Automated threshold-independent cortex segmentation by 3D-texture analysis of HR-pQCT scans

Automatic segmentation of trabecular and cortical compartments in HR-pQCT images using an embedding-predicting U-Net and morphological post-processing

High-resolution peripheral quantitative computed tomography (HR-pQCT) is an emerging in vivo imaging modality for quantification of bone microarchitecture. However, extraction of quantitative microarchitectural parameters from HR-pQCT images requires an accurate segmentation of the image. The current standard protocol using semi-automated contouring for HR-pQCT image segmentation is laborious, introduces inter-operator biases into research data, and poses a barrier to streamlined clinical implementation. In this work, we propose and validate a fully automated algorithm for segmentation of HR-pQCT radius and tibia images. A multi-slice 2D U-Net produces initial segmentation predictions, which are post-processed via a sequence of traditional morphological image filters. The U-Net was trained on a large dataset containing 1822 images from 896 unique participants. Predicted segmentations were compared to reference segmentations on a disjoint dataset containing 386 images from 190 unique participants, and 156 pairs of repeated images were used to compare the precision of the novel and current protocols. The agreement of morphological parameters obtained using the predicted segmentation relative to the reference standard was excellent (R² between 0.938 and > 0.999). Precision was significantly improved for several outputs, most notably cortical porosity. This novel and robust algorithm for automated segmentation will increase the feasibility of using HR-pQCT in research and clinical settings.

Automated segmentation of vertebral cortex with 3D U-Net-based deep convolutional neural network

Objectives: We developed a 3D U-Net-based deep convolutional neural network for the automatic segmentation of the vertebral cortex. The purpose of this study was to evaluate the accuracy of the 3D U-Net deep learning model.

Methods: In this study, a fully automated vertebral cortical segmentation method with 3D U-Net was developed, and ten-fold cross-validation was employed. Through data augmentation, we obtained 1,672 3D images of chest CT scans. Segmentation was performed using a conventional image processing method and manually corrected by a senior radiologist to create the gold standard. To compare the segmentation performance, 3D U-Net, Res U-Net, Ki U-Net, and Seg Net were used to segment the vertebral cortex in CT images. The segmentation performance of 3D U-Net and the other three deep learning algorithms was evaluated using DSC, mIoU, MPA, and FPS.

Results: The DSC, mIoU, and MPA of 3D U-Net are better than the other three strategies, reaching 0.71 ± 0.03, 0.74 ± 0.08, and 0.83 ± 0.02, respectively, indicating promising automated segmentation results. The FPS is slightly lower than that of Seg Net (23.09 ± 1.26 vs. 30.42 ± 3.57).

Conclusion: Cortical bone can be effectively segmented based on 3D U-net.

Deep learning methods allow fully automated segmentation of metacarpal bones to quantify volumetric bone mineral density

Arthritis patients develop hand bone loss, which leads to destruction and functional impairment of the affected joints. High resolution peripheral quantitative computed tomography (HR-pQCT) allows the quantification of volumetric bone mineral density (vBMD) and bone microstructure in vivo with an isotropic voxel size of 82 micrometres. However, image-processing to obtain bone characteristics is a time-consuming process as it requires semi-automatic segmentation of the bone. In this work, a fully automatic vBMD measurement pipeline for the metacarpal (MC) bone using deep learning methods is introduced. Based on a dataset of HR-pQCT volumes with MC measurements for 541 patients with arthritis, a segmentation network is trained. The best network achieves an intersection over union as high as 0.94 and a Dice similarity coefficient of 0.97 while taking only 33 s to process a whole patient yielding a speedup between 2.5 and 4.0 for the whole workflow. Strong correlation between the vBMD measurements of the expert and of the automatic pipeline are achieved for the average bone density with 0.999 (Pearson) and 0.996 (Spearman's rank) with [Formula: see text] for all correlations. A qualitative assessment of the network predictions and the manual annotations yields a 65.9% probability that the expert favors the network predictions. Further, the steps to integrate the pipeline into the clinical workflow are shown. In order to make these workflow improvements available to others, we openly share the code of this work.

