Automatic measurement of bone density from ct images and methods and systems for osteoporosis analysis

Abstract

The application relates to the technical field of medical image processing, and discloses a CT image bone density automatic measurement and osteoporosis analysis method and system, which comprises the following steps: step 1, acquiring CT body data and determining a target vertebra; step 2, extracting an internal reference area and converting equivalent bone density; step 3, constructing a cancellous bone core layer area and a shell layer area and dividing the areas; step 4, counting core layer and shell layer bone density and direction imbalance; step 5, determining a low-density voxel set, a low-density connected destruction amount and a shell-core imbalance amount; step 6, extracting an axial section and determining a weak section bearing capacity and a single vertebra analysis result; and step 7, weightedly collecting and determining a subject-level osteoporosis analysis result. The application realizes bone density automatic measurement and osteoporosis analysis.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, specifically relating to a method and system for automatic measurement of bone mineral density and analysis of osteoporosis in CT images. Background Technology

[0002] CT scans offer comprehensive slice information and high spatial resolution for bone tissue assessment. Clinically, imaging data from the lumbar spine and thoracolumbar region are frequently used to observe bone changes and, in conjunction with bone mineral density (BMD) results, to aid in the assessment of osteopenia and osteoporosis. However, current processing methods still rely on manual selection of vertebrae, manual delineation of measurement areas, or empirical sampling on a limited number of slices. This can lead to inconsistencies in the scope and slice selection between different operators, affecting the stability of BMD measurements. For images with truncated views, local artifacts, or complex vertebral boundaries, manual processing is time-consuming and repeatability is difficult to guarantee.

[0003] Furthermore, existing methods often focus more on obtaining a single average density value, failing to adequately consider the differences in density distribution across different regions within the vertebral body, the spatial aggregation of low-density areas, and the load-bearing capacity of weak sections of the vertebral body. In other words, even if the overall average values ​​are similar, there may still be uneven variations between the upper, lower, anterior, and posterior column regions within the vertebral body. If these variations cannot be further extracted during image processing, they can easily affect the completeness of osteoporosis analysis. Summary of the Invention

[0004] This invention provides a method and system for automatic measurement of bone mineral density and analysis of osteoporosis from CT images, solving the technical problems in the background art.

[0005] This invention provides a method for automatic measurement of bone mineral density and osteoporosis analysis of CT images, comprising the following steps:

[0006] Step 1: Obtain the CT body data corresponding to the target subject, process it to obtain candidate analysis body data, and determine the target vertebral body set and the integrity index corresponding to each target vertebral body;

[0007] Step 2: Extract the internal reference area for each target vertebral body, and determine the average CT value of the muscle region and the average CT value of the fat region; based on the average CT value of the muscle region, the average CT value of the fat region and the preset reference density, convert the original CT value of the voxels in each target vertebral body to obtain the equivalent bone mineral density data corresponding to each target vertebral body.

[0008] Step 3: Construct the cancellous bone core layer and shell layer based on the equivalent bone mineral density data corresponding to each target vertebra, and divide the cancellous bone core layer into subdivisions;

[0009] Step 4: Based on the results of the cancellous bone core layer, shell layer and partition, determine the overall core layer bone density, overall shell layer bone density, head-to-tail density loss and anteroposterior density loss.

[0010] Step 5: Based on the dispersion of the overall core layer bone density and the equivalent bone density in the cancellous bone core layer region, determine the low-density voxel set, the low-density connectivity disruption amount, and the shell-core loss amount.

[0011] Step 6: Extract axial sections along the head-to-tail direction of each target vertebra, and determine the load-bearing capacity of the weak section based on the density area and density inertia of each axial section; determine the single vertebra analysis results corresponding to each target vertebra based on the overall core layer bone density, density loss in the head-to-tail direction, density loss in the anterior-posterior direction, low-density connectivity disruption, shell-core loss, and load-bearing capacity of the weak section.

