Deep learning-based harmonization of trabecular bone microstructures between high- and low-resolution CT imaging.

BACKGROUND: Osteoporosis is a bone disease related to increased bone loss and fracture-risk. The variability in bone strength is partially explained by bone mineral density (BMD), and the remainder is contributed by bone microstructure. Recently, clinical CT has emerged as a viable option for in vivo bone microstructural imaging. Wide variations in spatial-resolution and other imaging features among different CT scanners add inconsistency to derived bone microstructural metrics, urging the need for harmonization of image data from different scanners. PURPOSE: This paper presents a new deep learning (DL) method for the harmonization of bone microstructural images derived from low- and high-resolution CT scanners and evaluates the method's performance at the levels of image data as well as derived microstructural metrics. METHODS: We generalized a three-dimensional (3D) version of GAN-CIRCLE that applies two generative adversarial networks (GANs) constrained by the identical, residual, and cycle learning ensemble (CIRCLE). Two GAN modules simultaneously learn to map low-resolution CT (LRCT) to high-resolution CT (HRCT) and vice versa. Twenty volunteers were recruited. LRCT and HRCT scans of the distal tibia of their left legs were acquired. Five-hundred pairs of LRCT and HRCT image blocks of 64 × 64 × 64 $64 \times 64 \times 64 $ voxels were sampled for each of the twelve volunteers and used for training in supervised as well as unsupervised setups. LRCT and HRCT images of the remaining eight volunteers were used for evaluation. LRCT blocks were sampled at 32 voxel intervals in each coordinate direction and predicted HRCT blocks were stitched to generate a predicted HRCT image. RESULTS: Mean ± standard deviation of structural similarity (SSIM) values between predicted and true HRCT using both 3DGAN-CIRCLE-based supervised (0.84 ± 0.03) and unsupervised (0.83 ± 0.04) methods were significantly (p < 0.001) higher than the mean SSIM value between LRCT and true HRCT (0.75 ± 0.03). All Tb measures derived from predicted HRCT by the supervised 3DGAN-CIRCLE showed higher agreement (CCC  ∈ $ \in $ [0.956 0.991]) with the reference values from true HRCT as compared to LRCT-derived values (CCC  ∈ $ \in $ [0.732 0.989]). For all Tb measures, except Tb plate-width (CCC = 0.866), the unsupervised 3DGAN-CIRCLE showed high agreement (CCC  ∈ $ \in $ [0.920 0.964]) with the true HRCT-derived reference measures. Moreover, Bland-Altman plots showed that supervised 3DGAN-CIRCLE predicted HRCT reduces bias and variability in residual values of different Tb measures as compared to LRCT and unsupervised 3DGAN-CIRCLE predicted HRCT. The supervised 3DGAN-CIRCLE method produced significantly improved performance (p < 0.001) for all Tb measures as compared to the two DL-based supervised methods available in the literature. CONCLUSIONS: 3DGAN-CIRCLE, trained in either unsupervised or supervised fashion, generates HRCT images with high structural similarity to the reference true HRCT images. The supervised 3DGAN-CIRCLE improves agreements of computed Tb microstructural measures with their reference values and outperforms the unsupervised 3DGAN-CIRCLE. 3DGAN-CIRCLE offers a viable DL solution to retrospectively improve image resolution, which may aid in data harmonization in multi-site longitudinal studies where scanner mismatch is unavoidable.

Continuum finite element analysis generalizesin vivotrabecular bone microstructural strength measures between two CT scanners with different image resolution.