A Comparison of Peripheral Imaging Technologies for Bone and Muscle Quantification: A Review of Segmentation Techniques

Musculoskeletal science has developed many overlapping branches, necessitating specialists from 1 area of focus to often require the expertise in others. In terms of imaging, this means obtaining a comprehensive illustration of bone, muscle, and fat tissues. There is currently a lack of a reliable resource for end users to learn about these tissues' imaging and quantification techniques together. An improved understanding of these tissues has been an important progression toward better prediction of disease outcomes and better elucidation of their interaction with frailty, aging, and metabolic disorders. Over the last decade, there have been major advances into the image acquisition and segmentation of bone, muscle, and fat features using computed tomography (CT), magnetic resonance imaging (MRI), and peripheral modules of these systems. Dedicated peripheral quantitative musculoskeletal imaging systems have paved the way for mobile research units, lower cost clinical research facilities, and improved resolution per unit cost paid. The purpose of this review was to detail the segmentation techniques available for each of these peripheral CT and MRI modalities and to describe advances in segmentation methods as applied to study longitudinal changes and treatment-related dynamics. Although the peripheral CT units described herein have established feasible standardized protocols that users have adopted globally, there remain challenges in standardizing MRI protocols for bone and muscle imaging.

An Algorithm for Automated Separation of Trabecular Bone From Variably Thick Cortices in High-Resolution Computed Tomography Data

OBJECTIVE

Structural measurements after separation of cortical from trabecular bone are of interest to a wide variety of communities but are difficult to obtain because of the lack of accurate automated techniques.

METHODS

We present a structure-based algorithm for separating cortical from trabecular bone in binarized images. Using the thickness of the cortex as a seed value, bone connected to the cortex within a spatially local threshold value is identified and separated from the remaining bone. The algorithm was tested on seven biological data sets from four species imaged using micro-computed tomography (μ-CT) and high-resolution peripheral quantitative computed tomography (HR-pQCT). Area and local thickness measurements were compared to images segmented manually.

RESULTS

The algorithm was approximately 11 times faster than manual measurements and the median error in cortical area was -4.47 ± 4.15%. The median error in cortical thickness was approximately 0.5 voxels for μ-CT data and less than 0.05 voxels for HR-pQCT images resulting in an overall difference of -28.1 ± 71.1 μm.

CONCLUSION

A simple and readily implementable methodology has been developed that is repeatable, efficient, and requires few user inputs, providing an unbiased means of separating cortical from trabecular bone.

SIGNIFICANCE

Automating the segmentation of variably thick cortices will allow for the evaluation of large data sets in a time-efficient manner and allow for full-field analyses that have been previously limited to small regions of interest. The MATLAB code can be downloaded from https://github.com/TBL-UIUC/downloads.git.

Contribution of high resolution peripheral quantitative CT to the management of bone and joint diseases

Many imaging modalities have been described to diagnose and monitor osteoporosis (OP), osteoarthritis and inflammatory rheumatic diseases. Over the last ten years, High Resolution peripheral Quantitative Computerized Tomography (HR-pQCT) was shown to be a precise and non invasive technique to study bone and joint diseases in clinical research. It allows the study of both cortical and trabecular bone microarchitecture at the distal tibia and radius, and further applications have been developed such as the study of mechanical properties by the finite element analysis. Thus, in case-control and cross-sectional studies, microarchitecture parameters discriminated fractured individuals independently of areal BMD. Also, microstructure parameters can predict incident fracture in postmenopausal women. In metabolic diseases associated with bone fragility, HR-pQCT may also be used to explore bone changes. In joint disease studies, HR-pQCT was a remarkable tool to assess bone erosion and joint space narrowing at the hand. This article gives an overview of this imaging technique.

Microarchitecture and Peripheral BMD are Impaired in Postmenopausal White Women With Fracture Independently of Total Hip T-Score: An International Multicenter Study

