Abstract
Objectives:
Radiomics refers to the extraction and analysis of advanced quantitative imaging from medical images to diagnose and/or predict diseases. In the dentistry field, the bone data from mandibular condyles could be computationally analyzed using the voxel information provided by high-resolution CBCT scans to increase the diagnostic power of temporomandibular joint (TMJ) conditions. However, such quantitative information demands innovative computational software, algorithm implementation, and validation. Our study's aim was to compare a newly developed BoneTexture application to two-consolidated software with previous applications in the medical field, Ibex and BoneJ, to extract bone morphometric and textural features from mandibular condyles.
Methods:
We used an imaging database of HR-CBCT TMJs scans with an isotropic voxel size of 0.08 mm3 . A single group with 66 distinct mandibular condyles composed the final sample. We calculated 18 variables for bone textural features and 5 for bone morphometric measurements using the Ibex, BoneJ and BoneTexture applications. Spearman correlation and Bland–Altman plot analyses were done to compare the agreement among software.
Results:
The results showed a high Spearman correlation among the software applications ( r = 0.7–1), with statistical significance for all variables, except Grey Level Non-Uniformity and Short Run Emphasis. The Bland–Altman vertical axis showed, in general, good agreement between the software applications and the horizontal axis showed a narrow average distribution for Correlation, Long Run Emphasis and Long Run High Grey Level Emphasis.
Conclusions:
Our data showed consistency among the three applications to analyze bone radiomics in high-resolution CBCT. Further studies are necessary to evaluate the applicability of those variables as new bone imaging biomarkers to diagnose bone diseases affecting TMJs.
Keywords: Cone-Beam computed tomography; temporomandibular joint disorders; software validation; tomography, X-Ray computed
Introduction
Quantitative imaging of mandibular condyles using cone beam CT (CBCT) is being developed in Oral and Maxillofacial Radiology research.1,2 Widespread incorporation of radiomics, that means, extracting quantitative information from medical images by converting them into minable high-dimensional data, will provide more accurate biological and clinical correlations as well as prognostic value.3 The data provided have the potential to change clinical practice from a qualitative analysis of imaging features to a more dynamic, spatial, and phenotypical characterization of suspected lesions. There is a need for studies of diagnostic methodologies to better classify the imaging findings and the discovery of new biomarkers, that were discussed in different imaging sources such as MR, PET and CT.4–8
The high resolution CBCT (HR-CBCT) is a new modality that provides more information and better spatial resolution. Increasing the amount of acquired voxels provides flexibility for the selection of smaller field of views with higher spatial resolution while following the ALARA/ALADA principle.9,10 HR-CBCT also facilitates the analysis of different bone parameters such as bone morphometry and textural features that were only possible using micro-CT, being a promising tool for the analysis of new bone-imaging biomarkers and its applications are promising.11,12 The applicability of these new biomarkers may directly impact health and disease diagnosis in future studies; however, the computational data extraction depends on precise software and biomarkers validation. Cancer studies have pointed out the role of radiomics in early disease prediction, and bone conditions, such as osteoarthritis, demand an early diagnose in order to reduce the chronic musculoskeletal damage that is usually detected just in late stages.3,13–15
For imaging data extraction, different software applications have been developed to improve the radiomics approaches and selection of significant imaging features. Ibex16 (open-source software) and BoneJ17 (application of Fiji software18 are examples of open-source methods; however, the software user interfaces are still complex and the applications are not specific to CBCT image. To simplify those approaches, Vimort et al,19 implemented a new application in the software 3D-Slicer (https://www.slicer.org) called BoneTexture.20 The software has a user-friendly interface and validation of ex vivo CT imaging has been performed.1,2,12
Due to the software applications complexity, algorithms implementation and data interpretation, a comparison of the available computational sources is needed before applying such tools for clinical decision-making. Thus, the aim of this study was to compare three software applications: BoneTexture, Ibex and BoneJ for analysis of 18 bone texture features and 5 bone morphometric variables from 66 condyles. We hypothesized that there are no differences among the variables using the three software applications.
Methods and materials
Study sample
This study was a secondary data analysis approved by the Institutional Review Board (HUM00113199) of the University of Michigan. One group composed our sample, obtained from a database with 34 patients, totalizing 68 HR-CBCT mandibular condyles scans. After outlier analysis, two condyles were excluded, [the values differed in three standard deviation from the 95% confidence interval (CI)], resulting in a single group with 66 condyles.