Fragility of trabecular bone (Tb) microstructure is increased in osteoporosis, which is associated with rapid bone loss and enhanced fracture-risk. Accurate assessment of Tb strength usingin vivoimaging available in clinical settings will be significant for management of osteoporosis and understanding its pathogenesis. Emerging CT technology, featured with high image resolution, fast scan-speed, and wide clinical access, is a promising alternative forin vivoTb imaging. However, variation in image resolution among different CT scanners pose a major hurdle in CT-based bone studies. This paper presents nonlinear continuum finite element (FE) methods for computation of Tb strength fromin vivoCT imaging and evaluates their generalizability between two scanners with different image resolution. Continuum FE-based measures of Tb strength under different loading conditions were found to be highly reproducible (ICC ≥ 0.93) using ankle images of twenty healthy volunteers acquired on low- and high-resolution CT scanners 44.6 ± 2.7 days apart. FE stress propagation was mostly confined to Tb micro-network (2.3 ± 1.7 MPa) with nominal leakages over the marrow space (0.4 ± 0.5 MPa) complying with the fundamental principle of mechanics atin vivoimaging. In summary, nonlinear continuum FE-based Tb strength measures are reproducible among different CT scanners and suitable for multi-site longitudinal human studies.

Computed Tomography-Based Stiffness Measures of Trabecular Bone Microstructure: Cadaveric Validation and In Vivo Application.

Osteoporosis causes bone fragility and elevates fracture risk. Applications of finite element (FE) analysis (FEA) for assessment of trabecular bone (Tb) microstructural strength at whole-body computed tomography (CT) imaging are limited due to challenges with Tb microstructural segmentation. We present a nonlinear FEA method for distal tibia CT scans evading binary segmentation of Tb microstructure, while accounting for bone microstructural distribution. First, the tibial axis in a CT scan was aligned with the FE loading axis. FE cubic mesh elements were modeled using image voxels, and CT intensity values were calibrated to ash density defining mechanical properties at individual elements. For FEA of an upright volume of interest (VOI), the bottom surface was fixed, and a constant displacement was applied at each vertex on the top surface simulating different loading conditions. The method was implemented and optimized using the ANSYS software. CT-derived computational modulus values were repeat scan reproducible (intraclass correlation coefficient [ICC] ≥ 0.97) and highly correlated (r ≥ 0.86) with the micro-CT (µCT)-derived values. FEA-derived von Mises stresses over the segmented Tb microregion were significantly higher (p < 1 × 10-11) than that over the marrow space. In vivo results showed that both shear and compressive modulus for males were higher (p < 0.01) than for females. Effect sizes for different modulus measures between males and females were moderate-to-high (≥0.55) and reduced to small-to-negligible (<0.40) when adjusted for pure lean mass. Among body size and composition attributes, pure lean mass and height showed highest (r ∈ [0.45 0.56]) and lowest (r ∈ [0.25 0.39]) linear correlation, respectively, with FE-derived modulus measures. In summary, CT-based nonlinear FEA provides an effective surrogate measure of Tb microstructural stiffness, and the relaxation of binary segmentation will extend the scope for FEA in human studies using in vivo imaging at relatively low-resolution. © 2022 The Authors. JBMR Plus published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research.

Finite element analysis of trabecular bone microstructure using CT imaging and continuum mechanical modeling.