Because single-center studies have reported conflicting associations between microarchitecture and fracture prevalence, we included high-resolution peripheral quantitative computed tomography (HR-pQCT) data from five centers worldwide into a large multicenter analysis of postmenopausal women with and without fracture. Volumetric BMD (vBMD) and microarchitecture were assessed at the distal radius and tibia in 1379 white postmenopausal women (age 67 ± 8 years); 470 (34%) had at least one fracture including 349 with a major fragility fracture. Age, height, weight, and total hip T-score differed across centers and were employed as covariates in analyses. Women with fracture had higher BMI, were older, and had lower total hip T-score, but lumbar spine T-score was similar between groups. At the radius, total and trabecular vBMD and cortical thickness were significantly lower in fractured women in three out of five centers, and trabecular number in two centers. Similar results were found at the tibia. When data from five centers were combined, however, women with fracture had significantly lower total, trabecular, and cortical vBMD (2% to 7%), lower trabecular number (4% to 5%), and thinner cortices (5% to 6%) than women without fracture after adjustment for covariates. Results were similar at the radius and tibia. Similar results were observed with analysis restricted to major fragility fracture, vertebral and hip fractures, and peripheral fracture (at the radius). When focusing on osteopenic women, each SD decrease of total and trabecular vBMD was associated with a significantly increased risk of major fragility fracture (OR = 1.55 to 1.88, p < 0.01) after adjustment for covariates. Moreover, trabecular architecture modestly improved fracture discrimination beyond peripheral total vBMD. In conclusion, we observed differences by center in the magnitude of fracture/nonfracture differences at both the distal radius and tibia. However, when data were pooled across centers and the sample size increased, we observed significant and consistent deficits in vBMD and microarchitecture independent of total hip T-score in all postmenopausal white women with fracture and in the subgroup of osteopenic women, compared to women who never had a fracture. © 2016 American Society for Bone and Mineral Research.

Automated cortical bone segmentation for multirow-detector CT imaging with validation and application to human studies

PURPOSE

Cortical bone supports and protects human skeletal functions and plays an important role in determining bone strength and fracture risk. Cortical bone segmentation at a peripheral site using multirow-detector CT (MD-CT) imaging is useful for in vivo assessment of bone strength and fracture risk. Major challenges for the task emerge from limited spatial resolution, low signal-to-noise ratio, presence of cortical pores, and structural complexity over the transition between trabecular and cortical bones. An automated algorithm for cortical bone segmentation at the distal tibia from in vivo MD-CT imaging is presented and its performance and application are examined.

METHODS

The algorithm is completed in two major steps-(1) bone filling, alignment, and region-of-interest computation and (2) segmentation of cortical bone. After the first step, the following sequence of tasks is performed to accomplish cortical bone segmentation-(1) detection of marrow space and possible pores, (2) computation of cortical bone thickness, detection of recession points, and confirmation and filling of true pores, and (3) detection of endosteal boundary and delineation of cortical bone. Effective generalizations of several digital topologic and geometric techniques are introduced and a fully automated algorithm is presented for cortical bone segmentation.

RESULTS

An accuracy of 95.1% in terms of volume of agreement with manual outlining of cortical bone was observed in human MD-CT scans, while an accuracy of 88.5% was achieved when compared with manual outlining on postregistered high resolution micro-CT imaging. An intraclass correlation coefficient of 0.98 was obtained in cadaveric repeat scans. A pilot study was conducted to describe gender differences in cortical bone properties. This study involved 51 female and 46 male participants (age: 19-20 yr) from the Iowa Bone Development Study. Results from this pilot study suggest that, on average after adjustment for height and weight differences, males have thicker cortex (mean difference 0.33 mm and effect size 0.92 at the anterior region) with lower bone mineral density (mean difference -28.73 mg/cm(3) and effect size 1.35 at the posterior region) as compared to females.

CONCLUSIONS

The algorithm presented is suitable for fully automated segmentation of cortical bone in MD-CT imaging of the distal tibia with high accuracy and reproducibility. Analysis of data from a pilot study demonstrated that the cortical bone indices allow quantification of gender differences in cortical bone from MD-CT imaging. Application to larger population groups, including those with compromised bone, is needed.