Sample characteristics
We aimed to test the agreement and performance of different software to extract bone textural features in a single group with a sample containing real clinical data. Towards this aim, 17 of our subjects had clinical diagnosis of TMJ-OA (RDC/TMD clinical criteria)13 and 17 were healthy. The goal was not to compare different conditions (disease x healthy) but to have both situations represented in our sample.
Imaging acquisition
The HR-CBCT scans contained in the database were acquired in a 3D Accuitomo (J. Morita MFG. CORP Tokyo, Japan), according to the following TMJ acquisition protocol: field of view 40 × 40 mm; 90 kV, 5 mA and 30.8 s and a voxel size of 0.08 mm3 at the University of Michigan—School of Dentistry.
Image pre-processing
For image and variable standardization each HR-CBCT DICOM volume was cropped into a small region of interest (ROI) greyscale sample of 50 × 50×50 slices (voxel size of 0.08 mm3) using the module “Crop-Volume” of the 3D-Slicer Software and saved as a single image volume (.dcm). Additionally, the ROI spatial position was kept to contain the trabecular/cortical condyle bone without extrapolated the condyle boundaries (Figure 1). Other image-processing steps were necessary in order to comply with specific software input file needs as explained in the next subsections.
Figure 1.
Computational Image processing and sample preparation. (A) 3D condyle rendering and its ROI in the center; (B) Original greyscale HR-CBCT in the sagittal slice showing the limits of the ROI; (C) ROI—3D Rendering; (D) ROI greyscale volume used to compute GLCM and GLRLM variables (50 × 50 × 50 slices); (E) ROI after binary transformation using the median thresholding algorithm available in BoneJ. This binary image was used to compute bone morphometry. 3D, three-dimensional; GLCM, Grey-level Co-occurrence Matrix; GLRLM, Grey-level Run Length matrix; HR-CBCT, high-resolution CBCT; ROI, region of interest.
Image processing for bone textural features
The Ibex16 software requires a ROI segmentation in addition to the 50 × 50 × 50 slices greyscale samples. The itk-SNAP21 software was used to create, select and export an annotation ROI (segmentation) as a single image volume, with the same volume and size for each sample. Then, the greyscale samples and their annotations were converted back into DICOM format using the 3D-Slicer. The files were loaded in the Ibex and BoneTexture applications in order to compute bone texture and bone morphometry features (Figure 2). 18 variables were divided in two main subgroups: (1) Grey-level Co-occurrence Matrix (GLCM)22 and (2) Grey-level Run Length matrix (GLRLM).23,24 The first group GLCM gives values for the distribution of co-occurring pixel values and includes: Energy, Entropy, Correlation, Inverse Difference Moment, Inertia, Cluster Shade, Cluster Prominence and Haralick Correlation. The second group, GLRLM,23,24 gives the size of homogeneous runs for each grey level and includes: Short Run Emphasis, Long Run Emphasis, Grey Level Non-Uniformity, Run Length Non-Uniformity, Low Grey Level Run Emphasis, High Grey Level Run Emphasis, Short Run Low Grey Level Emphasis, Short Run High Grey Level Emphasis, Long Run Low Grey Level Emphasis and Long Run High Grey Level Emphasis.
Figure 2.
Computational workflow; (A) ROI cropping (50 × 50 × 50 slices) from each condyle; (B) Annotation (segmentation) process of the ROI using ITK-snap and 3D-Slicer. This additional step was necessary to comply with Ibex requisites; (C) BoneTexture computation of the GLCM and GLRLM variables; (D) Ibex computation for GLCM and GLRLM variables. This step demands the annotation in addition to the ROI; (E) ROI binary conversion using the BoneJ plugin; (F) Input of the binary ROIs in the BoneJ and BoneTexture to compute the bone morphometry variables. GLCM, Grey-level Co-occurrence Matrix; GLRLM, Grey-level Run Length matrix; HR-CBCT, high-resolutionCBCT; ROI, region of interest.
Image processing for bone morphometry variables
Initially the 50 × 50 × 50 slices greyscale volume samples were converted into a binary image using the BoneJ Application18 tool called “median thresholding” and were then exported as Tagged Image File Format (tiff). These images were imported to the BoneJ application and BoneTexture for bone morphometry analysis (Figure 2). The following variables were examined: trabecular thickness (Tb.Th), trabecular separation (Tb.Sp), trabecular number (Tb.N), bone volume per total volume (BV/TV) and bone surface per bone volume (BS/BV).11,25 The values of Tb.N for the BoneJ software were obtained using the formula: Tb.N=(BV/TV)/Tb.Th, since this parameter is not explicitly computed in BoneJ.