[0012] Step 7: Based on the integrity index corresponding to each target vertebra, perform weighted summation on the overall nuclear layer bone density corresponding to each target vertebra, and combine the low-density connectivity disruption, weak cross-sectional load-bearing capacity and single vertebra analysis results to determine the osteoporosis analysis results of the tested object.

[0013] This invention provides an automated system for measuring bone mineral density and analyzing osteoporosis from CT images, comprising:

[0014] The data preprocessing module is used to acquire CT body data corresponding to the target subject, process it to obtain candidate analysis body data, and determine the target vertebral body set and the integrity index corresponding to each target vertebral body;

[0015] The reference conversion module is used to extract the internal reference area for each target vertebra, determine the average CT value of the muscle region and the average CT value of the fat region; based on the average CT value of the muscle region, the average CT value of the fat region and the preset reference density, the original CT value of the voxels in each target vertebra is converted to obtain the equivalent bone mineral density data corresponding to each target vertebra.

[0016] The partitioning module is used to construct the cancellous bone core layer and shell layer based on the equivalent bone mineral density data corresponding to each target vertebra, and to partition the cancellous bone core layer.

[0017] The density statistics module is used to determine the overall core layer bone density, overall shell layer bone density, head-to-tail density loss, and anteroposterior density loss based on the results of the cancellous bone core layer, shell layer, and regional divisions.

[0018] The degradation analysis module is used to determine the low-density voxel set, the amount of low-density connectivity disruption, and the shell-core loss based on the dispersion of the overall core layer bone density and the equivalent bone density in the cancellous bone core layer region.

[0019] The load-bearing determination module is used to extract axial sections along the head-to-tail direction of each target vertebra and determine the load-bearing capacity of weak sections based on the density area and density inertia of each axial section; and determine the single vertebra analysis results corresponding to each target vertebra based on the overall core layer bone density, density loss in the head-to-tail direction, density loss in the anterior-posterior direction, low-density connectivity disruption, shell-core loss, and load-bearing capacity of weak sections.

[0020] The results summary module is used to perform weighted summation of the overall nuclear layer bone density corresponding to each target vertebra based on the integrity index of each target vertebra, and combine the low-density connectivity disruption, weak cross-sectional load-bearing capacity and single vertebra analysis results to determine the osteoporosis analysis results of the tested object.

[0021] The beneficial effects of this invention are as follows: Based on CT scan data, this invention automatically completes data processing, internal reference conversion, regional stratification statistics, structural degradation extraction, and subject-level result summarization around the target vertebral body, forming a continuous workflow from voxel-level density processing to vertebral-level and subject-level analysis. This invention not only provides overall core-layer and overall shell-layer bone density, but also further reflects the differences in density distribution in the vertical and anteroposterior directions, as well as low-density connectivity disruption, core-shell imbalance, and weak cross-sectional load-bearing state. This improves the consistency of automatic bone density measurement and makes the osteoporosis analysis results correspond more closely to the actual distribution state within the vertebral body, helping to reduce fluctuations caused by manual layer selection and delineation. Attached Figure Description

[0022] Figure 1 This is a flowchart of the automatic measurement of bone mineral density and osteoporosis analysis method for CT images according to the present invention. Detailed Implementation

[0023] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0024] like Figure 1 As shown, the method for automatic measurement of bone mineral density and analysis of osteoporosis from CT images includes the following steps:

[0025] Step 1: Obtain CT body data corresponding to the target subject, process the data to obtain candidate analysis body data, and determine the target vertebral body set and the integrity index corresponding to each target vertebral body;

[0026] Step 2: Extract the internal reference area for each target vertebral body, and determine the average CT value of the muscle region and the average CT value of the fat region; based on the average CT value of the muscle region, the average CT value of the fat region and the preset reference density, convert the original CT value of the voxels in each target vertebral body to obtain the equivalent bone mineral density data corresponding to each target vertebral body.