PURPOSE: Osteoporosis is a bone disease associated with enhanced bone loss, microstructural degeneration, and fracture-risk. Finite element (FE) modeling is used to estimate trabecular bone (Tb) modulus from high-resolution three-dimensional (3-D) imaging modalities including micro-computed tomography (CT), magnetic resonance imaging (MRI), and high-resolution peripheral quantitative CT (HR-pQCT). This paper validates an application of voxel-based continuum finite element analysis (FEA) to predict Tb modulus from clinical CT imaging under a condition similar to in vivo imaging by comparing with measures derived by micro-CT and experimental approaches. METHOD: Voxel-based continuum FEA methods for CT imaging were implemented using linear and nonlinear models and applied on distal tibial scans under a condition similar to in vivo imaging. First, tibial axis in a CT scan was aligned with the coordinate z-axis at 150 µm isotropic voxels. FEA was applied on an upright cylindrical volume of interests (VOI) with its axis coinciding with the tibial bone axis. Voxel volume, edge, and vertex elements and their connectivity were defined as per the isotropic image grid. A calibration phantom was used to calibrate CT numbers in Hounsfield unit to bone mineral density (BMD) values, which was then converted into calcium hydroxyapatite (CHA) density. Mechanical properties at each voxel volume element was defined using its ash-density defined on CT-derived CHA density. For FEA, the bottom surface of the cylindrical VOI was fixed and a constant displacement was applied along the z-direction at each vertex element on the top surface to simulate a physical axial compressive loading condition. Finally, a Poisson's ratio of 0.3 was applied, and Tb modulus (MPa) was computed as the ratio of average von Mises stress (MPa) of volume elements on the top surface and the applied displacement. FEA parameters including mesh element size, substep number, and different tolerance values were optimized. RESULTS: CT-derived Tb modulus values using continuum FEA showed high linear correlation with the micro-CT-derived reference values (r ∈ [0.87 0.90]) as well as experimentally measured values (r ∈ [0.80 0.87]). Linear correlation of computed modulus with their reference values using continuum FEA with linear modeling was comparable with that obtained by nonlinear modeling. Nonlinear continuum FEA-based modulus values (mean of 1087.2 MPa) showed greater difference from their reference values (mean of 1498.9 MPa using micro-CT-based FEA) as compared with linear continuum methods. High repeat CT scan reproducibility (intra-class correlation [ICC] = 0.98) was observed for computed modulus values using both linear and nonlinear continuum FEA. It was observed that high stress regions coincide with Tb microstructure as fuzzily characterized by BMD values. Distributions of von Mises stress over Tb microstructure and marrow regions were significantly different (p < 10-8 ). CONCLUSION: Voxel-based continuum FEA offers surrogate measures of Tb modulus from CT imaging under a condition similar to in vivo imaging that alleviates the need for segmentation of Tb and marrow regions, while accounting for bone distribution at the microstructural level. This relaxation of binary segmentation will extend the scope of FEA application to assess mechanical properties of bone microstructure at relatively low-resolution imaging.

Quantitative CT-Based Methods for Bone Microstructural Measures and Their Relationships With Vertebral Fractures in a Pilot Study on Smokers.

Osteoporosis causes fragile bone, and bone microstructural quality is a critical determinant of bone strength and fracture risk. This study pursues technical validation of novel CT-based methods for assessment of peripheral bone microstructure together with a human pilot study examining relationships between bone microstructure and vertebral fractures in smokers. To examine the accuracy and reproducibility of the methods, repeat ultra-high-resolution (UHR) CT and micro-CT scans of cadaveric ankle specimens were acquired. Thirty smokers from the University of Iowa COPDGene cohort were recruited at their 5-year follow-up visits. Chest CT scans, collected under the parent study, were used to assess vertebral fractures. UHR CT scans of distal tibia were acquired for this pilot study to obtain peripheral cortical and trabecular bone (Cb and Tb) measures. UHR CT-derived Tb measures, including volumetric bone mineral density (BMD), network area, transverse trabecular density, and mean plate width, showed high correlation (r > 0.901) with their micro-CT-derived values over small regions of interest (ROIs). Both Cb and Tb measures showed high reproducibility-intra-class correlation (ICC) was greater than 0.99 for all Tb measures except erosion index and greater than 0.97 for all Cb measures. Female sex was associated with lower transverse Tb density (p < 0.1), higher Tb spacing (p < 0.05), and lower cortical thickness (p < 0.001). Participants with vertebral fractures had significantly degenerated values (p < 0.05) for all Tb measures except thickness. There were no statistically significant differences for Cb measures between non-fracture and fracture groups. Vertebral fracture-group differences of Tb measures remained significant after adjustment with chronic obstructive pulmonary disease (COPD) status. Although current smokers at baseline had more fractures-81.8% versus 63.2% for former smokers-the difference was not statistically significant. This pilot cross-sectional human study demonstrates CT-based peripheral bone microstructural differences among smokers with and without vertebral fractures. © 2021 The Authors. JBMR Plus published by Wiley Periodicals, Inc. on behalf of American Society for Bone and Mineral Research. © 2021 The Authors. JBMR Plus published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research.