Software parameters
We have based our parameters in an optimized protocol described by BoneTexture developers.20 The values used in BoneTexture to computed GLRLM variables were: number of bins: 10; voxel intensity range: min −250/max 4000; distance range: min 0/max 1 and neighbourhood radius: 4. For GLCM: number of bins: 10; voxel intensity range: min −250/max 4000 and neighbourhood radius: 4. For the Ibex software, GLCM values were: direction = 0, 45, 90, 135, 180, 225, 270 and 315; AdaptLimitLevel = 0; GreyLimits=-250/2100; NumLevels = 10; Offset = 1; and Symmetric = 0. For the GLRLM: Direction = 0, 45, 90, 135, 180, 225, 270, 315; GreyLimits=-250/4000; NumLevels = 10. Those parameters are sensitive to small changes and different software do not ask for the same inputs. In addition, for Ibex, the average of each direction was computed to have one value per variable and for the BoneTexture one final value is given. No filters were applied to analyze the GLCM and GLRM variables.
Variables description
An explanation and description of each textural features and bone morphometry are presented in Table 1.
Table 1.
Description of the variables for bone morphometry and bone textural features
| Bone Morphometric Features11 | Description |
| BV/TVa | Ratio between bone volume and total volume |
| Tb.Tha | Trabecular Thickness |
| Tb.Spa | Trabecular Separation |
| Tb.Na | Trabecular Number |
| BS/BVa | Ratio between bone surface and bone volume (Surface density) |
| GLRLM Features18,22 | The GLRLM gives the size and length of homogeneous runs/points with the same grey-level value. |
| Short Run Emphasisb | Large for fine textures. |
| Long Run Emphasisb | Large for coarse structural textures. |
| Grey Level Non-Uniformityb | Small if the grey-level values are alike in the image. |
| Run Length Non-Uniformityb | Small if the run lengths are alike through out the image. |
| Low Grey Level Run Emphasisb | Large if the image has many low runs grey-value. |
| High Grey Level Run Emphasisb | Large if the image has many high runs grey-value. |
| Short Run Low Grey Level Emphasisb | Large if the image has many short runs of low grey-value. |
| Short Run High Grey Level Emphasisb | Large if the image has many short runs of high grey-value. |
| Long Run Low Grey Level Emphasisb | Large if the image has many long runs of low grey-value. |
| Long Run High Grey Level Emphasisb | Large if the image has many long runs of high grey-value. |
| GLCM FEATURES18,26 | The grey-level co-occurrence matrix (GLCM) describes the distribution of co-occurring pixel/voxel values at a given offset/direction. |
| Energyb | Large if the image has textural uniformity and organization. |
| Entropyb | Large if the image has a random distribution of grey-level intensities and small for same grey-level distribution. |
| Correlationb | Indicates the Grey-level linear dependence between the pixels at the specified positions relative to each other. |
| Inverse Difference Momentb | Large value for homogeneous images. |
| Inertiab | Small for image with similar grey-level values. Also known as Contrast. |
| Cluster Shadeb | Large for asymmetric images |
| Cluster Prominenceb | Large for asymmetric images |
| Haralick Correlationb | Linear dependence between pixels relative to each other. **Ibex software gives the Autocorrelation. |
BS, bone surface; BV, bone volume; TV, total volume.
Calculated using BoneJ and BoneTexture applications
Calculated using Ibex and BoneTexture applications
Statistical analysis
A single examiner performed the study and, as the software applications automatically calculated the values, the interexaminer reproducibility was not assessed. The variables presented a non-normal distribution after analysis of the kurtosis and symmetry. Thus, the Spearman correlation test was used to evaluate the correlation among all the variables between the software. Due to the complex algorithms nature, language programming and differences in the scales of the software, normalization of the data was necessary prior to descriptive statistics and Bland–Altman analysis.27,28 The normalization formula was: Xi=(x-x̄)/S where Xi = new value, x = original value, x̄=average of the values and S = standard deviation.
Results
Table 2 shows the descriptive statistics for the normalized values for each variable and software compared. As expected, the average is close to 0 and the standard deviation close to 1 for all variables with different variation of the CI ranges.
Table 2.
