An automatic 3D joint erosion detection method for diagnosis and monitoring of rheumatoid arthritis

The automatic 3D joint erosion detection method constructs a closed cortical surface and uses curvature estimation and spatial clustering to identify erosion regions, addressing inefficiencies in existing 2D analysis and improving the accuracy and efficiency of rheumatoid arthritis diagnosis.

WO2024228055A1PCT designated stage expired Publication Date: 2024-11-07THE HONG KONG UNIV OF SCI & TECH

Patent Information

Application Number
PCT/IB2024/000202
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-02
Filing Date
2024-04-30
Publication Date
2024-11-07

AI Technical Summary

Technical Problem

Current methods for detecting joint erosion in rheumatoid arthritis, particularly using HR-pQCT images, are inefficient due to reliance on 2D slice analysis, which fails to capture 3D coherence and often results in missed subtle lesions and variability among radiologists, and lack effective algorithms for distinguishing erosion from cortical breaks.

Method used

An automatic 3D joint erosion detection method that constructs a closed cortical surface from HR-pQCT images using modified level set-based segmentation, variational methods for joint separation, and curvature estimation, followed by spatial clustering and ellipsoid shape descriptors to identify erosion regions, reducing false positives.

Benefits of technology

This method provides superior performance in detecting joint erosions compared to existing methods, with improved sensitivity and reduced false positives, enabling more accurate and efficient diagnosis and monitoring of rheumatoid arthritis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure IB2024000202_07112024_PF_FP_ABST
    Figure IB2024000202_07112024_PF_FP_ABST
Patent Text Reader

Abstract

An automatic three-dimensional (3D) joint erosion detection method and systems for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand is provided. The method includes following steps: providing HR-pQCT images containing joint volumes as inputs; segmenting joint bones from the HR-pQCT images; separating the joint bones from joint space based on the joint segmentation; constructing a cortical surface from the joint separation; estimating surface curvature from the cortical surface; aggregating concave regions based on the surface curvature; and determining erosion regions from the concave regions. The performance of the 3D joint erosion detection method is demonstrated to be superior to the performance of the existing methods.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] DESCRIPTION TITLE AN AUTOMATIC 3D JOINT EROSION DETECTION METHOD FOR DIAGNOSIS AND MONITORING OF RHEUMATOID ARTHRITIS CROSS-REFERENCE TO RELATED APPLICATION The subject application claims the benefit of U.S. Provisional Application Serial No. 63 / 499,494, filed May 2, 2023, which is incorporated herein by reference in its entirety. BACKGROUND OF THE INVENTION Rheumatoid arthritis (RA) is a chronic autoimmune disease that damages synovial joints, leading to irreversible disability. Globally, around 70 million people suffer from RA, underscoring the urgent need for early intervention to mitigate and potentially prevent irreversible damage to the joints. One of the early indicators of RA is bone erosion in the cortical shell [1]. High-resolution peripheral quantitative computed tomography (HR-pQCT) is a new promising advancement in imaging technology. It enables quantitative studies of microarchitectures in metacarpophalangeal (MCP) joints [2] and has demonstrated great potential in erosion detection at early stages. Traditionally, clinical detection procedures are carried out slice by slice in a two- dimensional (2D) format, which can be exceedingly time-consuming. Moreover, manual detection methods often overlook subtle lesions and exhibit significant variability among radiologists. Consequently, there is a pressing need for automated detection methods to streamline the process and ensure comprehensive lesion identification. Developing automatic erosion detection algorithms presents challenges from two primary perspectives. Firstly, the erosion regions take an extremely small proportion of the whole cortical surface and are sparsely distributed on the surface, posing difficulty in gathering ample samples for training machine learning-based methods. Secondly, distinguishing joint erosion from cortical breaks requires sophisticated features due to their similarity, further complicating algorithm development. Currently, manual or semi-automatic erosion detection methods [3, 4, 5, 6] are commonly used in the clinics. Many automatic image analysis methods have also been J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg proposed to study RA erosion detection. These methods fall into the following two categories: 1. Non-deep learning based methods. Lang et al. [7], Brett et al. [8] and Murakami et al. [9] develop algorithms for conventional radiography images, which are not suitable for CT image analysis for the low resolution of conventional radiography. Peter et al.

