Skip to main content
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2024 Apr 19.
Published in final edited form as: Proc SPIE Int Soc Opt Eng. 2023 Apr 19;12468:1246813. doi: 10.1117/12.2654304

Effects of non-stationary blur on texture biomarkers of bone using Ultra-High Resolution CT

G Shi a, F J Quevedo Gonzalez b, R E Breighner b, JA Carrino c, JH Siewerdsen d, W Zbijewski a
PMCID: PMC10788132  NIHMSID: NIHMS1957389  PMID: 38226358

Abstract

Purpose:

To advance the development of radiomic models of bone quality using the recently introduced Ultra-High Resolution CT (UHR CT), we investigate inter-scan reproducibility of trabecular bone texture features to spatially-variant azimuthal and radial blurs associated with focal spot elongation and gantry rotation.

Methods:

The UHR CT system features 250×250 μm detector pixels and an x-ray source with a 0.4×0.5 mm focal spot. Visualization of details down to ~150 μm has been reported for this device. A cadaveric femur was imaged on UHR CT at three radial locations within the field-of-view: 0 cm (isocenter), 9 cm from the isocenter, and 18 cm from the isocenter; we expect the non-stationary blurs to worsen with increasing radial displacement. Gray level cooccurrence (GLCM) and gray level run length (GLRLM) texture features were extracted from 237 trabecular regions of interest (ROIs, 5 cm diameter) placed at corresponding locations in the femoral head in scans obtained at the different shifts. We evaluated concordance correlation coefficient (CCC) between texture features at 0 cm (reference) and at 9 cm and 18 cm. We also investigated whether the spatially-variant blurs affect K-means clustering of trabecular bone ROIs based on their texture features.

Results:

The average CCCs (against the 0 cm reference) for GLCM and GLRM features were ~0.7 at 9 cm. At 18 cm, the average CCCs were reduced to ~0.17 for GLCM and ~0.26 for GLRM. The non-stationary blurs are incorporated in radiomic features of cancellous bone, leading to inconsistencies in clustering of trabecular ROIs between different radial locations: an intersection-over-union overlap of corresponding (most similar) clusters between 0 cm and 9 cm shift was >70%, but dropped to <60% for the majority of corresponding clusters between 0 cm and 18 cm shift.

Conclusion:

Non-stationary CT system blurs reduce inter-scan reproducibility of texture features of trabecular bone in UHR CT, especially for locations >15 cm from the isocenter. Radiomic models of bone quality derived from UHR CT measurements at isocenter might need to be revised before application in peripheral body sites such as the hips.

Keywords: bone microstructure, high resolution CT, quantitative CT, bone imaging, radiomics, texture analysis

1. INTRODUCTION

Recent years have seen an emergence of multi-detector CT (MDCT) systems with Ultra-High Resolution (UHR) capabilities. Such devices use new x-ray source and detector hardware to enable visualization of anatomical details at a ~100 – 200 μm scale, almost 2x finer than in current MDCT. Examples include conventional energy-integrating MDCT designs, such as the Canon Precision CT used in this work (maximal spatial resolution of ~150 μm)1, as well as the recently introduced photon-counting CT (PCCT) with direct conversion detectors – in fact, UHR imaging is often perceived as one of the immediate clinical benefits of PCCT.2

Among the promising applications of UHR CT is bone health evaluation - e.g., in early detection of osteoporosis (OP)3 and osteoarthritis (OA)4 – where high spatial resolution might enable quantitative assessment of trabecular microstructure to augment conventional measurements of bone mineral density (BMD). Similar to radiomics in oncology, image texture features from conventional CT have been demonstrated to provide useful indirect markers of cancellous microarchitecture amenable to predictive statistical modeling5. We investigate the potential benefits of using UHR CT to derive texture features for such radiomic bone quality analysis.

Image texture is affected both by the underlying tissue morphology and the image formation process. Therefore, the robustness of radiomic models to changes in image acquisition settings needs to be assessed6. We evaluate the sensitivity of UHR bone texture features to spatially-variant spatial resolution within the scanner field-of-view (FOV). Such nonstationarities include azimuthal blur caused by gantry motion and radial blur due to increase in the projected x-ray focal spot size (both worse at the periphery of the FOV). These effects might dominate local spatial resolution in UHR CT, where detector pixel aperture blur is minimized7.

The results of the experimental study presented below will help determine whether radiomic bone quality models in UHR CT need to be adapted to the position of target bone in the FOV. This might be particularly relevant for peripheral body sites such as the hip.

2. METHODS

2.1. UHR CT system and imaging protocols

The UHR CT is equipped with a 160-row x-ray detector with 250×250 μm pixels (measured at isocenter) and an x-ray source with a 0.4×0.5 mm focal spot. The scanner can also be operated in an NR mode with 500×500 μm pixels and 0.8 × 1.3 mm focal spot. A cadaveric femur was imaged in the UHR and NR modes at 0 cm, 9 cm, and 18cm away from the center of the gantry (Fig. 1(A)). All acquisitions used 120 kVp tube voltage, 200mAs exposure and 1.5 sec rotation time. The UHR reconstruction protocol involved FC30 kernel, 0.127 mm in-plane voxel size and 0.25 mm slice thickness. The NR reconstruction settings were FC30 kernel, 0.256 mm in-plane voxel size, and 0.5 mm slice thickness.