Descriptive statistics for each variable and respective software
| BoneTexture Application | BoneJ Application | |||||||
| Variables (n = 66) | 95% CI | 95% CI | ||||||
| Mean | Min | Max | SD | Mean | Min | Max | SD | |
| BV/TV (%) | −0.01 | −3.15 | 3.22 | 1.01 | −0.01 | −3.15 | 3.21 | 1.01 |
| Tb.Th (mm) | −0.03 | −1.62 | 4 | 0.98 | −0.01 | −1.4 | 3.04 | 1.01 |
| Tb.Sp (mm) | −0.01 | −1.75 | 3.58 | 1.01 | −0.02 | −2.04 | 2.74 | 0.99 |
| Tb.N (mm−1) | 0.02 | −2.44 | 3.08 | 0.99 | 0 | −2.22 | 2.02 | 1.01 |
| BS/BV (mm−1) | 0.03 | −3.19 | 2.41 | 0.98 | 0.03 | −3.05 | 3.56 | 0.98 |
| BoneTexture Application | IBEX Application | |||||||
| GLRLM_Short Run Emphasis | −0.07 | −2.66 | 1.53 | 0.93 | 0.06 | −1.48 | 2.53 | 0.87 |
| GLRLM_Long Run Emphasis | 0.07 | −1.53 | 2.66 | 0.93 | −0.11 | −0.9 | 2.38 | 0.61 |
| GLRLM_Grey Level Non-Uniformity | 0.04 | −1.58 | 2.45 | 0.86 | 0.06 | −2.88 | 1.86 | 0.93 |
| GLRLM_Run Length Non-Uniformity | 0.07 | −1.63 | 3.13 | 0.92 | 0.07 | −2.07 | 2.93 | 0.93 |
| GLRLM_Low Grey Level Run Emphasis | −0.02 | −2.75 | 1.72 | 1.01 | −0.03 | −2.47 | 2.72 | 0.97 |
| GLRLM_High Grey Level Run Emphasis | 0.03 | −1.43 | 3.56 | 1 | 0.03 | −2.02 | 3.66 | 0.98 |
| GLRLM_Short Run Low Grey Level Emphasis | −0.02 | −2.74 | 1.68 | 1.01 | −0.01 | −1.6 | 2.36 | 0.91 |
| GLRLM_Short Run High Grey Level Emphasis | 0.03 | −1.45 | 3.56 | 1 | 0.06 | −1.49 | 2.92 | 0.96 |
| GLRLM_Long Run Low Grey Level Emphasis | −0.01 | −2.79 | 1.92 | 1.01 | −0.11 | −1.05 | 2.34 | 0.64 |
| GLRLM_Long Run High Grey Level Emphasis | 0.03 | −1.36 | 3.54 | 0.99 | −0.12 | −0.85 | 2.38 | 0.57 |
| GLCM_Energy | −0.08 | −1.61 | 2.17 | 0.87 | −0.06 | −1.75 | 2.3 | 0.91 |
| GLCM_Entropy | 0.07 | −1.6 | 1.95 | 0.92 | 0.06 | −1.89 | 2.02 | 0.93 |
| GLCM_Correlation | −0.11 | −1.25 | 1.56 | 0.69 | 0.02 | −1.83 | 1.86 | 0.99 |
| GLCM_Inverse Difference Moment | −0.09 | −2.28 | 1.9 | 0.84 | −0.09 | −2.12 | 2.34 | 0.87 |
| GLCM_Inertia | 0.09 | −1.9 | 2.28 | 0.84 | 0.09 | −2.34 | 2.12 | 0.87 |
| GLCM_Cluster Shade | 0.02 | −1.7 | 4.16 | 1.01 | 0.02 | −2.83 | 4.01 | 1.01 |
| GLCM_Cluster Prominence | 0.02 | −0.81 | 4.5 | 1.01 | 0.02 | −0.8 | 4.2 | 1.01 |
| GLCM_Haralick Correlation | 0.02 | −0.79 | 5.15 | 1.01 | 0.02 | −1.53 | 4.05 | 1.00 |
BS, bone surface; BV, bone volume; CI, confidence interval; GLCM, Grey-level Co-occurrence Matrix; GLRLM, Grey-level Run Length matrix; SD, standard deviation; TV, total volume.
High Spearman correlations (r = 0.7–1) among the different software applications are shown in Table 3. The only variables that did not show a significant correlation were Short Run Emphasis and Grey Level Non-Uniformity. The variables: Correlation, Long Run Emphasis and Long Run High Grey Level Emphasis showed statistically significance but negative correlation, suggesting that the applications used different mathematical approach’s to obtain these values.
Table 3.