[0010] use several basic image processing filters with fixed parameters to obtain the detection results with three-dimensional (3D) HR-pQCT volumes, but it cannot handle the complicated variations among different samples. 2. Deep-learning based methods. Ren et al.

[0011] and Folle et al.

[0012] apply popular deep-learning based methods to implement 2D slice-by-slice analysis with HR-pQCT images. However, relying solely on single 2D slices fails to capture the coherence among adjacent slices which is essential for modeling 3D joint erosion accurately. Furthermore, the above approaches may not ensure optimal performance, especially in erosion regions characterized by intricate details. BRIEF SUMMARY OF THE INVENTION There continues to be a need in the art for improved system designs and methods for erosion detection for the diagnosis and monitoring of rheumatoid arthritis. According to an embodiment of the subject invention, an automatic three-dimensional (3D) joint erosion detection method for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand is provided. The method comprises following steps: (a) providing HR-pQCT images containing joint volumes as inputs; (b) segmenting joint bones from the HR-pQCT images; (c) separating the joint bones from joint space based on the joint segmentation; (d) constructing a cortical surface from the joint separation; (e) estimating surface curvature from the cortical surface; (f) aggregating concave regions based on the surface curvature; and (g) determining erosion regions from the concave regions. The step (b) may comprise segmenting the HR-pQCT images by a segmentation method. The segmentation method may be a modified level set-based segmentation method. The modified level set-based segmentation method may comprise performing an initialization-robust modification. The step (c) may comprise separating the joint bones by a variational method. The variational method may be a method of finding the separating regular plane of the joint segmentation. The step (d) may comprise constructing the cortical surface from the joint separation by a deformation model. The deformation model may be a modified active contour J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg model. The modified active contour model may comprise performing a negative-curvature sensitive modification. The step (e) may comprise applying a curvature estimation method to the cortical surface. The curvature estimation method may be a modified isosurface-based curvature estimation method. The modified isosurface-based curvature estimation method may comprise performing a scale selection modification. The step (f) may comprise applying the surface curvature to aggregate concave regions with a spatial clustering method. Moreover, the spatial clustering method is a modified connected region decomposition. The modified connected region decomposition comprises performing a neighborhood-enlarging modification. The step (g) may comprise selecting erosion regions by a shape descriptor. The shape descriptor is a modified ellipsoid shape descriptor. The modified ellipsoid shape descriptor is based on an axis length ratio-based modification. According to certain embodiments of the subject invention, an automatic three- dimensional (3D) joint erosion detection method for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand is provided. The method comprises constructing a closed cortical surface; and performing an erosion detection. The constructing a closed cortical surface comprises MCP volume cropping, initialization-robust region competition level set segmentation, joint bone separation by finding the regular surface across the joint space, and surface gap filling. The performing erosion detection comprises surface curvature estimation, spatial cluster along with ellipsoid representation of erosions, and false positive reduction by thresholding the ratio of max–min axes of ellipsoids with erosion shape refinement. According to some embodiment of the subject invention, the non-transitory computer readable medium having stored therein program instructions executable by a computing system to cause the computing system to perform an automatic three-dimensional (3D) joint erosion detection method for the diagnosis and monitoring of rheumatoid arthritis using HR- pQCT images of a hand is provided. The method comprises constructing a closed cortical surface; and performing an erosion detection. The construction of a closed cortical surface comprises MCP volume cropping, initialization-robust region competition level set segmentation, joint bone separation by finding the regular surface across the joint space, and surface gap filling. The performing erosion detection comprises surface curvature estimation, spatial cluster along with the ellipsoid representation of erosions, and false positive reduction by thresholding the ratio of max–min axes of ellipsoids with erosion shape refinement. