Method for constructing a mathematical biomarker for detecting osteoporosis
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- UNIVERISTE PARIS CITE
- Filing Date
- 2026-01-30
- Publication Date
- 2026-08-06
Smart Images
Figure FR2026000020_06082026_PF_FP_ABST
Abstract
Description
[0001] Description
[0002] Title of the invention: Method for constructing a mathematical biomarker for the detection of osteoporosis
[0003] [1] The technical context of the present invention is that of osteoporosis prevention. More particularly, the invention relates to a method for constructing a mathematical biomarker for the detection, prognosis and monitoring of osteoporosis.
[0004] [2] Certain severe osteoporotic fractures, such as those of the upper end of the femur or vertebrae, lead to serious consequences and are responsible for dependency and increased mortality. Fracture prevention is therefore a major public health issue.
[0005] [3] To this end, bone indices can already be determined from panoramic radiographs to try to predict the risk of osteoporosis. The majority of scientific studies on the subject evaluate the link between bone mineral density or osteoporosis status and radiological indices. These are primarily cross-sectional studies, often limited by the choice and size of their population, the relevance of the clinical information collected, or the small sample size.
[0006] [4] In particular, there are known methods for diagnosing osteoporosis which combine clinical elements, such as loss of height, metabolic bone pathology, consumption of drugs inducing bone loss, ..., biomarkers, such as for example a FRAX® score index or Trabecular Bone Score (TBS), and a measurement of bone mineral density (BMD) by dual-energy X-ray absorptiometry.
[0007] [5] However, nearly 30% of patients with severe fractures did not have bone densitometric osteoporosis. In particular, half of all fractures of the proximal femur occur in the general population without a prior history of fracture, thus justifying early detection during non-specific examinations.
[0008] [6] In humans, studies have also found a positive correlation between peripheral bone mineral density and mandibular bone density. Panoramic radiographic examinations are routinely performed by radiologists and dentists. This examination is accessible, simple, and inexpensive. Screening for osteoporosis based on panoramic radiographs can thus allow for rapid patient referral and early intervention to reduce the risk of fracture. In this context, bone indices assessed from panoramic radiographs are used to try to predict the risk of osteoporosis and monitor its progression.
[0009] [7] However, the use of these indices suffers from two weaknesses: first, they are difficult to interpret; second, they have been constructed and evaluated on populations that are too small or not relevant for the implementation of population monitoring.
[0010] [8] Also known is document US20230263463 A1, which describes an apparatus for measuring the thickness, roughness, and morphological index of mandibular cortical bone using a panoramic dental image to aid in the diagnosis of osteoporosis. This document describes how the thickness, roughness, and morphological index of cortical bone are measured more accurately, and how the diagnosis of osteoporosis can be supported more precisely. The diagnostic aid includes a unit for extracting a mandibular contour from an image of mandibular cortical bone, in order to define line segments of the mandibular cortical bone image along which the cortical bone thickness is calculated.
[0011] [9] The drawback of the device described in US20230263463 is that it defines reference points—that is, fixed points—located on either side of the chin. The algorithm then "counts" the pixels from these reference points. Therefore, the method described in document US20230263463 only works if the algorithm is able to locate the reference points on the image of the mandible: the method only works on mandibles in good condition and taken from a predetermined viewing angle.
[0012]
[0010] In general, the method described in US20230263463 makes numerous assumptions about a typical mandibular morphology and the expected variation in bone density from a lower border of the mandible. Therefore, the method described in this document might not function correctly, for example, in the case of a severely damaged lower mandible. Furthermore, the method described in US20230263463 employs numerous digital filters to "process" the image and construct the osteoporosis indicator. However, the implementation of such filters raises many questions regarding their usefulness for detecting osteoporosis and their reproducibility.
[0013]
[0011] The present invention aims to propose a new method for constructing a mathematical biomarker for the detection of osteoporosis, particularly early osteoporosis, in order to address at least in large part the previous problems and to lead to other advantages.
[0014]
[0012] Another objective of the invention is to be able to use such a biomarker with a large number of mandibular radiographic images, freeing oneself from the biases introduced by the positioning of the patient's head.
[0015]
[0013] Another object of the invention is to improve the accuracy and versatility of such a biomarker by also offering easier interpretation thereof.
[0016]
[0014] The method of the invention can be applied to the regular monitoring of osteoporotic patients receiving treatment with bisphosphonates, in order to evaluate the effectiveness of the treatment on bone density and the evolution of osteoporosis by monitoring the value of the biomarker.
[0017]
[0015] According to a first aspect of the invention, at least one of the aforementioned objectives is achieved with a method for constructing a mathematical biomarker for the detection of osteoporosis, the mathematical biomarker being representative of the bone density of a person's mandible and obtained from at least one panoramic radiographic image of the mandible, the method comprising the following steps:
[0018]
[0016] - a localization step of a lower border of the mandible on at least one image, called digitized lower border;
[0019]
[0017] - a step of defining a sequence of normal vectors, each normal vector being associated with a gray level value of a plurality of pixels extending along a direction of the normal vector from the digitized lower edge of the mandible, taken at predetermined intervals from said lower edge;
[0018] - a step of determining the largest subsets of homogeneous or substantially homogeneous gray levels extending along the mandible, across the set of normal vectors defined previously and for each predetermined interval taken at a distance from the digitized lower edge of the mandible;
[0020]
[0019] - a step of calculating indicators representative of grey level distributions representative of bone density and which corresponds to the biomarker, the grey level distributions being calculated for the largest subsets for each predetermined interval.
[0021]
[0020] The method according to the first aspect of the invention is implemented by a control unit. In the context of the present invention, the control unit comprises a microcontroller and / or a printed circuit board and / or a microprocessor. Optionally, the control unit may also include memory. The control unit may be, for example, a computer or a remote server. The advantage of the method according to the invention is that it can be implemented asynchronously with respect to the acquisition of the panoramic photographs.
[0022]
[0021] In the context of the present invention, the mathematical biomarker for the detection of osteoporosis is therefore of the type of a mathematical tensor.
[0023]
[0022] In the context of the present invention, the digitized lower edge localization step makes it possible to determine a starting direction for the application of the mathematical biomarker. In particular, the objective here is to be able to distinguish the lower mandible from the background of the image.