Spearman correlation among the BoneTexture, Ibex and BoneJ applications
| Variables (n = 66) | r value | p-value |
| BV/TVa | 1 | <0.001 |
| Tb.Tha | 0.620 | <0.001 |
| Tb.Spa | 0.853 | <0.001 |
| BS/BVa | 0.974 | <0.001 |
| Tb.Na | 0.891 | <0.001 |
| GLCM_Energyb | 0.603 | <0.001 |
| GLCM_Entropyb | 0.857 | <0.001 |
| GLCM_Correlationb | −0.835 | <0.001 |
| GLCM_Inverse Difference Momentb | 0.755 | <0.001 |
| GLCM_Inertiab | 0.747 | <0.001 |
| GLCM_Cluster Shadeb | 0.747 | <0.001 |
| GLCM_Cluster Prominenceb | 0.796 | <0.001 |
| GLCM_Haralick Correlationb | 0.896 | <0.001 |
| GLRLM_Short Run Emphasisb | −0.230 | 0.060 |
| GLRLM_Long Run Emphasisb | −0.882 | <0.001 |
| GLRLM_Grey Level Non-Uniformityb | −0.061 | 0.627 |
| GLRLM_Run Length Non-Uniformityb | 0.855 | <0.001 |
| GLRLM_Low Grey Level Run Emphasisb | 0.734 | <0.001 |
| GLRLM_High Grey Level Run Emphasisb | 0.925 | <0.001 |
| GLRLM_Short Run Low Grey Level Emphasisb | 0.482 | <0.001 |
| GLRLM_Short Run High Grey Level Emphasisb | 0.770 | <0.001 |
| GLRLM_Long Run Low Grey Level Emphasisb | 0.762 | <0.001 |
| GLRLM_Long Run High Grey Level Emphasisb | −0.499 | <0.001 |
BS, bone surface; BV, bone volume; GLCM, Grey-level Co-occurrence Matrix; GLRLM, Grey-level Run Length matrix; TV, total volume.
α = 95% and 2-tailed p value; p < 0.001 statistically significant.
Comparison between BoneTexture and Ibex
Comparison between BoneTexture and BoneJ
The Bland-Altman plots for comparison among software applications show a good agreement (vertical axis) between each software applications for all of the features as shown in the Figures 3–5 with most values within the CI. The horizontal dots distribution (horizontal axis) represent the average for each sample computed by the software applications. A larger distribution is seen for BV/TV, Tb.Th, Tb.Sp, Tb.N (Figure 3 A–E); GLCM: Energy, Entropy, Inverse Difference Moment, Inertia, Cluster Shade, Cluster Prominence, Haralick Correlation (Figure 4: A, B, D, E, F, G and H) and GLRLM: Short Run Emphasis, Grey level Non-Uniformity, Run Lenght Non-Uniformity, Low Grey Level Run Emphasis, High Grey Level Run Emphasis, Short Run Low Grey Level Emphasis, Short Run High Grey Level Emphasis and Long Run Low Grey Level Emphasis (Figure 5 A, C, D, E, F, G, H and I). On the other hand, the variables that did not presented such distinction between the sample averages (horizontal axis) were: Correlation for GLCM (Figure 4C) and for GLRLM: Long Run Emphasis and Long Run High Grey Level Emphasis (Figure 5 B and J).
Figure 3.
Bland–Altman analysis of GLCM features between BoneTexture and BoneJ applications. The normalized values were used to compare.
Figure 4.
Bland–Altman analysis of GLCM features between BoneTexture and Ibex applications. The normalized values were used to compare. GLCM,Grey-level Co-occurrence Matrix.
Figure 5.
Bland–Altman analysis of GLRLM features between BoneTexture and Ibex applications. The normalized values were used to compare. GLRLM,Grey-level Run Length matrix.
Discussion
The textural features described in this paper have been widely investigated in computed-medical research. Quantitative information extracted from volumetric images (CT and MRI) in cancer research have improved the integration of imaging, clinical and genomic features.16,29,30 In bone research, the big challenge is to discover specific imaging biomarkers that identify earlier stages and/or predict progression of bone diseases.
Caramella et al,31 demonstrated that caution should be taken when analyzing grey-level textures in CT images, mainly due to the differences in image acquisition protocols and reconstruction algorithms. The present study has standardized these variables and tested only the effects and differences among the three software applications. We focused on previously published textural features such as GLRLM and GLCM, due to their nature in identifying grey-level patterns22,23,26,27,32,33 and morphometric bone features.11,25 Our results aid the choice of a precise software to assess bone imaging markers, the selection of the best imaging bone biomarkers and evaluate the software parameters, demonstrating the potential applicability of the evaluated software as a method for future bone studies.