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 provides an overview of the 3D joint erosion detection method, wherein the upper panel describes the construction of closed cortical surface, including MCP volume cropping, initialization-robust region competition level set segmentation, joint bone separation by finding the regular surface across the joint space, and surface gap filling; and wherein the lower panel describes the erosion detection part, comprising surface curvature estimation, spatial cluster along with ellipsoid representation of erosions, and false positive reduction by thresholding the ratio of max–min axes of ellipsoids with erosion shape refinement, according to an embodiment of the subject invention. FIGS.2A-2D illustrate the performance comparison between the cortical surface construction pipeline and the deep learning segmentation methods, indicating that the deep learning-based methods do not perform well on the important potential erosion regions, which is displayed in the axial views, wherein FIGS.2A-2D show the performance of the 2D U-net method, the 3D U-net method, the automatic three-dimensional (3D) joint erosion detection method of the subject invention, and the Ground Truth method performed on a same object , respectively, according to an embodiment of the subject invention. FIG. 3 illustrates two examples of erosion detection results compared with manual annotation results, wherein different erosions are shown by different colors, and wherein in each pair, the left one is the manual annotation result and the right one is the detection result of the subject invention, according to an embodiment of the subject invention. FIGs. 4A-4B illustrates P–R curves of the algorithm of the subject invention on: P-R curve of classification performance between healthy joints and eroded joints as shown in FIG. 4A, and P-R curve of location detection of erosions as shown in FIG. 4B, wherein the Area-Under-Curve (AUC) of eroded joint detection and erosion detection are 0.8604 and 0.5148, respectively, the sensitivities of eroded joint detection and erosion detection are 0.9714 and 0.8889, respectively, according to an embodiment of the subject invention. Table 1: Default parameter setting of the automatic 3D joint erosion detection method of the subject invention 1. Parameters 2. Values 3. Time step in 4.1 5. Coefficients (µ1, 6. (6, 0.2) J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg DETAILED DISCLOSURE OF THE INVENTION The embodiments of the subject invention pertain to an automatic three-dimensional (3D) joint erosion detection method and systems for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term “and / or” includes any and all combinations of one or more of the associated listed items. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well as the singular forms, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” and / or “comprising,” when used in this specification, specify the presence of stated features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one having ordinary skill in the J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg art to which this invention pertains. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein. When the term “about” is used herein, in conjunction with a numerical value, it is understood that the value can be in a range of 90% of the value to 110% of the value, i.e. the value can be + / - 10% of the stated value. For example, “about 1 kg” means from 0.90 kg to 1.1 kg. In describing the invention, it will be understood that a number of techniques and steps are disclosed. Each of these has individual benefits and each can also be used in conjunction with one or more, or in some cases all, of the other disclosed techniques. Accordingly, for the sake of clarity, this description will refrain from repeating every possible combination of the individual steps in an unnecessary fashion. Nevertheless, the specification and claims should be read with the understanding that such combinations are entirely within the scope of the invention and the claims. Rheumatoid arthritis (RA) is a chronic inflammatory disease, and high-resolution peripheral quantitative computed tomography (HR-pQCT) stands out as a promising tool for diagnosing and tracking disease progression in RA. Despite its potential, an automatic erosion detection tool utilizing HR-pQCT images has yet to be developed. An automatic 3D joint erosion detection method is provided for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand. The erosion detection method comprises two steps: constructing a closed surface of the cortical layer from an HR- pQCT volume image, and then detecting erosion regions using surface curvature information. In the first step, the segmentation of a joint bone whose boundary aligns well with the cortical layer is obtained, and then the gaps along the bone surface are filled. In the second step, the voxel-wise concavity and region-wise concavity are hierarchically used to characterize erosions and use ellipsoid shape indicator to detect the most possible erosions. The parameter of ellipsoid shape indicator in the method of the subject invention is adjustable so that physicians can control the false positive rate according to individual preference. In one embodiment of the subject invention, the automatic 3D joint erosion detection method and