and four were reserved for blind testing. Reference segmentation Reference segmentations were prepared using Avizo 3D 2024.1 (Thermo Fisher Scientific, Waltham, MA, USA). The 16-bit grayscale contrast of each volume was improved using the “Sigmoid Intensity Remapping” operation set to 3D mode. Bone tissue was initially segmented using the “Auto Thresholding” module (default settings) and manually inspected for mislabeling. Trabecular bone that was missed by auto thresholding was added to the bone tissue segmentation using the “Top Hat” filter. Medullary spaces and large vascular canals (hereafter referred to as “pores”) were isolated using the 3D ray tracing “Compute Ambient Occlusion” operation on the bone segmentation. Trabecular bone was segmented following the “shrink-wrap” approach of Herbst et al. (2021). Briefly, this approach creates a masking volume that circumscribes major pores. Bone tissue within this mask volume is classified as trabecular bone, whereas bone tissue outside the volume is classified as cortical bone. Image preparation culminated in two sets of reference data for each volume: the first with labels for bone tissue and pores (BP); and the second with labels for cortical bone, trabecular bone, and pores (CTP). Fitting the deep learning models Several deep learning models were fitted using Avizo 3D with TensorFlow (Google, Mountain View, California) as the backend. The architecture of each model was pre-built consisting of a U-Net convolutional neural network with a ResNet-18 encoder. We fitted models separately on 2D (e.g., single slice) and 3D (e.g., volumetric) data. We also fitted models to segment bone and medullary pores separately from those to segment cortical bone, trabecular bone, and medullary pores. Each model was pre-trained on UWBM 78743 and subsequently fine-tuned on a larger composite dataset containing several specimens. Note: a pre-trained model represents an intermediate stage in fitting that we share for data transparency. The fine-tuned model is finalized and ready for general use. List of 2D models Abbreviations: B = bone; P = medullary pores; C = cortical bone; T = trabecular bone Bold = finalized model Name Training stage Segmentation goal BP-2D-02 Pre-train Bone tissue and medullary pores BP-2D-02a Fine-tune Bone tissue and medullary pores CTP-2D-02 Pre-train Cortical bone, trabecular bone, medullary pores CTP-2D-02a Fine-tune Cortical bone, trabecular bone, medullary pores List of 3D models Abbreviations: B = bone; P = medullary pores; C = cortical bone; T = trabecular bone Bold = finalized model Name Training stage Segmentation goal BP-3D-02 Pre-train Bone tissue and medullary pores BP-3D-02a Fine-tune Bone tissue and medullary pores CTP-3D-02 Pre-train Cortical bone, trabecular bone, medullary pores CTP-3D-02a Fine-tune Cortical bone, trabecular bone, medullary pores

应用场景：

Dataset: Segmentation of cortical bone, trabecular bone, and medullary pores from micro-CT images using 2D and 3D deep learning models