Fig. 1.

Fig. 1.

(A) Femur imaging setup. (B) Resolution phantom and 3D edge spread function sampling regions.

2.2. Modulation Transfer Function (MTF) measurement

To provide an objective assessment of non-stationary blurs in the system, we measured radial and azimuthal Modulation Transfer Function (MTF) using the resolution phantom shown in Fig. 1(B). The phantom consisted of 5 cylindrical acrylic inserts (~3cm diameter, 4.4cm length) inserted 5 cm apart into a foam block. It was positioned such that the inserts were located 0 to 20 cm from the center of the gantry and imaged using the same protocols as for the cadaveric femur. 3D MTF was derived from edge spread of cylindrical inserts8.

2.4. Texture analysis

The reconstructions of the cadaveric femur were resampled to isotropic voxels (0.25 mm size for UHR and 0.5 mm size for NR). Spherical Region of Interests (ROIs) 5 mm in diameter were placed in the femoral head (237 ROIs) of the 0 cm scan and propagated to corresponding locations in the 9 cm and 18 cm scans using rigid transformations obtained from registration of the bone volumes. Gray level cooccurrence (GLCM) and gray level run length (GLRLM) texture features were extracted from each ROI at each radial shift using Pyradiomics.

Reproducibility of texture features across different radial shifts was assessed using Concordance Correlation Coefficient against values measured from the scan of the femur placed at 0 cm.

The subsequent analyses were performed after dimensionality reduction of ROI texture measurements using Principle Component Analysis (PCA). We report on two experiments. First, the PCA was computed jointly across all ROIs at all radial shifts. A Support Vector machine (SVM) classifier was trained on the PCA scores of each ROI to build a multiclass model distinguishing ROIs from the different radial shifts. The classification accuracy was assessed using leave-one-out cross-validation.

In the second experiment, the PCA was applied to ROI texture measurements separately for each radial shift of the femur. At each shift, unsupervised K-mean clustering was performed on PCA texture scores of the ROIs. The number of clusters was determined in the dataset imaged at isocenter (0 cm shift) by minimizing the Davies–Bouldin index. We then used the same number of clusters at 0 cm and 9 cm shifts. Corresponding clusters between the 0 cm dataset and the radially shifted acquisitions were established by finding cluster pairs with the largest number of common ROIs. We then evaluated the degree of ROI overlap between the corresponding clusters using an intersection over union metric.

3. RESULTS AND BREAKTHROUGH WORK

Fig. 2 shows the radial and azimuthal MTF of UHR CT at different distances from the isocenter. There is a substantial degradation in radial and azimuthal spatial resolution toward the periphery of the FOV. At 20 cm away from isocenter, UHR MTF approaches NR MTF at 0 cm. Fig. 3 illustrates the effects of this spatially-variant blur on visualization of trabecular bone using femoral head scans acquired at different radial shifts.

Fig. 2.

Fig. 2.

Radial and azimuthal MTFs of UHR CT at 0 cm, 10 cm, 15 cm, and 20 cm from isocenter. The MTFs of NR mode at 0 cm are shown for comparison. The UHR mode yields substantially improved spatial resolution compared to NR. However, UHR resolution diminishes towards the periphery of the FOV due to non-stationary effects of radial focal spot elongation and azimuthal gantry rotation blur. We investigate whether those spatially-variant blurs affects the reproducibility of texture features of bone across the scanner FOV.

Fig. 3.

Fig. 3.

Femoral head reconstructions at different resolutions and radial distances from the isocenter. At 0 cm and 9 cm from the isocenter, the UHR mode provides clearly improved delineation of trabecular bone compared to NR at 0 cm. This advantage becomes less pronounced at 18 cm because of spatially-variant resolution loss. The yellow box indicates the location of magnified detail views in the bottom row.

The loss of resolution associated with off-center locations affects the reproducibility of texture metrics of trabecular bone between data acquired at isocenter and at a radial shift, as demonstrated by the CCC values in Fig. 4 (A). The average CCCs for GLCM and GLRM features are ~0.7 at 9 cm. At 18 cm, the average CCCs are reduced to ~0.17 for GLCM and ~0.26 for GLRM. The clear separation of ROIs imaged at 0 cm and 18 cm in the PCA projection in Fig. 4B indicates that non-stationary blurs become incorporated into texture features of trabecular bone. In fact, the leave-one-out SVM experiments showed that the radial shift of an ROI can be predicted from texture features with 96%, 72%, and 79% classification accuracies for 0, 9, and 18 cm radial shifts, respectively.

Fig. 4.

Fig. 4.