systems for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand comprises constructing a closed cortical surface and performing erosion J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg detection. In particular, the construction of a closed cortical surface comprises MCP volume cropping, initialization-robust region competition level set segmentation, joint bone separation by finding the regular surface across the joint space, and surface gap filling. The performing erosion detection comprises surface curvature estimation, spatial cluster along with the ellipsoid representation of erosions, and false positive reduction by thresholding the ratio of max–min axes of ellipsoids with erosion shape refinement. Automatic Construction of Cortical Surface Following initialization-robust region competition level set method is utilized to firstly segment the cortical layers from a raw volume: Concretely, the foreground and background probability density functions (PDFs) read as follows: where m, a and b are the location, shape and scale parameters of the Gamma distribution, respectively; and ^^ and ^^ଶare the mean and variance of the Gaussian distribution, respectively. During each iteration, first, ( ^^, ^^, ^^) and ( ^^, ^^) are estimated with maximum likelihood estimation (MLE) in form of smoothed Heaviside function, and then update ^^ with the updated PDF parameters. In the Gamma part, the location parameter m is first estimated. For the estimation of a and b, the MLE is adopted as follows: J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg Parameter estimation of Gaussian distribution in the level set framework reads To fill in the discontinuity arising from small cortical breaks and under-segmentation, while preserving the geometrical structures of the cortical surface such as joint space and concave regions, the lower and upper joints are initially separated in order to preserve the joint space by optimizing the following variational function: where ^^(∙). is the 3D segmentation mask by binarizing the level set of (1). The functional derivative of (6) can be calculated as: whereడ^డ௭|௭ୀ^denotes the partial derivative of ^^ w.r.t z dimension and takes ^^ = ℎ. When h is not an integer, the linear interpolation is used to achieve the approximation ofப^ ^ୀ୦. Then variational morphological operators are used to fill all the discontinuities on the surface. To recover the buried concave regions, a concave region sensitive active contour model (ACM) as follows is further provided: J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg where ^^ > 0 is a predefined parameter indicating the curvature range of concave regions. 3D Automatic Joint Erosion Detection Surface curvature estimation becomes essential when utilizing surface concavity to indicate possible joint erosion. Two principal curvature ^^^, ^^ ^^ ^^ ^^ଶare first estimated. A point is concave if its Gaussian curvature K and mean curvature H satisfy ^^ = ^^^^^ଶ> 0 and ^^ =^భା^మ< 0. Furthermore, due to resoluti^ଶ on restriction, both of the curvature radii ^^^=|^భ|and ^^ଶ=^|^మ|should be larger than 1. In a nutshell, the curvature of the concave candidate erosion points should satisfy: Given a segmentation volume V and its corresponding surface S, the isosurface is derived from the signed distance map of V. To apply differential operators to the digital volume under certain scale priors, a 3D isotropic Gaussian kernel ^^ఙ( ^^, ^^, ^^) = is used to convert the isosurface ^^ into ^^^space = ^^ ∗ ^^ where ^^ involves the predefined scale priors of the target objects. The curvatures estimation with the gradient and Hassian matrix of ϕ^is in the following steps: 1. Given gradient = ( ^^ ^^ , ^^ ^^ , ^^ ఙ ௬ ఙ ௭^^ఙ) , compute its normal vector ^^^ = − ^^ = ^^ − ^^^ ^^^், where ^^ is the identity matrix. 2. Calculate the Hessian matrix ^^ of ^^ఙas Because ^^ ^^ ^^ ^^( ^^) = 2 , and the two non-zero eigenvalues are the two principal ures, the Gaussian Curvature reads ^^ =(் మ మcurvat^(ீ)) ି|ீ|ಷ ்^(ீ)ଶ and mean curvature is ^^ =ଶ, where ^^ ^^ is the trace and | ∙ |ிis the Frobenius norm. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg Now it is required to aggregate the concave voxels into disjoints regions according to the spatial proximity, as outlined in Algorithm 1. Algorithm 1 Spatial clustering for aggregating erosions Require: Binary volume E for the erosion volume; minimum distance among erosions r; Ensure: Clustering mapping: ⊂ ^^; cluster number ^^ 1: Initialize ^^( ^^) ← 0 ∀∈ Ω, cluster number ^^ ← 0; 2: for each ^^ ∈ Ω with ^^( ^^) = 1 do 3: if f(x) = 0 then 4: then 5: ^^( ^^) = 1 6: else 7: ^^( ^^) ← mode of ^^( ^^) ∀ ^^ ∈ ^^^( ^^) 8: end if 9: end if 10: end for 11: return ^^and K. Given a cluster ^^ of voxels indicating candidate erosion, the minimal outer ellipsoid of ^^, named Lowner-John ellipsoid is used to represent the erosion as an entirety. Suppose the non-degenerate Lowner-John ellipsoid of 3D point set C parameterized as ℇ^: = ^ ^^|( ^^ − ^^)்^^ି^( ^^ − ^^) ≤ 1^, where ^^ is some 3 × 3 positive-definit matrix and ^^ ∈ ℝଷ, it is modeled as the following convex optimization problem and it is solved according to