[0024]
[0023] In the context of the present invention, the normal vector sequence definition step determines a succession of gray-level values—representative of bone density from the radiographic image—along a particular direction extending from the previously determined lower edge. In other words, the definition step discretizes the radiographic image into a plurality of normal vectors representative of the mandibular bone density; that is, each normal vector is associated with several values, all of which provide an indication of the mandibular bone density in its direction at a given point on the mandibular edge. In the context of the present invention, the term "normal vector sequence" or "normal vector bundle" refers to the definition step, which aims to define a plurality of normal vectors.
[0025]
[0024] In the context of the present invention, the determination step aims to interpret the gray level values in the direction of the various normal vectors taken from the edge of the mandible. By considering all the values taken at a fixed distance from the edge, the determination step makes it possible to define "level lines" that extend parallel or substantially parallel to the lower edge of the mandible. Thus, in the determination step, the largest determined subsets are taken parallel or substantially parallel to the lower edge of the mandible. By "homogeneous or substantially homogeneous" it is understood that we are looking for gray level values that are equal or substantially equal for adjacent normal vectors at positions equidistant from the lower edge of the mandible.
[0026]
[0025] In the context of the present invention, the at least one image from which the method according to the invention applies is of the type of a radiographic image of a human mandible, in frontal and panoramic view, so as to see on said image the teeth, as well as, in particular, the mandible and the maxilla.
[0027]
[0026] Thus, the distribution of these homogeneous grayscale values yields a denoised grayscale value towards the interior of the mandible. This value distribution thus forms, in a particularly ingenious way, an excellent qualitative indicator of the health status of the lower mandible, by characterizing the variation in bone density within the lower mandible. By extension, this distribution forms a relevant biomarker for detecting osteoporosis in the subject wearing the imaged mandible, the precise interpretation and characterization of which must then be measured and interpreted by a healthcare professional, not provided for by the present invention. In other words, the shape of the distribution thus obtained, towards the interior of the mandible, makes it possible to obtain geometries representative of the state of the mandibular bone and offers a biomarker for the detection, prediction, and monitoring of osteoporosis.
[0027] The method according to the first aspect of the invention resolves the technical problems mentioned above by allowing the biomarker to be used with a large number of mandibular radiographs, eliminating the biases introduced by the patient's head position: the definition and determination steps allow the method to be implemented without relying on rigid assumptions about the images, the shape of the lower mandible, or the patient's head position. Consequently, the method according to the invention makes it possible to offer such a biomarker that is more precise and suitable for a wider range of situations.
[0028]
[0028] The method according to the first aspect of the invention advantageously comprises at least one of the improvements below, the technical characteristics forming these improvements being able to be taken alone or in combination:
[0029]
[0029] - the localization step includes a localization step of a lower border of the mandible on the right and left parts of at least one image. In the context of the present invention, the right and left parts of at least one image are considered from a median vertical axis of said image. These "left" and "right" borders are referred to as the digitized lower border;
[0030]
[0030] - the digitized lower edge localization step further includes a step of smoothing said digitized lower edge;
[0031]
[0031] - The definition step includes a step for defining a sequence of normal vectors oriented inwards towards the head along the smoothed digitized lower edges, each normal vector defining a semi-axis (or a semi-segment) oriented inwards towards the jaw, one closed end of which is located on the digitized lower edge and one open part of which propagates inwards towards the head. Each normal vector thus first traverses the mandible. Subsequently, along each normal vector, gray level values of the pixels it overlaps and their neighboring pixels are associated, at a distance of 2, 4, or 8 neighboring pixels. Then, at a regular step denoted a, successive distances of a, 2a, 3a,... are formed.From the normal vector, an associated grey level value is calculated from the pixel containing the point at the considered distance and its neighboring pixels, so as to define a grey level curve perpendicular to the edge of the mandible calculated at different successive distances, taken according to the step a. This advantageous configuration then allows the construction of a grey level curve parallel to the edge at distance a, 2a, 3a, ... which connects the points of the previous curves at this distance when the normal vector is moved along the edge;
[0032]
[0032] - The step of determining the largest homogeneous or substantially homogeneous grayscale subsets is preferentially performed on each of the grayscale curves parallel to the edge of the mandible, at distances a, 2a, 3a, ... from it. These largest homogeneous grayscale subsets on the curves at a fixed distance from the edge of the mandible allow the calculation of indicators representative of the grayscale distribution of the pixels they overlap. For example, an indicator representative of the grayscale distribution consists of the set of grayscale levels of the overlapping pixels or their average, possibly calculated using the neighbors of the pixels. As an output of the determination step, the method according to the invention provides a set of indicators that respect the geometry of the mandible and that characterize the grayscale levels above it.These indicators are preferably provided in the form of tensors which can then be easily used in common statistical learning methods and ultimately allow the construction of a biomarker within the meaning of the invention;
[0033]
[0033] - in the normal vector definition step, the gray level values are determined at different distances calculated along each direction. Preferably, the gray level values are calculated at constant intervals along each direction, for example, for an interval equivalent to one pixel of the mandible image or at constant intervals of pixels along each direction. This advantageous configuration makes it possible to normalize the discretization of the gray level information contained in the radiographic image and transferred to the normal vectors;
[0034]
[0034] - furthermore, the grey level value for each element of each normal vector thus defined is either equal to the value of a pixel of the mandible image taken at the considered position of the image along the direction of said normal vector, or equal to a statistical quantity representative of several pixels considered at the considered position of the image along the direction of said normal vector, such as, for example, an average value of the grey levels of said pixels. This advantageous configuration makes it possible to propose a representative and more robust statistical value of the local grey levels of the radiographic image, thereby reducing the dependence on noise levels or other artifacts present on the radiographic image;
[0035]
[0035] - during the normal vector definition step, each direction is preferably defined perpendicular to the digitized lower edge of the mandible. Even more preferably, each direction is perpendicular or substantially perpendicular to the digitized lower edge of the mandible from which it is generated. The directions of the normal vectors may all be parallel to each other or they may extend in several different directions, perpendicular to the digitized lower edge of the mandible;
[0036]
[0036] - The step of localizing the lower border of the mandible is performed by detecting a contrast variation on the radiographic image itself or on a derivative of said radiographic image, such as, for example, gradients calculated from the radiographic image. In particular, during the localization step, the digitized lower border of the mandible is determined on the radiographic image of the mandible or a derivative of said radiographic image by a step of searching for a plurality of points adjacent to each other and all exhibiting a contrast value or a contrast variation greater than a threshold value, a step possibly enhanced by statistical learning methods allowing adaptation to natural imperfections in the images. This advantageous configuration thus makes it possible to detect any shape of the lower border of the mandible.In other words, this advantageous configuration makes it possible to avoid starting from restrictive assumptions about the lower edge of the mandible and to adapt to any type of radiographic image;
[0037]
[0037] - the contrast variation is determined by the variation in grey level between a group of adjacent pixels on the image, for example between two adjacent pixels. Consequently, the contrast value or the contrast variation makes it possible to define a level line, or more generally a level curve, associated with the lower edge of the mandible;
[0038] - By manipulating the contrast value, the contrast variation value, and using statistical learning, the method according to the invention aims to find a curve of maximum length and sufficient regularity that is attracted to the interface between the mandible and the void (which surrounds the head and is located lower in the image) without being too susceptible to artifacts that may arise, for example, due to the presence of the hyoid bone. By a sufficiently regular line, we mean a curve whose curvature is sufficiently small, for example. This line of maximum length and sufficient regularity defines the lower edge of the mandible;
[0038]
[0039] The method according to the invention includes a step of limiting the digitized lower edge by determining lateral extremities of the digitized lower edge beyond which the digitized lower edge is no longer representative of the lower edge of the mandible, the normal vectors determined during the determination step being defined only between the two lateral extremities thus defined. In particular, according to a first embodiment, the step of limiting the lateral extremities of the digitized lower edge includes a step of applying an active contour model to the digitized lower edge of the mandible in order to determine the lateral extremities.Alternatively, according to a second embodiment, the step of limiting the lateral extremities of the digitized lower border includes a step of calculating a derivative of the digitized lower border, the lateral extremities of the digitized lower border being determined when the value of the derivative exceeds a threshold value. These advantageous configurations thus make it possible to locate the lower border of the mandible unaffected by the hyoid bone or chin artifacts.