[0027] Step 3: Construct the cancellous bone core layer and shell layer based on the equivalent bone mineral density data corresponding to each target vertebra, and divide the cancellous bone core layer into subdivisions;

[0028] Step 4: Based on the results of the cancellous bone core layer, shell layer and partition, determine the overall core layer bone density, overall shell layer bone density, head-to-tail density loss and anteroposterior density loss.

[0029] Step 5: Based on the dispersion of the overall core layer bone density and the equivalent bone density in the cancellous bone core layer region, determine the low-density voxel set, the low-density connectivity disruption amount, and the shell-core loss amount.

[0030] Step 6: Extract axial sections along the head-to-tail direction of each target vertebra, and determine the load-bearing capacity of the weak section based on the density area and density inertia of each axial section; determine the single vertebra analysis results corresponding to each target vertebra based on the overall core layer bone density, density loss in the head-to-tail direction, density loss in the anterior-posterior direction, low-density connectivity disruption, shell-core loss, and load-bearing capacity of the weak section.

[0031] Step 7: Based on the integrity index corresponding to each target vertebra, perform weighted summation on the overall nuclear layer bone density corresponding to each target vertebra, and combine the low-density connectivity disruption, weak cross-sectional load-bearing capacity and single vertebra analysis results to determine the osteoporosis analysis results of the tested object.

[0032] In one embodiment of the present invention, CT volume data corresponding to the target subject is acquired, processed to obtain candidate analysis volume data, and the set of target vertebrae and the integrity index corresponding to each target vertebrae are determined. This part belongs to the preprocessing stage of the entire automatic bone density measurement and osteoporosis analysis process, mainly focusing on the processing of three-dimensional image data, the main location of the vertebrae, and the quantification of usability. Here, CT volume data refers to volume data composed of continuous CT slices in spatial order; candidate analysis volume data refers to volume data after removing the bed board and obvious metal artifacts while retaining the trunk area and the soft tissue area around the vertebrae; the integrity index refers to the result obtained by quantifying whether there are missing, truncated, and residual artifact interference in the target vertebrae.

[0033] Specifically, in step 11, CT volume data is first acquired, and slice thickness, pixel spacing, slice number, and spatial orientation information are extracted. Then, based on this information, 3D reconstruction and standard voxel resampling are performed on the CT volume data to ensure that the volume data generated under different scanning conditions maintains a consistent spatial scale. Standard voxel resampling refers to the process of converting the original volume data into a uniform voxel size, facilitating subsequent region extraction and quantity statistics by voxel.

[0034] After completing the 3D reconstruction, volume domain extraction, bed plate removal, and removal of obvious metal artifacts are performed on the volume data, preserving the trunk region and the soft tissue region surrounding the vertebral bodies to obtain candidate analysis volume data. The soft tissue region surrounding the vertebral bodies is preserved here because the paravertebral muscle and fat regions need to be extracted from the adjacent areas of the vertebral bodies as internal reference regions later. In other words, this step does not simply preserve bone tissue, but rather retains the information of neighboring tissues required for subsequent density conversion while eliminating irrelevant interference. Through the above processing, the influence of the bed plate and artifacts on vertebral body identification is reduced, and a stable data foundation is maintained for subsequent internal reference region extraction.

[0035] In step 12, bony regions are extracted based on the candidate analysis volume data, and multiple vertebral body candidate regions are separated according to connectivity relationships. Here, connectivity relationship refers to the spatial connection between continuously connected bony voxels in the three-dimensional volume data. After separation, transverse processes, spinous processes, and adjacent non-target bony structures are further removed from each vertebral body candidate region to obtain the main vertebral body contour. The main vertebral body contour refers to the boundary of the main vertebral body remaining after removing accessory bony structures; subsequent determination of the target vertebral body assembly, construction of the cancellous bone nucleus layer, and cross-sectional load-bearing analysis are all based on this.