[0014] : where ^^ ^^ ^^(∙) denotes the determinant. Then the square root of three eigenvalues of ^^, length of three axes of the ellipsoid, as shape indicator are used to distinguish the true erosion from the false positives, where is treated as the maximal curvature radium and as the minimal curvature radium, and their ratio ^^: = can serve as a size-invariant measure for shape measurement. The ratio k is adopted to exclude those obvious false positives, because natural concave structures shared by joints are often J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg extremely flat in ellipsoid representation, corresponding to relatively large ^^, whereas the ellipsoid of the true erosion tends to have a relatively small ^^ , indicating those those significant locally-bending regions. In the detection method of the subject invention, a higher threshold ^^ will lead to higher sensitivity at the cost of higher false positive rate. At the last step, the erosion regions are refined and the other regions are kept unchanged. The closed joint volumes and the detected ellipsoid erosions are denoted as ^^ and^^, respectively. Then, the volume ^ത^ = ^^ ∪ ^^ is obtained. The objective is to get the volumewithout erosion J by minimizing the following energy functional: where ^^^is the signed distance function of ^^, ^^^^^is the regularization term in (4), and ^^ଶare hyper-parameters. With the modified joint volume J, the refined RA erosion regions can be obtained by applying connected region decomposition to ^^ − ^^. The automatic 3D joint erosion detection method using HR-pQCT images for early diagnosis and monitoring of RA is provided, comprising two main steps: constructing closed cortical surface and detecting erosion regions on the surface. In the first step of constructing closed cortical surface, an initialization-robust region competition method for joint segmentation is performed, and then the surface gaps are filled by using joint bone separation and curvature-based surface alignment. In the second step of detecting erosion regions on the surface, the curvature information of each voxel is estimated and then aggregate the candidate voxels into concave surface regions and the shape information of the regions is used to detect erosions. Experiments show that performance of the method of the subject invention is superior to the performance of the existing methods. Embodiment 1. An automatic 3D joint erosion detection for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand, the method comprising: (a) providing HR-pQCT images containing joint volumes as inputs; (b) segmenting joint bones from the HR-pQCT images; (c) separating the joint bones from joint space based on the joint segmentation; (d) constructing a cortical surface from the joint separation; (e) estimating surface curvature from the cortical surface; (f) aggregating concave regions based on the surface curvature; and (g) determining erosion regions from the concave regions. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg Embodiment 2. The method of embodiment 1, wherein the step (b) comprises segmenting the HR-pQCT images by a segmentation method. Embodiment 3. The method of embodiment 2, wherein the segmentation method is a modified level set-based segmentation method. Embodiment 4. The method of embodiment 3, wherein the modified level set-based segmentation method comprises performing an initialization-robust modification. Embodiment 5. The method of embodiment 1, wherein the step (c) comprises separating the joint bones by a variational method. Embodiment 6. The method of embodiment 5, wherein the variational method is a method of finding the separating regular plane