Recent studies by Paniagua et al,12 and Vimort et al,19 showed that GLCM and GLRLM textural features are potential diagnostic markers of TMJ OA. However, those studies focused on methodological developments of a novel software application called BoneTexture. This software is based on the implementation of ITK features, such as fast neighbourhood operators, reuse of intermediate computations, and minimization of memory use. These feature computations are n-dimensional.20 Our study showed the correlation and agreement of BoneTexture with two other software applications previously applied in the medical field: Ibex and BoneJ.16 The spearman correlation (Table 3) showed a high and significant correlation for Bone Morphometry and all the GLCM features, except for two GLRLM variables: Grey Level Non-Uniformity (p = 0.627) and Short Run Emphasis (p = 0.06). However, the negative correlation between BoneTexture and Ibex values for Correlation, Long Run Emphasis and Long Run High Grey Level Emphasis suggests that the applications might use different mathematical approach’s to obtain these values. For this reason, further studies should have caution when evaluating these biomarkers using BoneTexture or Ibex.
Since the algorithms implementation demands different computational mathematical libraries and language programming, the results obtained within software differed in terms of scale. For this reason, the results were normalized before the Bland–Altman and descriptive analyses. For the Bland–Altman plots (Figures 3–5), the closer the distribution to the horizontal median line, the better the agreement between the software. The distribution in the horizontal axis shows the variability in the sample average.27,28 This variability is expected since the sample included 66 different ROIs from human mandibular condyles.
In addition, to better understand the CI for Bland–Altman, a direct comparison with the individual confidence interval values (Table 2) makes necessary. As observed, the limits of agreement in Bland–Altman (Figure 3) for Bone morphometric features were larger than each software’s 95% CI as shown in Table 2. For GLCM features (Figure 4), the 95% CIs in Table 2 were larger than the limits of agreement in the Bland–Altman for almost all the variables tested, except for the variable Correlation. On the other hand, the limits of agreement for GLRLM features (Figure 5) presented wider deviation in comparison to the 95% CIs shown in Table 2. The six variables, with narrower CIs were: Run Length Non-Uniformity, Low Grey Level Run Emphasis, High Grey Level Run Emphasis, Short Run Low Grey Level Emphasis, Short Run High Grey Level Emphasis and Long Run Low Grey Level Emphasis. Those results suggest that the GLRLM features are more affected by the software choice. The variables, Short Run Emphasis, Long Run Emphasis, Grey Level Non-Uniformity and Long Run High Grey Level Emphasis, presented a wider CI and should be further investigated in next studies, focusing on how much the software settings and parameters could affect the results.
Additionally, Ebrahim et al,1 showed that the measurements of Bone morphometry are suitable bone imaging biomarker. They compared the histological findings with biomarkers from CBCT condyles and found that the GLRLM and GLCM features not present a strong correlation with osteoclast number. This suggests that these biomarkers are mainly involved with other bone characteristics, such as: 3D morphology and greyscale organization, instead of a direct bone resorption response.
Figure 6 displays a qualitative description of how complex the interpretation of GLCM and GLRM features could be in clinical research. The BoneTexture application was used to compute and display the three selected extreme cases from our sample. The observed results allow us to see how much each biomarker depends on the greylevel organization. A general good ability to differentiate the greyscale patterns is observed among the variables. Future studies, with larger samples for adequate power of statistical analysis, are needed to determine the applicability of these biomarkers to diagnose and/or predict progression of bone diseases in longitudinal assessments. Prior to testing the sensitivity and specificity of these surrogate biomarkers to diagnose condylar health and disease, future investigations will require radiology experts and clinical interpretation as gold standard references, as well as standardization of other image processing procedures not tested in the present study, such as segmentation and selection of the same ROI among the subjects, avoiding bias due to the ROI selection. In summary, we showed the applicability and agreement among three different software applications to assess bone textural features and morphometry; additional studies are necessary to elucidate the clinical significance of these variables.
Figure 6.
Computation of three selected extreme cases from our sample and the quantitative results for our proposed biomarkers using the BoneTexture application with the parameters described in this paper.
Conclusion
The BoneTexture application computed textural features and bone morphometry measurements with good agreement and high correlations compared to the BoneJ and Ibex applications. Further studies using HR-CBCT must be conducted to describe the applicability of these textural features to diagnose health and disease in mandibular condyles.
Footnotes
Acknowledgment: This study was partially funded by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001 and supported by NIH grants DE R01DE024450 and R21DE025306.
Contributor Information
Jonas Bianchi, Email: jonasbianchi.unesp@gmail.com.
João Roberto Gonçalves, Email: joaogonc2002@yahoo.com.br.
Antonio Carlos de Oliveira Ruellas, Email: aruellas@umich.edu.