[0039]
[0040] - During the step of defining the sequence of normal vectors, these are taken in an equidistant manner, that is to say at regular intervals, along the digitized lower edge between the two lateral extremities of the digitized lower edge of the mandible. Their number is constant and fixed by the user, for example: 100 intervals, 100 intervals being the preferred embodiment of the invention.
[0040]
[0041] - In the step of determining the largest subsets, the largest subsets are determined parallel to the digitized lower edge of the mandible, across all the previously defined normal vectors. In other words, they are calculated by taking the i-th grayscale curve parallel to the edge and associated with the normal vector bundle, and identifying the largest homogeneous grayscale segment for this i-th curve. Put another way, the largest subsets define contour lines that extend parallel to the digitized lower edge of the mandible, at various depths along the previously defined normal vectors.In the context of the present invention, a group of values of several normal vectors is considered to define a homogeneous subset if the gray level value of each component of the normal vectors forming said subset deviates little from a reference value (for example, the mean or the median). In particular, a gray level value of each component of the normal vectors forming the subset is considered to deviate little from a mean reference value or the median if said gray level value is equal to a given value—for example, the mean or the median of said subset—or if it is distant from said given value with a sufficiently low Z-score.
[0041]
[0042] - In the determination stage, the largest subsets of homogeneous or substantially homogeneous grey levels extend along curvilinear lines taken parallel to the lower edge of the mandible. In other words, the grey level subsets are constructed at substantially constant distances from the digitized lower edge of the mandible, so as to form iso-distance lines conforming to the curvilinear geometry of said lower edge. These curvilinear lines are obtained by considering, for a given distance from the lower edge, the grey level values resulting from the plurality of adjacent normal vectors, which makes it possible to preserve the natural morphology of the mandible independently of its local shape;
[0042]
[0043] - During the determination stage, the largest subsets of homogeneous grey levels corresponding to homogeneous segments along the curvilinear lines are selected. In other words, for each curvilinear line considered, a continuous segment of maximum length is sought for which the grey levels remain homogeneous or substantially homogeneous. These homogeneous segments are preferentially centered with respect to a median area of the mandible, so as to limit the influence of artifacts present near the spine or the lateral extremities of the radiographic image;
[0044] - During the determination step, the average gray level is calculated for each subset of homogeneous gray levels. This provides gray level values associated with these subsets, which together form a curve, denoted m(a), representing the variation of gray level as a function of distance from the lower border of the mandible. This curve is taken along the normal vector and is characteristic of a state of the mandibular bone, forming the biomarker. In other words, the average values of the homogeneous segments determined for different distances from the lower border of the mandible are aggregated to form a density profile curve m(a) representative of the variation in bone density within the mandible. Each m(a) value corresponds to a denoised estimate of the gray level of the mandibular bone at a given depth from the lower border of the mandible.The curve m(a) thus obtained constitutes a grey level enhancement profile, whose overall geometry, including in particular its local variations, maxima and damping zones, is directly correlated to the state of density and structuring of the mandibular bone;
[0043]
[0045] - The m(a) curve is compared to families of reference m(a) curves, constructed from a database of panoramic radiographic images of the mandible, called annotated images, in order to determine a bone state belonging to a predefined class or cluster. In other words, the m(a) curve associated with a given mandible is positioned relative to a set of typical curves previously established from a reference population. These families of curves are constructed from radiographic images associated with clinical and biological metadata, so that each family corresponds to an identified bone state, such as a healthy state, an intermediate state, or an osteoporotic state;
[0044]
[0046] - The reference curve families m(a) are constructed from a training database comprising approximately 40,000 panoramic radiographic images of the mandible. The database is structured into successive five-year age groups, each containing approximately 2,500 images, with a roughly balanced distribution between male and female subjects. The images in the database are associated with corresponding biological and clinical data, allowing the extracted bone density profiles to be linked to the bone status of the subjects. For each image in the database, a density profile curve m(a) is constructed from the lower border of the mandible by defining vectors normal to said lower border and estimating, along these vectors, a grayscale enhancement curve from the lower border towards the interior of the mandible.Statistical analysis of the resulting m(a) curves shows that, for young subjects, particularly men and women between the ages of forty and forty-five, the density profiles exhibit a high degree of similarity, reflecting a homogeneous bone state with little impact from osteoporosis. Conversely, for older subjects, and especially for older women, the analysis reveals a greater diversity of m(a) profiles, corresponding to the emergence of new families of curves characteristic of progressive bone structure deterioration and osteoporotic conditions.The construction of these families of curves relies essentially on the use of an adaptive grey level estimator, defined perpendicular to the mandible and calculated on a limited number of adaptively selected points, which makes it possible to obtain robust density profiles, comparable between individuals and representative of the real variations in bone density;
[0045]
[0047] - The comparison is performed using a statistical grouping or machine learning method, allowing the m(a) curve to be classified into several classes or clusters representing distinct bone states. In other words, the similarity between the studied m(a) curve and the reference curves is evaluated using statistical or algorithmic classification tools. These tools allow curves with similar morphological characteristics to be grouped into clusters, each cluster being associated with a bone state determined based on analyses performed on a large number of radiographic images;
[0046]
[0048] The method includes a step of selecting one of the normal vectors, called the selected normal vector. From this selected normal vector, the determination step is performed, such that, for a given distance from the digitized lower edge of the mandible, a gray level value of directly adjacent normal vectors located on either side of the selected normal vector is compared. If the gray level value of the adjacent normal vectors, taken at the given distance, is equal to or comparable to that of the selected normal vector taken at the same distance, then the largest set of gray levels is defined by said adjacent normal vectors and said selected normal vector, taken at the given distance. In other words, a reference normal vector is chosen and serves as the starting point for constructing the homogeneous segment under consideration.The grey level value measured on this selected normal vector is used as a comparison value to determine how far adjacent normal vectors can be integrated without loss of homogeneity, thus ensuring an adaptive construction of the homogeneous segment;