of the joint segmentation. Embodiment 7. The method of embodiment 1, wherein the step (d) comprises constructing the cortical surface from the joint separation by a deformation model. Embodiment 8. The method of embodiment 7, wherein the deformation model is a modified active contour model. Embodiment 9. The method of embodiment 8, wherein the modified active contour model comprises performing a negative-curvature sensitive modification. Embodiment 10. The method of embodiment 1, wherein the step (e) comprises applying a curvature estimation method to the cortical surface. Embodiment 11. The method of embodiment 10, wherein the curvature estimation method is a modified isosurface-based curvature estimation method. Embodiment 12. The method of embodiment 11, wherein the modified isosurface- based curvature estimation method comprises performing a scale selection modification. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg Embodiment 13. The method of embodiment 1, wherein the step (f) comprises applying the surface curvature to aggregate concave regions with a spatial clustering method. Embodiment 14. The method of embodiment 13, wherein the spatial clustering method is a modified connected region decomposition. Embodiment 15. The method of embodiment 14, wherein the modified connected region decomposition comprises performing a neighborhood-enlarging modification. Embodiment 16. The method of embodiment 1, wherein the step (g) comprises selecting erosion regions by a shape descriptor. Embodiment 17. The method of embodiment 16, wherein the shape descriptor is a modified ellipsoid shape descriptor. Embodiment 18. The method of embodiment 17, wherein the modified ellipsoid shape descriptor is based on an axis length ratio-based modification. Embodiment 19. An automatic three-dimensional (3D) joint erosion detection method for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand, the method comprising: constructing a closed cortical surface; and performing an erosion detection; wherein the constructing a closed cortical surface comprises MCP volume cropping, initialization-robust region competition level set segmentation, joint bone separation by finding the regular surface across the joint space, and surface gap filling, and wherein the performing erosion detection comprises surface curvature estimation, spatial cluster along with ellipsoid representation of erosions, and false positive reduction by thresholding the ratio of max–min axes of ellipsoids with erosion shape refinement. Embodiment 20. A non-transitory computer readable medium having stored therein program instructions executable by a computing system to cause the computing system to J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg perform an automatic three-dimensional (3D) joint erosion detection method for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand, the method comprising: constructing a closed cortical surface; and performing an erosion detection; wherein the constructing a closed cortical surface comprises MCP volume cropping, initialization-robust region competition level set segmentation, joint bone separation by finding the regular surface across the joint space, and surface gap filling, and wherein the performing erosion detection comprises surface curvature estimation, spatial cluster along with ellipsoid representation of erosions, and false positive reduction by thresholding the ratio of max–min axes of ellipsoids with erosion shape refinement. All patents, patent applications, provisional applications, and publications referred to or cited herein are incorporated by reference in their entirety, including all figures and tables, to the extent