Jean-Baptiste Vimort, Email: jb.vimort@kitware.com.
Marília Yatabe, Email: msyatabe@umich.edu.
Beatriz Paniagua, Email: beatriz.paniagua@kitware.com.
Pablo Hernandez, Email: pablo.hernandez@kitware.com.
Erika Benavides, Email: benavid@umich.edu.
Fabiana Naomi Soki, Email: fabisoki@umich.edu.
Lucia Helena Soares Cevidanes, Email: luciacev@umich.edu.
REFERENCES
- 1.Ebrahim FH, Ruellas ACO, Paniagua B, Benavides E, Jepsen K, Wolford L, et al. Accuracy of biomarkers obtained from cone beam computed tomography in assessing the internal trabecular structure of the mandibular condyle. Oral Surg Oral Med Oral Pathol Oral Radiol 2017; 124: 588–99. doi: 10.1016/j.oooo.2017.08.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Vimort J-B, Ruellas A, Prothero J, Marron JS, McCormick M, Cevidanes L, et al. Detection of bone loss via subchondral bone analysis. Proc SPIE-the Int Soc Opt Eng 2018; 10578: 25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Hochhegger B, Zanon M, Altmayer S, Pacini GS, Balbinot F, Francisco MZ, et al. Advances in imaging and automated quantification of malignant pulmonary diseases: a state-of-the-art review. Lung 2018; 196: 633–42. doi: 10.1007/s00408-018-0156-0 [DOI] [PubMed] [Google Scholar]
- 4.Johnson VL, Giuffre BM, Hunter DJ, App B, Ex S, Sc S. Osteoarthritis: what does imaging tell us about its etiology? Semin Musculoskelet Radiol 2012; 16: 410–8. doi: 10.1055/s-0032-1329894 [DOI] [PubMed] [Google Scholar]
- 5.Microarchitecture BML. The key to bone quality. Rheumatology 2009; 48(Suppl 4): iv3–8. [DOI] [PubMed] [Google Scholar]
- 6.Lewis EL, Dolwick MF, Abramowicz S, Reeder SL. Contemporary imaging of the temporomandibular joint. Dental Clinics of North America 2008; 52: 875–90. doi: 10.1016/j.cden.2008.06.001 [DOI] [PubMed] [Google Scholar]
- 7.Neogi T, Felson DT. Osteoarthritis: bone as an imaging biomarker and treatment target in oa. Nat Rev Rheumatol 2016; 12: 503–4. doi: 10.1038/nrrheum.2016.113 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Starkov P, Aguilera TA, Golden DI, Shultz DB, Trakul N, Maxim PG, et al. The use of texture-based radiomics CT analysis to predict outcomes in early-stage non-small cell lung cancer treated with stereotactic ablative radiotherapy. Br J Radiol 2019; 92: 20180228. doi: 10.1259/bjr.20180228 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Farman AG. ALARA still applies.. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2005; 100: 395–7. doi: 10.1016/j.tripleo.2005.05.055 [DOI] [PubMed] [Google Scholar]
- 10.Rehani MM, Gupta R, Bartling S, Sharp GC, Pauwels R, Berris T, et al. Radiological protection in cone beam computed tomography (CBCT). ICRP publication 129. Ann ICRP 2015; 44: 9–127. doi: 10.1177/0146645315575485 [DOI] [PubMed] [Google Scholar]
- 11.Dempster DW, Compston JE, Drezner MK, Glorieux FH, Kanis JA, Malluche H, et al. Standardized nomenclature, symbols, and units for bone histomorphometry: a 2012 update of the report of the ASBMR histomorphometry Nomenclature Committee. J Bone Miner Res 2013; 28: 2–17. doi: 10.1002/jbmr.1805 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Paniagua B, Ruellas AC, Benavides E, Marron S, Wolford L, Cevidanes L. Validation of CBCT for the computation of textural biomarkers. SPIE Med Imaging 2015; 9417: 1–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Schiffman E, Ohrbach R, Truelove E, Look J, Anderson G, Goulet J-P, et al. Diagnostic criteria for temporomandibular disorders (DC/TMD) for clinical and research applications: recommendations of the International RDC/TMD Consortium Network* and orofacial pain special interest Group†. J Oral Facial Pain Headache 2014; 28: 6–27. doi: 10.11607/jop.1151 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Murphy MK, MacBarb RF, Wong ME, Athanasiou KA. Temporomandibular disorders: a review of etiology, clinical management, and tissue engineering strategies. Int J Oral Maxillofac Implants 2013; 28: e393–414. doi: 10.11607/jomi.te20 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Castellano G, Bonilha L, Li LM, Cendes F. Texture analysis of medical images. Clinical Radiology 2004; 59: 1061–9. doi: 10.1016/j.crad.2004.07.008 [DOI] [PubMed] [Google Scholar]