A comparative study of trabecular bone micro-structural measurements using different CT modalities.

Osteoporosis, characterized by reduced bone mineral density and micro-architectural degeneration, significantly enhances fracture-risk. There are several viable methods for trabecular bone micro-imaging, which widely vary in terms of technology, reconstruction principle, spatial resolution, and acquisition time. We have performed an excised cadaveric bone specimen study to evaluate different computed tomography (CT)-imaging modalities for trabecular bone micro-structural analysis. Excised cadaveric bone specimens from the distal radius were scanned using micro-CT and fourin vivoCT imaging modalities: high-resolution peripheral quantitative computed tomography (HR-pQCT), dental cone beam CT (CBCT), whole-body multi-row detector CT (MDCT), and extremity CBCT. A new algorithm was developed to optimize soft thresholding parameters for individualin vivoCT modalities for computing quantitative bone volume fraction maps. Finally, agreement of trabecular bone micro-structural measures, derived from differentin vivoCT imaging, with reference measures from micro-CT imaging was examined. Observed values of most trabecular measures, including trabecular bone volume, network area, transverse and plate-rod micro-structure, thickness, and spacing, forin vivoCT modalities were higher than their micro-CT-based reference values. In general, HR-pQCT-based trabecular bone measures were closer to their reference values as compared to otherin vivoCT modalities. Despite large differences in observed values of measures among modalities, high linear correlation (rε [0.94 0.99]) was found between micro-CT andin vivoCT-derived measures of trabecular bone volume, transverse and plate micro-structural volume, and network area. All HR-pQCT-derived trabecular measures, except the erosion index, showed high correlation (rε [0.91 0.99]). The plate-width measure showed a higher correlation (rε [0.72 0.91]) amongin vivoand micro-CT modalities than its counterpart binary plate-rod characterization-based measure erosion index (rε [0.65 0.81]). Although a strong correlation was observed between micro-structural measures fromin vivoand micro-CT imaging, large shifts in their values forin vivomodalities warrant proper scanner calibration prior to adopting in multi-site and longitudinal studies.

Deep Learning Based High-Resolution Reconstruction of Trabecular Bone Microstructures from Low-Resolution CT Scans using GAN-CIRCLE.

Osteoporosis is a common age-related disease characterized by reduced bone density and increased fracture-risk. Microstructural quality of trabecular bone (Tb), commonly found at axial skeletal sites and at the end of long bones, is an important determinant of bone-strength and fracture-risk. High-resolution emerging CT scanners enable in vivo measurement of Tb microstructures at peripheral sites. However, resolution-dependence of microstructural measures and wide resolution-discrepancies among various CT scanners together with rapid upgrades in technology warrant data harmonization in CT-based cross-sectional and longitudinal bone studies. This paper presents a deep learning-based method for high-resolution reconstruction of Tb microstructures from low-resolution CT scans using GAN-CIRCLE. A network was developed and evaluated using post-registered ankle CT scans of nineteen volunteers on both low- and high-resolution CT scanners. 9,000 matching pairs of low- and high-resolution patches of size 64×64 were randomly harvested from ten volunteers for training and validation. Another 5,000 matching pairs of patches from nine other volunteers were used for evaluation. Quantitative comparison shows that predicted high-resolution scans have significantly improved structural similarity index (p < 0.01) with true high-resolution scans as compared to the same metric for low-resolution data. Different Tb microstructural measures such as thickness, spacing, and network area density are also computed from low- and predicted high-resolution images, and compared with the values derived from true high-resolution scans. Thickness and network area measures from predicted images showed higher agreement with true high-resolution CT (CCC = [0.95, 0.91]) derived values than the same measures from low-resolution images (CCC = [0.72, 0.88]).