they are not inconsistent with the explicit teachings of this specification. It should be understood that the examples and embodiments described herein are for illustrative purposes only and that various modifications or changes in light thereof will be suggested to persons skilled in the art and are to be included within the spirit and purview of this application and the scope of the appended claims. In addition, any elements or limitations of any invention or embodiment thereof disclosed herein can be combined with any and / or all other elements or limitations (individually or in any combination) or any other invention or embodiment thereof disclosed herein, and all such combinations are contemplated with the scope of the invention without limitation thereto. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg REFERENCES [1] Christian M Stach, Michael B¨auerle, Matthias Englbrecht, Gerhard Kronke, Klaus Engelke, Bern- hard Manger, and Georg Schett. Periarticular bone structure in rheumatoid arthritis patients and healthy individuals assessed by high-resolution computed tomography. Arthritis & Rheumatism: Official Journal of the American College of Rheumatology, 62(2):330–339, 2010. [2] Angela M Cheung, Jonathan D Adachi, David A Hanley, David L Kendler, K Shawn Davison, Robert Josse, Jacques P Brown, Louis-Georges Ste-Marie, Richard Kremer, Marta C Erlandson, et al. High- resolution peripheral quantitative computed tomography for the assessment of bone strength and structure: a review by the canadian bone strength working group. Current Osteoporosis Reports, 11(2):136–146, 2013. [3] Dominique T¨opfer, Stephanie Finzel, Oleg Museyko, Georg Schett, and Klaus Engelke. Segmentation and quantification of bone erosions in high-resolution peripheral quantitative computed tomography datasets of the metacarpophalangeal joints of patients with rheumatoid arthritis. Rheumatology, 53(1):65–71, 2014. [4] Dominique T¨opfer, Bastian Gerner, Stephanie Finzel, Sebastian Kraus, Oleg Museyko, Georg Schett, and Klaus Engelke. Automated three-dimensional registration of high- resolution peripheral quan- titative computed tomography data to quantify size and shape changes of arthritic bone erosions. Rheumatology, 54(12):2171–2180, 2015. [5] Andrea Scharmga, Michiel Peters, Joop P van den Bergh, Piet Geusens, Daan Loeffen, Bert van Rietbergen, Thea Schoonbrood, Debby Vosse, Ren´e Weijers, and Astrid van Tubergen. Development of a scoring method to visually score cortical interruptions on high-resolution peripheral quantitative computed tomography in rheumatoid arthritis and healthy controls. PloS One, 13(7):e0200331, 2018. [6] Anne Fouque-Aubert, St´ephanie Boutroy, Hubert Marotte, Nicolas Vilayphiou, Justine Bacchetta, Pierre Miossec, Pierre Dominique Delmas, and Roland D Chapurlat. Assessment of hand bone loss in rheumatoid arthritis by high-resolution peripheral quantitative ct. Annals of the Rheumatic Diseases, 69(9):1671–1676, 2010. [7] Georg Langs, Philipp Peloschek, Horst Bischof, and Franz Kainberger. Automatic quantification of joint space narrowing and erosions in rheumatoid arthritis. IEEE Transactions on Medical Imaging, 28(1):151–164, 2008. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg [8] Alan Brett, Jane Haslam, and Anthony Holmes. Computer program products and methods for detection and tracking of rheumatoid arthritis, February 28 2012. US Patent 8,126,242. [9] Seiichi Murakami, Kazuhiro Hatano, JooKooi Tan, Hyoungseop Kim, and Takatoshi Aoki. Auto- matic identification of bone erosions in rheumatoid arthritis from hand radiographs based on deep convolutional neural network. Multimedia Tools and Applications, 77(9):10921–10937, 2018.