[0047]
[0049] - The integration of adjacent normal vectors is performed iteratively by increasing, at each iteration, the length between the selected normal vector and the adjacent normal vectors on either side of the normal vector, at a given distance from the digitized lower edge of the mandible. If the grayscale values taken at the adjacent normal vectors, at the given distance, are equal to or comparable to the largest grayscale set, then the largest grayscale set is augmented by the aforementioned adjacent normal vectors, taken at the given distance. In other words, determining a homogeneous segment involves a step of progressive lateral extension, starting from the selected normal vector, until a break in the homogeneity of grayscale levels is detected along the adjacent normal vectors, for a given distance from the digitized lower edge of the mandible.In other words, the construction of the homogeneous segment is performed by an incremental process, progressively extending left and right from the selected normal vector. This advantageous configuration allows for the automatic identification of the maximum width of a homogeneous segment while excluding areas where variations in gray level indicate the presence of imaging artifacts or structural changes in the bone.
[0048]
[0050] - A homogeneous greyscale set is defined as a contiguous set of greyscale values whose statistical dispersion is less than a predetermined threshold, said dispersion being evaluated by variance, standard deviation, or distance from a mean or median value. In other words, the greyscale values constituting a homogeneous segment exhibit limited variability around a common representative value. This statistical definition makes it possible to distinguish truly homogeneous areas of the mandibular bone from areas affected by noise or artifacts related to the radiographic image acquisition conditions;
[0049]
[0051] - Each normal vector is defined as a vector locally perpendicular to the digitized lower border of the mandible, said direction vector being determined from a local derivative or tangent to the lower border, taken at a considered point on the mandible. In other words, the orientation of each normal vector is adjusted locally according to the geometry of the lower border of the mandible. This local definition ensures that the normal vectors penetrate the mandibular bone in a relevant direction, regardless of morphological variations in the mandible;
[0050]
[0052] - Each normal vector is discretized into a succession of positions corresponding to constant steps – representing a distance from the digitized lower edge of the mandible – and equivalent to one pixel, each position being associated with a gray level value. In other words, gray levels are sampled at regular intervals along each normal vector, allowing for a discrete and normalized representation of bone density in depth. This discretization facilitates the comparison of density profiles between different normal vectors and between different radiographic images;
[0051]
[0053] Each normal vector has a width—measured perpendicular to its direction—of one pixel, so that the grayscale values correspond to a local elementary slice of the radiographic image. In other words, each normal vector represents a minimal, localized extraction of information from the radiographic image. This unit width preserves the spatial resolution of the image while limiting the effects of excessive averaging that could mask local variations in bone density.
[0052]
[0054] According to a second aspect of the invention, a device for determining a mathematical biomarker for the detection of osteoporosis is proposed, the mathematical biomarker being representative of a bone density of a mandible and obtained from at least one panoramic radiographic image of the mandible, the device comprising a control unit having means configured to implement the method according to the first aspect of the invention or according to any of its improvements.
[0053]
[0055] By way of example, but not limited to, the control unit includes a microcontroller and / or a printed circuit board and / or a microprocessor. Additionally, the control unit may also include memory. Examples of control units include a computer, a laptop, a tablet, a remote server, etc.
[0054]
[0056] According to a first embodiment, the device according to the invention implements the method according to the invention "online," that is, directly after the acquisition of the corresponding radiographic images of a patient's jaw. In other words, in this first embodiment, the device is, for example, located at a dentist's office to provide such a biomarker directly after acquiring panoramic radiographic images of the jaw for dental purposes, the biomarker providing additional information to the patient, if applicable. In particular, at least one radiographic image is obtained by a radiographic device of the apparatus, the radiographic device being configured to acquire at least one panoramic image of a patient's mandible, and wherein the method is implemented directly after the acquisition of said radiographic image.
[0055]
[0057] According to a second embodiment, the device according to the invention implements the method according to the invention "offline," that is, after the acquisition of the corresponding radiographic images of a patient's jaw. In other words, in this second embodiment, the device is, for example, located at a healthcare center or in a facility other than the one where the dentist takes the radiographic images. In this case, the radiographic images acquired are transmitted and stored in a data center that includes a database containing, in addition to the radiographic images, patient information, including their identification.The availability of this stored data and / or radiographic images allows, subsequently—that is, independently of when the radiographic images were acquired—the determination of such a biomarker according to the invention by a processing unit that queries this database. In other words, in this second embodiment, at least one radiographic image is obtained from a database, remote or local, containing radiographic images of multiple patients. The method is implemented on at least a portion of the radiographic images in the database, after the images have been acquired. The biomarker can thus be calculated for each image present in the database.
[0056]
[0058] In either of its embodiment variants, the device conforming to the second aspect of the invention has computing means for performing statistical grouping or machine learning of biomarkers, allowing the biomarkers obtained for each person to be classified among several classes or clusters representative of distinct bone states.
[0057]
[0059] According to a third aspect of the invention, an assembly is proposed comprising an apparatus for determining a mathematical biomarker for the detection of osteoporosis in accordance with the second aspect of the invention or according to any of its variants or improvements, a radiography device enabling the production of panoramic radiographic images, and a means of communication / transmission of images between the radiography device and the apparatus for determining a biomarker.