(A) CCC between texture metrics measured in the femoral head imaged at isocenter and at a radial shift. Results for a representative subset of GLRM and GLCM metrics are shown. Poor concordance at 18 cm shift is apparent. (B) Projection of UHR texture measurements of 237 trabecular ROIs onto the first two principal components of the distribution of GLRM+GLCM features. The distance from isocenter appears to be partly encoded in the texture metrics, leading to a separation of measurements taken at 18 cm (yellow) from those obtained at 0 cm (blue).

Figure 5 shows the clustering of ROIs based on texture features. The corresponding clusters identified in the 0 cm and 9 cm datasets are fairly consistent, with the intersection-over-union overlap metric of 90% for cluster #1 (i.e., cluster #1 in the 0 cm dataset and its most similar cluster at 9 cm), 74% for cluster #2, 72% for cluster #3, and 76% for cluster #4. Between 0 cm and 18cm datasets, the overlap metrics of the corresponding clusters decrease to 56% for cluster #1, 74% for cluster #2, 45% for cluster #3, and 56% for cluster #4.

Fig. 5.

Fig. 5.

Unsupervised K-means clustering of trabecular ROIs performed in the PCA component space of ROI texture features. The clusters of trabecular ROIs are fairly consistent between datasets acquired at 0 cm and 9 cm from isocenter. Cluster consistency with 0 cm data is diminished for measurements obtained using the UHR femoral head scan at 18cm from isocenter. ROIs close to cluster centers of mass are shown for illustration.

4. CONCLUSIONS

Non-stationary CT system blurs associated with focal spot elongation and gantry rotation reduce inter-scan reproducibility of texture features of trabecular bone in UHR CT, especially for regions >15 cm from the isocenter. Resolution loss at the periphery of scanner FOV produces substantial enough change in image texture that the location of trabecular ROI can be predicted based on its GLRM and GLCM features. As a result, trabecular microstructure classes identified by clustering texture metrics at the isocenter might not be reproducible at other radial shifts9. Radiomic models of bone quality derived from UHR CT measurements at isocenter might need to be revised before application in peripheral body sites such as the hips.

ACKNOWLEDGEMENT

The work was supported by NIH R01-EB-029446.

REFERENCES

  • [1]:Yanagawa M, Hata A, Honda O, et al. Subjective and objective comparisons of image quality between ultra-high-resolution CT and conventional area detector CT in phantoms and cadaveric human lungs. European Radiology. 2018;28(12):5060–5068. doi: 10.1007/s00330-018-5491-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [2]:Wehrse E, Sawall S, Klein L et al. Potential of ultra-high-resolution photon-counting CT of bone metastases: initial experiences in breast cancer patients. npj Breast Cancer 7, 3 (2021). 10.1038/s41523-020-00207-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [3]:Samelson EJ, Broe KE, Xu H, et al. Cortical and trabecular bone microarchitecture as an independent predictor of incident fracture risk in older women and men in the Bone Microarchitecture International Consortium (BoMIC): a prospective study. The Lancet Diabetes & Endocrinology. 2018;7(January):34–43. doi: 10.1016/S2213-8587(18)30308-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [4]:Coughlin TR, Kennedy OD. The role of subchondral bone damage in post-traumatic osteoarthritis. Ann N Y Acad Sci. 2016;1383(1):58–66. doi: 10.1111/nyas.13261 [DOI] [PubMed] [Google Scholar]
  • [5]:Nagarajan MB, Checefsky WA, Abidin AZ, et al. Characterizing trabecular bone structure for assessing vertebral fracture risk on volumetric quantitative computed tomography. In: Gimi B, Molthen RC, eds. Proc of SPIE Medical Imaging. Vol 9417.; 2015:94171E. doi: 10.1117/12.2082059 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [6]:Midya A, Chakraborty J, Gönen M, Do RKG, Simpson AL. Influence of CT acquisition and reconstruction parameters on radiomic feature reproducibility. Journal of Medical Imaging. 2018;5(01):1. doi: 10.1117/1.jmi.5.1.011020 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [7]:Hernandez A, Wu P, Mahesh M, Siewerdsen JH, Boone JM. Location and direction dependence in the 3D MTF for a high-resolution CT system. Med Phys. 48(6):2760–2771, 2021 [DOI] [PubMed] [Google Scholar]
  • [8]:Wu P, Boone JM, Hernandez AM, Mahesh M and Siewerdsen JH (2021), Theory, method, and test tools for determination of 3D MTF characteristics in cone-beam CT. Med. Phys, 48: 2772–2789. 10.1002/mp.14820 [DOI] [PubMed] [Google Scholar]
  • [9]:Valentinitsch A, Patsch JM, Burghardt AJ, Link TM, Majumdar S, Fischer L, Schueller-Weidekamm C, Resch H, Kainberger F, Langs G. Computational identification and quantification of trabecular microarchitecture classes by 3-D texture analysis-based clustering. Bone. 2013. May;54(1):133–40. doi: 10.1016/j.bone.2012.12.047. Epub 2013 Jan 10. [DOI] [PubMed] [Google Scholar]

RESOURCES