[0010] Michiel Peters, A Scharmga, J De Jong, A Van Tubergen, PPMM Geusens, JJ Arts, D Loeffen, R Weijers, B Van Rietbergen, and J Van Den Bergh. An automated algorithm for the detection of cortical interruptions on high resolution peripheral quantitative computed tomography images of finger joints. PLoS One, 12(4):e0175829, 2017.

[0011] Jintao Ren, Arash Moaddel, Ellen M Hauge, Kresten K Keller, Rasmus K Jensen, and Fran¸cois Lauze. Automatic detection and localization of bone erosion in hand hr- pqct. In Medical Imaging 2019: Computer-Aided Diagnosis, pages 524 – 532, 2019.

[0012] Lukas Folle, Timo Meinderink, David Simon, Anna-Maria Liphardt, Gerhard Kr¨onke, Georg Schett, Arnd Kleyer, and Andreas Maier. Deep learning methods allow fully automated segmentation of metacarpal bones to quantify volumetric bone mineral density. Scientific Reports, 11(1):9697, 2021.

[0013] Manfredo P Do Carmo. Differential Geometry of Curves and Surfaces. Courier Dover Publications, 2016.

[0014] Stephen Boyd, Stephen P Boyd, and Lieven Vandenberghe. Convex Optimization. Cambridge university press, 2004. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg

Claims

CLAIMS We claim:

1. An automatic three-dimensional (3D) joint erosion detection method for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand, the method comprising: (a) providing HR-pQCT images containing joint volumes as inputs; (b) segmenting joint bones from the HR-pQCT images; (c) separating the joint bones from joint space based on the joint segmentation; (d) constructing a cortical surface from the joint separation; (e) estimating surface curvature from the cortical surface; (f) aggregating concave regions based on the surface curvature; and (g) determining erosion regions from the concave regions.

2. The method of claim 1, wherein the step (b) comprises segmenting the HR-pQCT images by a segmentation method.

3. The method of claim 2, wherein the segmentation method is a modified level set- based segmentation method.

4. The method of claim 3, wherein the modified level set-based segmentation method comprises performing an initialization-robust modification.

5. The method of claim 1, wherein the step (c) comprises separating the joint bones by a variational method.

6. The method of claim 5, wherein the variational method is a method of finding the separating regular plane of the joint segmentation.

7. The method of claim 1, wherein the step (d) comprises constructing the cortical surface from the joint separation by a deformation model. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg8. The method of claim 7, wherein the deformation model is a modified active contour model.

9. The method of claim 8, wherein the modified active contour model comprises performing a negative-curvature sensitive modification.

10. The method of claim 1, wherein the step (e) comprises applying a curvature estimation method to the cortical surface.

11. The method of claim 10, wherein the curvature estimation method is a modified isosurface-based curvature estimation method.

12. The method of claim 11, wherein the modified isosurface-based curvature estimation method comprises performing a scale selection modification.

13. The method of claim 1, wherein the step (f) comprises applying the surface curvature to aggregate concave regions with a spatial clustering method.

14. The method of claim 13, wherein the spatial clustering method is a modified connected region decomposition.

15. The method of claim 14, wherein the modified connected region decomposition comprises performing a neighborhood-enlarging modification.

16. The method of claim 1, wherein the step (g) comprises selecting erosion regions by a shape descriptor.

17. The method of claim 16, wherein the shape descriptor is a modified ellipsoid shape descriptor.

18. The method of claim 17, wherein the modified ellipsoid shape descriptor is based on an axis length ratio-based modification. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg19. An automatic three-dimensional (3D) joint erosion detection method for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand, the method comprising: constructing a closed cortical surface; and performing an erosion detection; wherein the constructing a closed cortical surface comprises MCP volume cropping, initialization-robust region competition level set segmentation, joint bone separation by finding the regular surface across the joint space, and surface gap filling, and wherein the performing erosion detection comprises surface curvature estimation, spatial cluster along with ellipsoid representation of erosions, and false positive reduction by thresholding the ratio of max–min axes of ellipsoids with erosion shape refinement.

20. A non-transitory computer readable medium having stored therein program instructions executable by a computing system to cause the computing system to perform an automatic three-dimensional (3D) joint erosion detection method for the diagnosis and monitoring of rheumatoid arthritis using HR-pQCT images of a hand, the method comprising: constructing a closed cortical surface; and performing an erosion detection; wherein the constructing a closed cortical surface comprises MCP volume cropping, initialization-robust region competition level set segmentation, joint bone separation by finding the regular surface across the joint space, and surface gap filling, and wherein the performing erosion detection comprises surface curvature estimation, spatial cluster along with ellipsoid representation of erosions, and false positive reduction by thresholding the ratio of max–min axes of ellipsoids with erosion shape refinement. J:\HKUS\193XPCT\Application\HKUS-193XPCT-Application.docx / brb / jyg

Citation Information

Patent Citations

  • Method and system for ultrasound based automated detection, quantification and tracking of pathologies

    CN102973293A

  • Image-based method to measure joint deformity

    US20180263586A1

  • System and method for quantitative assessment of joint diseases and the change over time of joint diseases

    WO2003012724A1

Cited By

  • Bone joint disease artificial intelligence auxiliary diagnosis and treatment system based on Internet hospital

    CN120977553A