[0058]
[0060] Finally, according to a fourth aspect of the invention, it is proposed to use a mathematical biomarker obtained according to the method described in the first aspect of the invention or according to any of its improvements, for:
[0059]
[0061] - the detection, particularly early detection, of osteoporosis or
[0060]
[0062] - Regular monitoring of osteoporotic patients receiving medical treatment, for example treatment with bisphosphonates,
[0061]
[0063] in order to assess the effectiveness of the treatment on bone density and the progression of osteoporosis.
[0062]
[0064] Various embodiments of the invention are envisaged, incorporating, according to all their possible combinations, the different optional features described herein.
[0065] Other features and advantages of the invention will become apparent from the following description on the one hand, and from several illustrative and non-limiting examples of embodiments given with reference to the attached schematic drawings on the other hand, in which:
[0063]
[0066] [Fig.1] illustrates a synoptic diagram of an example of an embodiment of the method conforming to the first aspect of the invention;
[0064]
[0067] [Fig.2] illustrates a radiographic image of a mandible and on which the method illustrated in FIGURE 1 is applied;
[0065]
[0068] [Fig.3] illustrates, from two panoramic images of healthy subjects, an indicator, and more particularly the curve of average grey levels when the distance to the edge of the mandible increases, obtained by the method according to the invention;
[0066]
[0069] [Fig.4] illustrates, from two panoramic images of osteoporotic subjects, an indicator, and more particularly here the curve of average grey levels when the distance to the edge of the mandible increases, obtained by the method according to the invention;
[0067]
[0070] [Fig.5] illustrates a radiographic image of a mandible on which the extraction areas are defined;
[0068]
[0071] [Fig.6] illustrates the grid on which the estimators are created and the grey levels are extracted; this grid is itself an indicator within the meaning of the invention;
[0069]
[0072] [Fig.7] illustrates a figure showing the grey level curves as a function of the distance to the edge estimated by several competing estimators which can serve as an indicator within the meaning of the invention;
[0070]
[0073] [Fig.8] illustrates a dendrogram obtained from grey level curves estimated as a function of the distance to the edge extracted according to the method of the invention, grouped into similarity clusters; the horizontal dotted line shows a choice allowing the definition of 4 clusters;
[0071]
[0074] [Fig.9] illustrates the results for the four clusters (1 to 4) obtained according to the method of the invention and previously defined, comprising healthy patients (clusters 2 and 3) and osteoporotic patients (cluster 1).
[0075] Of course, the features, variants, and different embodiments of the invention can be combined in various ways, provided they are not incompatible or mutually exclusive. In particular, variants of the invention may include only a selection of features, described hereafter in isolation from the other described features, if this selection of features is sufficient to confer a technical advantage or to differentiate the invention from prior art.
[0072]
[0076] In particular, all the variants and embodiments described can be combined with each other if there are no technical obstacles to this combination.
[0073]
[0077] In the figures, elements common to several figures retain the same reference.
[0074]
[0078] With reference to Figure 1, the invention addresses a method 1 for constructing a mathematical biomarker for the detection of osteoporosis, the mathematical biomarker being representative of the bone density of a mandible 2 and obtained from at least one panoramic IR radiographic image of the mandible 2. Method 1 is implemented, for example, by a computer or a remote server. More generally, Method 1 is implemented by an apparatus for determining a mathematical biomarker for the detection of osteoporosis, the apparatus comprising a control unit having means configured to implement Method 1 according to the invention.
[0075]
[0079] Method 1 according to the invention comprises the following steps:
[0076]
[0080] - a localization step 11 of a lower edge 21 of the mandible 2 on at least one image, called lower edge 21 digitized;
[0077]
[0081] - a definition step 12 of a sequence of normal vectors aO, a1, a2, ai, ai+1, an, each normal vector aO, a1, a2, ai, ai+1, an being associated with a grey level value NGX of a plurality of pixels extending along a direction of the normal vector from the digitized lower edge 21 of the mandible 2, taken at predetermined intervals from said lower edge 21. The normal vectors aO, a1, a2, ai, ai+1, an are visible in particular on FIGURE 6;
[0082] - a determination step 13 of the largest subsets of homogeneous or substantially homogeneous NGX grey levels extending along the mandible 2, across the set of normal vectors aO, a1, a2, ai, ai+1, an defined previously and for each predetermined interval taken at a distance from the lower edge 21 digitized of the mandible 2;
[0078]
[0083] - a calculation step of 14 representative indicators of NGX grey level distributions calculated for the largest subsets for each predetermined interval.
[0079]
[0084] Figures 2, 3 and 4 illustrate in more detail the operation of method 1 according to the invention.
[0080]
[0085] As seen in FIGURE 2, the localization step 11 of the lower edge 21 allows us to define the region of the mandible 2 from which the normal vectors aO, a1, a2, ai, ai+1, .., an, are calculated.
[0081]
[0086] As specified, the localization step 11 of the lower border 21 of the mandible 2 is carried out by detecting a variation in contrast on the IR radiographic image itself or on a derivative of said IR radiographic image, such as for example gradients calculated from the IR radiographic image.
[0082]
[0087] In Figure 2, it can be seen that the lower edge 21, digitized during the localization step 11, is formed from a collection of points on the IR radiographic image of the mandible 2. This localization is determined on the IR radiographic image of the mandible 2, or a derivative thereof, by a step involving the search for a plurality of points adjacent to one another, all exhibiting a contrast value or contrast variation greater than a threshold value. In particular, Figure 2 shows that the digitized lower edge 21 of the mandible 2 is identified by an apparent contrast with the background 22 of the IR radiographic image, i.e., a gray level NGX sufficiently different from that of the background 22 of the IR radiographic image.
[0083]
[0088] The background 22 of the IR radiographic image is formed, for example, by the area of the IR radiographic image peripheral to the mandible 2 itself, and / or by the area of the IR radiographic image formed by neck tissues.
[0084]
[0089] The lower border 21 of the mandible 2 extends from a part of the mandible 2 proximal to the chin, and towards a posterior part of the mandible 2. To this end, in order to laterally limit the digitized lower border 21, method 1 according to the invention includes a limitation step 15 of the digitized lower border 21 by determining lateral extremities of the digitized lower border 21 beyond which the digitized lower border 21 is no longer representative of the lower border 21 of the mandible 2. After this limitation step 15, the normal vectors aO, a1, a2, ai, ai+1, an, subsequently determined in the determination step 13 will be defined only between the two lateral extremities thus defined.