Inverse Correlation at the Hip Between Areal Bone Mineral Density Measured by Dual-Energy X-ray Absorptiometry and Cortical Volumetric Bone Mineral Density Measured by Quantitative Computed Tomography

Quantitative computed tomography (QCT) is considered to measure true volumetric bone mineral density (vBMD; mg/cm3) and enables differentiation between cortical and trabecular bone. We aimed to determine the value of QCT by correlating areal BMD (aBMD) by dual-energy X-ray absorptiometry (DXA) with vBMD when using a fixed threshold to delineate cortical from trabecular bone. In a cross-sectional study, 98 postmenopausal women had their hip scanned by DXA and by QCT. At the total hip and the trabecular bone compartment, aBMD correlated significantly with vBMD (r=0.74 and r=0.63; p<0.01, respectively). A significant inverse correlation was found between aBMD and cortical vBMD (r=-0.57; p<0.01). Total hip volume by QCT did not change with aBMD. However, increased aBMD was associated with a decreased trabecular bone volume (r=-0.36; p<0.01) and an increased cortical volume (r=0.69; p<0.01). Changing the threshold used to delineate cortical from trabecular bone from default 350 mg/cm3 to either 300 or 400 mg/cm3 did not affect integral vBMD (p=89) but had marked effects on estimated vBMD at the cortical (p<0.001) and trabecular compartments (p<0.001). Furthermore, increasing the threshold decreased cortical thickness (p<0.001), whereas the strength parameter in terms of buckling ratio increased (p<0.001). Our results show good agreement between aBMD and integral vBMD. However, using a fixed threshold to differentiate cortical from trabecular bone causes an apparent increase in cortical volume with a decrease in cortical density as aBMD increases. This may be caused by the classification of a larger part of the transition zone as cortical bone with increased aBMD.

A new algorithm to improve assessment of cortical bone geometry in pQCT

High-resolution peripheral quantitative computed tomography (HR-pQCT) is now considered the leading imaging modality in bone research. However, access to HR-pQCT is limited and image acquisition is mainly constrained only for the distal third of appendicular bones. Hence, the conventional pQCT is still commonly used despite inaccurate threshold-based segmentation of cortical bone that can compromise the assessment of whole bone strength. Therefore, this study addressed whether the use of an advanced image processing algorithm, called OBS, can enhance the cortical bone analysis in pQCT images and provide similar information to HR-pQCT when the same volumes of interest are analyzed. Using pQCT images of European Forearm Phantom (EFP), and pQCT and HR-pQCT images of the distal tibia from 15 cadavers, we compared the results from the OBS algorithm with those obtained from common pQCT analyses, HR-pQCT manual analysis (considered as a gold standard) and common HR-pQCT analysis dual threshold technique.We found that the use of OBS segmentation method for pQCT image analysis of EFP data did not result in any improvement but reached similar performance in cortical bone delineation as did HR-pQCT image analyses. The assessments of cortical cross-sectional bone area and thickness by OBS algorithm were overestimated by less than 4% while area moments of inertia were overestimated by ~5–10%, depending on reference HR-pQCT analysis method. In conclusion, this study showed that the OBS algorithm performed reasonably well and it offers a promising practical tool to enhance the assessment of cortical bone geometry in pQCT.

A New Algorithm for Cortical Bone Segmentation with Its Validation and Applications to In Vivo Imaging

Cortical bone supports and protects our skeletal functions and it plays an important in determining bone strength and fracture risks. Cortical bone segmentation is needed for quantitative analyses and the task is nontrivial for in vivo multi-row detector CT (MD-CT) imaging due to limited resolution and partial volume effects. An automated cortical bone segmentation algorithm for in vivo MD-CT imaging of distal tibia is presented. It utilizes larger contextual and topologic information of the bone using a modified fuzzy distance transform and connectivity analyses. An accuracy of 95.1% in terms of volume of agreement with true segmentations and a repeat MD-CT scan intra-class correlation of 98.2% were observed in a cadaveric study. An in vivo study involving 45 age-similar and height-matched pairs of male and female volunteers has shown that, on an average, male subjects have 16.3% thicker cortex and 4.7% increased porosity as compared to females.

High-resolution peripheral quantitative computed tomography for the assessment of bone strength and structure: a review by the Canadian Bone Strength Working Group

Bone structure is an integral determinant of bone strength. The availability of high resolution peripheral quantitative computed tomography (HR-pQCT) has made it possible to measure three-dimensional bone microarchitecture and volumetric bone mineral density in vivo, with accuracy previously unachievable and with relatively low-dose radiation. Recent studies using this novel imaging tool have increased our understanding of age-related changes and sex differences in bone microarchitecture, as well as the effect of different pharmacological therapies. One advantage of this novel tool is the use of finite element analysis modelling to non-invasively estimate bone strength and predict fractures using reconstructed three-dimensional images. In this paper, we describe the strengths and limitations of HR-pQCT and review the clinical studies using this tool.