- 16.Zhang L, Fried DV, Fave XJ, Hunter LA, Yang J, Court LE. Ibex: an open infrastructure software platform to facilitate collaborative work in radiomics.Ibex: an open infrastructure software platform to facilitate collaborative work in radiomics Med Phys 2015; 42: 1341–5310.1118/1.4908210.. doi: 10.1118/1.4908210 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Doube M, Kłosowski MM, Arganda-Carreras I, Cordelières FP, Dougherty RP, Jackson JS, et al. BoneJ: free and extensible bone image analysis in ImageJ. Bone 2010; 47: 1076–9. doi: 10.1016/j.bone.2010.08.023 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Paniagua B, Ruellas AC, Benavides E, Marron S, Wolford L, Cevidanes L, et al. Validation of CBCT for the computation of textural biomarkers. Proc SPIE-the Int Soc Opt Eng 2015; 94171B9417. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Vimort J-B, McCormick M, Budin F, Paniagua B. Computing textural feature maps for N-dimensional images. Insight J 2017;. [Google Scholar]
- 20. Documentation/Nightly/Extensions/BoneTextureExtension - SlicerWiki n.d.. accessed April 2, 2019.
- 21.Yushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, Gee JC, et al. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. Neuroimage 2006; 31: 1116–28. doi: 10.1016/j.neuroimage.2006.01.015 [DOI] [PubMed] [Google Scholar]
- 22.Galloway MM. Texture analysis using gray level run lengths. Computer Graphics and Image Processing 1975; 4: 172–9. doi: 10.1016/S0146-664X(75)80008-6 [DOI] [Google Scholar]
- 23.Haralick RM, Shanmugam K, Dinstein Its'Hak, Dinstein I. Textural features for image classification. IEEE Trans Syst Man Cybern 1973; SMC-3: 610–21. doi: 10.1109/TSMC.1973.4309314 [DOI] [Google Scholar]
- 24.Haralick RM, Shapiro LG. Computer and Robot Vision. Addison-Wesley Longman Publishing Co., Inc., Boston : Comput. Robot Vis. Vol. 1. Boston, MA: The British Institute of Radiology.; 1992. 459.1992, p.. [Google Scholar]
- 25.Bouxsein ML, Boyd SK, Christiansen BA, Guldberg RE, Jepsen KJ, Müller R. Guidelines for assessment of bone microstructure in rodents using micro-computed tomography. J Bone Miner Res 2010; 25: 1468–86. doi: 10.1002/jbmr.141 [DOI] [PubMed] [Google Scholar]
- 26.Dasarathy BV, Holder EB. Image characterizations based on joint gray level—run length distributions. Pattern Recognition Letters 1991; 12: 497–50210.1016/0167-8655(91)80014-2.. doi: 10.1016/0167-8655(91)80014-2 [DOI] [Google Scholar]
- 27.Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: 307–10. [PubMed] [Google Scholar]
- 28.Giavarina D. Understanding Bland Altman analysis. Biochem Med 2015; 25: 141–51. doi: 10.11613/BM.2015.015 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology 2016; 278: 563–77. doi: 10.1148/radiol.2015151169 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Nioche C, Orlhac F, Boughdad S, Reuzé S, Goya-Outi J, Robert C, et al. LIFEx: a freeware for radiomic feature calculation in multimodality imaging to accelerate advances in the characterization of tumor heterogeneity. Cancer Res 2018; 78: 4786–9. doi: 10.1158/0008-5472.CAN-18-0125 [DOI] [PubMed] [Google Scholar]
- 31.Caramella C, Allorant A, Orlhac F, Bidault F, Asselain B, Ammari S, et al. Can we trust the calculation of texture indices of CT images? a phantom study. Med Phys 2018; 45: 1529–36. doi: 10.1002/mp.12809 [DOI] [PubMed] [Google Scholar]
- 32.Amadasun M, King R. Textural features corresponding to textural properties. IEEE Trans Syst Man Cybern 1989; 19: 1264–74. doi: 10.1109/21.44046 [DOI] [Google Scholar]
- 33.Tang X. Texture information in run-length matrices. IEEE Trans. on Image Process. 1998; 7: 1602–9. doi: 10.1109/83.725367 [DOI] [PubMed] [Google Scholar]