[0085]
[0090] According to a preferred embodiment of the invention, the limiting step 15 of the lateral ends of the digitized lower edge 21 includes a step of applying an active contour model 16 to the digitized lower edge 21 of the mandible 2 in order to determine the lateral ends.
[0086]
[0091] Following these preliminary steps, and as seen in FIGURE 2, the lower edge 21 digitized during the localization step 11 of method 1 allows to perfectly adhere to the lower bone of the mandible 2. The digitized lower edge 21 thus forms a curvilinear curve m(a) which coincides with the lower end of the mandibular bone of the mandible 2.
[0087]
[0092] From this digitized lower edge 21, the definition step 12 allows the definition of several normal vectors aO, a1, a2, ai, ai+1, ..., an, which all extend perpendicularly with respect to the lower edge 21 of the mandible 2. In particular, each normal vector aO, a1, a2, ai, ai+1, ..., an is calculated so as to extend perpendicularly to a local segment of the digitized lower edge 21 of the mandible 2. For this purpose, the local derivative of the digitized lower edge 21 can be calculated at position x along said lower edge 21 to determine the direction to take for a given normal vector aO, a1, a2, ai, ai+1, ..., an.
[0088]
[0093] Along the lower edge 21 digitized, the normal vectors aO, a1, a2, ai, ai+1, an, are regularly distributed, so that a distance between two directly adjacent normal vectors aO, a1, a2, ai, ai+1, .. an is constant.
[0089]
[0094] Preferably, in order to obtain sufficient information density, we choose to cut the digitized lower edge 21 into several tens of normal vectors aO, a1, a2, ai, ai+1, an, preferably one hundred.
[0095] Each normal vector aO, a1, a2, ai, ai+1, an, has a plurality of values that are representative of a gray level NGX of the IR radiographic image along the elongation direction of said normal vectors aO, a1, a2, ai, ai+1, an. These gray level NGX values can correspond to the value of the given pixel located at a distance a from the lower edge 21 on the given normal vector aO, a1, a2, ai, ai+1, ..., an, or correspond to a statistical representative of a group of pixels located around the distance a from the lower edge 21 on the given normal vector aO, a1, a2, ai, ai+1, ...an, such as, for example, an average.
[0090]
[0096] At the end of the definition step 12 of the normal vectors aO, a1, a2, ai, ai+1, an, each normal vector aO, a1, a2, ai, ai+1, ..., an, comprises a plurality of NGX grey level values which represents a bone density of the mandible 2, from the lower edge 21 digitized of the mandible 2, towards the interior of the mandible 2.
[0091]
[0097] The curve m(a) shown in FIGURE 2 in the shape of an "S" gives an example of the NGX grey levels on one of the normal vectors aO, a1, a2, ai, ai+1, .., an, defined as previously.
[0092]
[0098] Next, during determination step 13, method 1 aims to define curvilinear lines b1, b2, bi, .. bn that extend transversely, perpendicularly from the normal vectors a0, a1, a2, ai, ai+1, .. an, defined previously. These curvilinear lines b1, b2, bi, bn all extend parallel or substantially parallel to the digitized lower edge 21 of the mandible 2.
[0093]
[0099] In other words, the largest subsets are calculated by taking the ith distance a on each normal vector aO, a1, a2, ai, ai+1, an, and identifying, at all these distances a on the normal vectors aO, a1, a2, ai, ai+1, an, in the direction along the curvilinear lines b1, b2, bi, ..., bn, the largest homogeneous segment in NGX grayscale. By homogeneous, we consider that the NGX grayscale values of these subsets are equal or that they exhibit a reduced value distribution relative to an average value of the subset in question.
[0100] Preferably, these larger subsets are taken centered between the two lateral ends of the digitized lower edge 21 of the mandible 2 and as defined previously.
[0094]
[0101] Next, the average of these homogeneous NGX grayscale values for each largest homogeneous NGX grayscale segment extending along the curvilinear lines b1, b2, bi, ..., bn, and located at a distance a, yields a denoised NGX grayscale value at the center of mandible 2, and thus a denoised bone density value for mandible 2 at a distance from the digitized lower edge 21. This allows us to obtain a curve m(a) representative of bone density, forming a mathematical indicator that can provide a relevant, robust, and reliable biomarker for osteoporosis. In particular, the geometry of this curve m(a) characterizes the state of the mandibular bone in mandible 2 and offers a biomarker for the detection, prediction, and monitoring of osteoporosis.
[0095]
[0102] These m(a) curves - visible in FIGURES 3 and 4 associated respectively with two healthy subjects and two osteoporotic subjects - represent the average evolution of NGX grey levels - and therefore of bone density - as a function of the distance from the lower edge 21 digitized of the mandible 2. Depending on the level of exposure to osteoporosis, such as for cases of "absence of osteoporosis", "early osteoporosis" and "severe osteoporosis", the NGX grey level cascade defined by the m(a) curves in question presents a characteristic geometry which goes for example from a geometry with strong localized enhancement in the case of absence of osteoporosis - as seen in FIGURES 3 - to a damped and irregular geometry in the case of strong presence of osteoporosis - as seen in FIGURES 4.
[0096]
[0103] Figure 5 illustrates a panoramic IR radiograph of a mandible on which the extraction zones are defined and represented by two rectangles located on either side of the chin. These extraction zones are defined so as to be located at a distance from the central area of the panoramic IR radiograph because the gray levels represented at the central area are affected by the presence of the patient's spine in the background, artificially brightening the imaged areas. In other words, the gray level of the area associated with the chin is not a faithful representation of the lower jawbone at the chin because it is affected by the presence of the spine in the background of the panoramic IR radiograph. Therefore, this area is an exclusion zone that should not be considered by Method 1 according to the invention.
[0097]
[0104] On the other side, at the level of a rear outer edge of the lower jaw, each extraction zone is delimited near the last teeth present on the panoramic IR radiographic image.
[0098]
[0105] The extraction zones are advantageously symmetrical with respect to each other, with respect to a median vertical axis of the jaw passing approximately through the middle of the chin, although the extraction zones can also be asymmetrical in the case of an asymmetrical jaw, for example.