Computational identification and quantification of trabecular microarchitecture classes by 3-D texture analysis-based clustering

High resolution peripheral quantitative computed tomography (HR-pQCT) permits the non-invasive assessment of cortical and trabecular bone density, geometry, and microarchitecture. Although researchers have developed various post-processing algorithms to quantify HR-pQCT image properties, few of these techniques capture image features beyond global structure-based metrics. While 3D-texture analysis is a key approach in computer vision, it has been utilized only infrequently in HR-pQCT research. Motivated by high isotropic spatial resolution and the information density provided by HR-pQCT scans, we have developed and evaluated a post-processing algorithm that quantifies microarchitecture characteristics via texture features in HR-pQCT scans. During a training phase in which clustering was applied to texture features extracted from each voxel of trabecular bone, three distinct clusters, or trabecular microarchitecture classes (TMACs) were identified. These TMACs represent trabecular bone regions with common texture characteristics. The TMACs were then used to automatically segment the voxels of new data into three regions corresponding to the trained cluster features. Regional trabecular bone texture was described by the histogram of relative trabecular bone volume covered by each cluster. We evaluated the intra-scanner and inter-scanner reproducibility by assessing the precision errors (PE), intra class correlation coefficients (ICC) and Dice coefficients (DC) of the method on 14 ultradistal radius samples scanned on two HR-pQCT systems. DC showed good reproducibility in intra-scanner set-up with a mean of 0.870±0.027 (no unit). Even in the inter-scanner set-up the ICC showed high reproducibility, ranging from 0.814 to 0.964. In a preliminary clinical test application, the TMAC histograms appear to be a good indicator, when differentiating between postmenopausal women with (n=18) and without (n=18) prevalent fragility fractures. In conclusion, we could demonstrate that 3D-texture analysis and feature clustering seems to be a promising new HR-pQCT post-processing tool with good reproducibility, even between two different scanners.

Multicenter precision of cortical and trabecular bone quality measures assessed by high-resolution peripheral quantitative computed tomography

High-resolution peripheral quantitative computed tomography (HR-pQCT) has recently been introduced as a clinical research tool for in vivo assessment of bone quality. The utility of this technology to address important skeletal health questions requires translation to standardized multicenter data pools. Our goal was to evaluate the feasibility of pooling data in multicenter HR-pQCT imaging trials. Reproducibility imaging experiments were performed using structure and composition-realistic phantoms constructed from cadaveric radii. Single-center precision was determined by repeat scanning over short-term (<72 hours), intermediate-term (3-5 months), and long-term intervals (28 months). Multicenter precision was determined by imaging the phantoms at nine different HR-pQCT centers. Least significant change (LSC) and root mean squared coefficient of variation (RMSCV) for each interval and across centers was calculated for bone density, geometry, microstructure, and biomechanical parameters. Single-center short-term RMSCVs were <1% for all parameters except cortical thickness (Ct.Th) (1.1%), spatial variability in cortical thickness (Ct.Th.SD) (2.6%), standard deviation of trabecular separation (Tb.Sp.SD) (1.8%), and porosity measures (6% to 8%). Intermediate-term RMSCVs were generally not statistically different from short-term values. Long-term variability was significantly greater for all density measures (0.7% to 2.0%; p < 0.05 versus short-term) and several structure measures: cortical thickness (Ct.Th) (3.4%; p < 0.01 versus short-term), cortical porosity (Ct.Po) (15.4%; p < 0.01 versus short-term), and trabecular thickness (Tb.Th) (2.2%; p < 0.01 versus short-term). Multicenter RMSCVs were also significantly higher than short-term values: 2% to 4% for density and micro-finite element analysis (µFE) measures (p < 0.0001), 2.6% to 5.3% for morphometric measures (p < 0.001), whereas Ct.Po was 16.2% (p < 0.001). In the absence of subject motion, multicenter precision errors for HR-pQCT parameters were generally less than 5%. Phantom-based multicenter precision was comparable to previously reported in in vivo single-center precision errors, although this was approximately two to five times worse than ex vivo short-term precision. The data generated from this study will contribute to the future design and validation of standardized procedures that are broadly translatable to multicenter study designs.