[0099]
[0106] Figure 6 illustrates the creation of the estimators and the extraction of gray levels. In particular, Figure 6 illustrates all the normal vectors that are created from the digitized lower edge of the jaw. These normal vectors are all perpendicular to the local portion of the digitized lower edge of the jaw from which they extend into the image. Each normal vector has a width of 1 pixel, that is, perpendicular to its own elongation, and a length of several pixels, taken along its own elongation. The normal vectors thus define a gray-level analysis grid for the panoramic radiographic image of the patient's jaw.And it is from at least one of them that the homogeneous sets of grey levels are then determined, in a direction perpendicular to each normal vector - the curvilines - and preferably from a selected normal vector forming the starting point of this analysis and from which we extend, on the curvilines, to the right and to the left, on the different adjacent normal vectors, the homogeneous set provided that the grey level of the adjacent normal vector is indeed homogeneous, that is to say comparable, that is to say equal or substantially equal, to the value of the grey level of the selected normal vector and / or to the average value of the grey level of the largest set of grey levels obtained so far.Through successive iterations, the method thus stretches, for a given distance *a* from the lower edge of the jaw, along the curvilines, the homogeneous set of gray levels, starting from the selected normal vector, until the largest set is found. As soon as, on one side or the other of the selected eigenvector, a gray level sufficiently different from the gray level of the selected normal vector or from the average value of the homogeneous set of gray levels obtained so far, then this iterative process is interrupted.
[0100]
[0107] This iterative determination of the largest homogeneous set of grey levels, taken from the selected normal vector, is carried out symmetrically with respect to said selected normal vector or, possibly, asymmetrically and independently of each side of said selected normal vector.
[0101]
[0108] Advantageously, for each extraction zone, we define:
[0102]
[0109] - a first normal vector selected as being located at or near a central area of said extraction zone; and
[0103]
[0110] A second selected normal vector taken at three-quarters of the extraction zone and outwards from said extraction zone.
[0104]
[0111] The method according to the invention further includes a validation step for the homogeneous sets of gray levels thus obtained: if the results obtained for at least one of the two selected normal vectors thus defined, for a given extraction area, are satisfactory and valid, then the use of the data and the mathematical biomarker thus defined is considered viable. Conversely, if the results obtained for both selected normal vectors thus defined, for a given extraction area, are incorrect and / or unusable and / or erroneous and / or of poor quality, then the use of the data and the mathematical biomarker thus defined is considered not viable.
[0105]
[0112] Figure 7 illustrates a graph where the x-axis represents the distance from the lower border of the mandible, while the y-axis represents the intensity of the gray level. In other words, the y-axis represents the thickness of the jawbone measured at a given distance from the lower border of the mandible. The different curves are obtained either at different positions within the extraction area or for several different patients. In particular, Figure 6 illustrates profiles of healthy jaws, with a nominal profile for bone thickness, despite a slight variation in thickness between radiographed subjects.
[0113] We observe that, starting from the lower edge of the jaw, the bone initially thickens, reaching a maximum at a depth of approximately 5-7 arbitrary units, before thinning to a minimum at approximately 20 arbitrary units, and then thickening again at greater depths. This latter upward shift in profiles is not entirely accurate, as it is also explained by the nature of the radiographic images obtained and the presence of artifacts in the peripheral outer areas of the jaw, where the radiographic imaging reveals teeth and / or bones located behind the jaw itself.
[0106]
[0114] Figure 9 illustrates the results for four clusters (1 to 4) obtained according to the method of the invention. Clusters 2 and 3 represent data obtained from healthy patients, i.e., those with a healthy and nominal jaw, while clusters 1 and 4 represent non-nominal jaws, and in particular jaws showing osteoporosis for cluster 1.
[0107]
[0115] As shown in Figure 8, a dendrogram with profiles obtained according to the method of the invention, grouped into clusters. This dendrogram, obtained by hierarchical clustering according to Ward's criterion D2, highlights a non-arbitrary structuring of the analyzed profiles, characterized by the existence of several distinct families separated by clear breaks in dissimilarity. Cross-sectioning the dendrogram at the threshold indicated by a horizontal dashed line naturally leads to the identification of three main clusters, corresponding to homogeneous but distinct mandibular density profiles. This reflects the existence of differentiated bone states and supports the use of these groupings as reference classes for classifying an individual's bone status.
[0108]
[0116] In summary, the invention relates to a method 1 for constructing a mathematical biomarker for the detection of osteoporosis, the mathematical biomarker being representative of the bone density of a mandible 2 and obtained from at least one panoramic IR radiographic image of the mandible 2, the method 1 comprising a localization step 11 of a digitized lower edge 21 of the mandible 2, a definition step 12 of a plurality of normal vectors extending perpendicularly to the digitized lower edge 21, each normal vector comprising several NGX gray level values of the pixels of the IR radiographic image taken along the elongation direction of the normal vectors a0, a1, a2, ai, ai+1, an, a determination step 13 of the largest subsets of homogeneous NGX gray levels extending across the normal vectors, for example perpendicularly,a calculation step 14 of average values of these large subsets at different distances from the lower edge 21 digitized and towards the interior of the mandible 2.,
[0109]
[0117] Of course, the invention is not limited to the examples just described, and many modifications can be made to these examples without departing from the scope of the invention. In particular, the various features, forms, variants, and embodiments of the invention can be combined in various ways, provided they are not incompatible or mutually exclusive. Specifically, all the variants and embodiments described above are combinable.
Claims
Demands
1. Method (1) of constructing a mathematical biomarker for the detection of osteoporosis, the mathematical biomarker being representative of the bone density of a mandible (2) and obtained from at least one panoramic radiographic (IR) image of the mandible (2) of a person, method (1) being implemented by a control unit comprising a microcontroller and / or a printed circuit board and / or a microprocessor and memory, method (1) comprising the following steps: - a localization step (11) of a lower edge (21) of the mandible (2) on at least one image, called lower edge (21) digitized; - a definition step (12) of a plurality of normal vectors, each normal vector having a grey level value (NGX), representative of a bone density from the radiographic image, of a plurality of pixels extending along a plurality of proper directions from the digitised lower edge of the mandible (2), taken at predetermined intervals from said lower edge (21); - a determination step (13) of the largest homogeneous or substantially homogeneous grey level subsets (NGX) extending along the mandible (2), across the set of normal vectors defined previously and for each predetermined interval taken at a distance from the digitized lower edge (21) of the mandible (2); - a calculation step (14) of at least one indicator representative of the grey level distributions (NGX) representative of bone density and which corresponds to the biomarker, the grey level distributions being calculated for the largest subsets for each predetermined interval.
2. Method (1) according to the preceding claim, wherein, during the normal vector definition step (12), each eigendirection is preferably defined by a straight line direction and secant with respect to the digitized lower edge (21) of the mandible (2).
3. Method (1) according to any one of claims 1 or 2, wherein the localization step (11) of the lower edge (21) of the mandible (2) is performed by detecting a contrast variation.
4. Method (1) according to any one of the preceding claims, wherein the method (1) comprises a limiting step (15) of the digitized lower edge (21) by determining lateral extremities of the digitized lower edge (21) beyond which the digitized lower edge (21) is no longer representative of the lower edge (21) of the mandible (2), the normal vectors determined in the determination step (13) being defined only between the two lateral extremities thus defined.
5. Method (1) according to claim 4, wherein the limiting step (15) of the lateral ends of the digitized lower edge (21) comprises: - an application step (16) of an active contour model onto the digitized lower edge (21) of the mandible (2) in order to determine the lateral extremities; or - a calculation step (14) of a derivative of the digitized lower edge (21), the lateral ends of the digitized lower edge (21) being determined when the value of the derivative exceeds a threshold value.
6. Method (1) according to any one of the preceding claims, wherein in the determination step (13), the largest homogeneous or substantially homogeneous grey level (NGX) subsets extend along curvilinear lines (b1, b2, bi, .. bn) taken parallel with respect to the lower edge (21) of the mandible (2),
7. Method (1) according to the preceding claim, wherein, during the determination step (13), the largest homogeneous grey level (NGX) subsets corresponding to homogeneous segments along the curvilinear lines (b1, b2, bi, .. bn) are selected.
8. Method (1) according to the preceding claim, wherein, during the determination step (13), the average of grey levels is calculated for each homogeneous grey level subset (NGX), so as to provide grey level values associated with said grey level subsets (NGX), and which together represent a curve m(a) of variation of the grey level as a function of the distance from the lower edge (21) of the mandible (2), taken along the normal vector and characteristic of a state of the mandibular bone and forming the biomarker.
9. Method (1) according to the preceding claim, wherein the curve m(a) is compared to families of reference curves m(a), constructed from a database of panoramic radiographic (IR) images of the mandible (2), called annotated images, in order to determine a bone state belonging to a predefined class or cluster.
10. Method (1) according to the preceding claim, wherein the comparison is carried out by a statistical grouping or machine learning method, allowing the curve m(a) to be classified among several classes or clusters representative of distinct bone states.
11. Method (1) according to any one of the preceding claims, wherein method (1) includes a step of selecting one of the normal vectors, said selected normal vector, and from which selected normal vector, the determination step (13) is carried out, such that for a given distance from the digitized lower edge (21) of the mandible (2), a gray level value of the directly adjacent normal vectors located on either side of the selected normal vector is compared, such that if the gray level value of the adjacent normal vectors, taken at the given distance, is equal to or comparable to that of the selected normal vector taken at that same distance, then the largest gray level set (NGX) is defined by said adjacent normal vectors and said selected normal vector, taken at the given distance.
12. Method (1) according to the preceding claim, wherein the integration of the adjacent normal vectors is carried out iteratively by increasing at each iteration a length between the selected normal vector and the adjacent normal vectors, on either side of the normal vector, at the given distance from the digitized lower edge (21) of the mandible (2), and such that if the grey level values taken at the level of the adjacent normal vectors, at the given distance, are equal or comparable to that of the largest grey level set (NGX), then the largest grey level set (NGX) is augmented by said adjacent normal vectors, taken at the given distance.
13. Method (1) according to any one of the preceding claims, wherein a homogeneous set of grey levels (NGX) is defined as a contiguous set of grey level values whose statistical dispersion is less than a predetermined threshold, said dispersion being evaluated by a variance, a standard deviation or a distance to a mean or median value.
14. Method (1) according to any one of the preceding claims, wherein each normal vector is defined as a vector locally perpendicular to the digitized lower edge (21) of the mandible (2), said normal vector being determined from a local derivative or tangent to the lower edge (21), taken at a considered point of the mandible (2).
15. Method (1) according to any one of the preceding claims, wherein each normal vector is discretized into a succession of positions corresponding to constant steps - representing a distance from the digitized lower edge (21) of the mandible (2) - and equivalent to a pixel, each position being associated with a gray level value.
16. Method (1) according to any one of the preceding claims, wherein each normal vector has a width - taken perpendicular to the proper direction of the normal vector - of one pixel, so that the gray level values correspond to a local elementary slice of the radiographic image.
17. Apparatus for determining a mathematical biomarker for the detection of osteoporosis, the mathematical biomarker being representative of a bone density of a mandible (2) and obtained from at least one panoramic radiographic (IR) image of the mandible (2), the apparatus comprising a control unit having means configured to implement the method (1) according to any one of the preceding claims and having a microcontroller and / or a printed circuit board and / or a microprocessor, and a memory.
18. Apparatus according to claim 17, wherein at least one radiographic image (RI) is obtained by a radiography device of the apparatus, the radiography device being configured to obtain at least one panoramic image of the mandible (2) of a patient, and wherein method (1) is implemented directly after the acquisition of said radiographic image (RI).
19. Apparatus according to claim 17, wherein at least one radiographic image (RI) is obtained from a remote or local database grouping the radiographic images (RI) of a plurality of patients, method (1) being implemented on at least a part of said radiographic images (RI) of the database.
20. An apparatus according to any one of claims 17 to 19, wherein the apparatus comprises computing means for performing statistical or machine learning grouping of biomarkers, enabling the classification of the biomarkers obtained for each person into several classes or clusters representative of distinct bone states.
21. An assembly comprising the apparatus for determining a mathematical biomarker for the detection of osteoporosis according to any one of claims 17 to 20, a radiography device for acquiring panoramic radiographic (IR) images, and a means for communication / transmission of images between the radiography device and the apparatus for determining a biomarker.
22. Use of a mathematical biomarker obtained according to the method of claims 1 to 16, for: - the detection of osteoporosis or - monitoring of osteoporotic patients receiving medical treatment, for example treatment with bisphosphonates, in order to assess the effectiveness of the treatment on bone density and the progression of osteoporosis by monitoring the value of the biomarker, i