[0036] Next, the vertebral bodies are arranged in a head-to-tail sequence. This head-to-tail direction refers to the anatomical arrangement from the head to the tail. Through this process, the main outlines of multiple vertebrae are organized into an ordered sequence, providing a clear spatial basis for subsequent target vertebral localization. For example, when the candidate analysis data includes multiple thoracic and lumbar vertebrae, a continuous vertebral sequence can be formed first, and then target vertebral screening can be performed within this sequence, rather than directly judging individual isolated bone fragments.

[0037] In step 13, based on the head-to-tail order of the vertebral bodies, the first to third lumbar vertebrae are preferentially identified as the target vertebral body set. When there are gaps in the first to third lumbar vertebrae, continuous vertebrae are identified from the twelfth thoracic vertebrae to the fourth lumbar vertebrae. When only one vertebral body outline exists without any gaps or visual field truncation, the vertebrae corresponding to that outline are identified as the target vertebral body set. Here, visual field truncation refers to the boundary loss caused by a portion of the vertebral body exceeding the scan coverage area.

[0038] After the target vertebrae set is determined, the total number of voxels, the number of voxels with truncation in the field of view, and the number of voxels with residual artifact interference are counted within the main contour of each target vertebra. The ratio of the sum of the number of voxels with truncation in the field of view and the number of voxels with residual artifact interference to the total number of voxels is determined as the reduction amount, and the difference between the reduction amount and the reduction amount is determined as the integrity index. The total number of voxels refers to the total number of voxels contained within the main contour of each target vertebra; the number of voxels with truncation in the field of view refers to the number of voxels corresponding to the missing areas at the edge of the main contour of the vertebra due to insufficient scanning coverage; the number of voxels with residual artifact interference refers to the number of voxels that still affect the boundary and internal voxel distribution of the main contour of the vertebra after the removal of obvious metal artifacts. This not only represents the usability of the target vertebrae, but also provides a unified basis for subsequent weighted summarization at the subject level. For example, when a lumbar vertebra is located at the edge of the scan and the main contour is obviously truncated, its integrity index will be reduced accordingly, and its impact in subsequent summarization will also be reduced.

[0039] Through the above implementation process, the volumetric data is first processed, followed by the extraction of the vertebral body contour, the determination of the target vertebral body set, and the calculation of the integrity index. This ensures that subsequent internal reference area extraction, equivalent bone mineral density conversion, regional stratification statistics, and structural degradation analysis are all based on a consistent three-dimensional image processing foundation. After this processing, the boundaries of the vertebral body objects are clearer, the reference sources of nearby soft tissues are more stable, and the data chain upon which subsequent automatic bone mineral density measurement and osteoporosis analysis rely is smoother.

[0040] In one embodiment of the present invention, an internal reference region is extracted for each target vertebra, and the average CT value of the muscle region and the average CT value of the fat region are determined. Then, based on the average CT value of the muscle region, the average CT value of the fat region, and a preset reference density, the original CT values ​​of the voxels within each target vertebra are converted to obtain the equivalent bone mineral density data corresponding to each target vertebra. This part follows the processing results of the aforementioned target vertebra set, vertebral body outline, and integrity index. Its core lies in establishing an internal reference chain around the soft tissue information near the target vertebra, and then converting the original CT values ​​into equivalent bone mineral density values ​​that can directly participate in subsequent regional statistics and osteoporosis analysis.

[0041] The internal reference region here refers to the soft tissue area located near the target vertebral body that reflects the local grayscale baseline under the current scanning conditions; the average CT value of the muscle region refers to the average result of the original CT values ​​of all voxels in the paravertebral muscle region; the average CT value of the fat region refers to the average result of the original CT values ​​of all voxels in the fat region; the preset reference density refers to the pre-set reference density corresponding to the muscle region and the reference density corresponding to the fat region. The equivalent bone mineral density volumetric data represents the bone mineral density conversion result of the target voxel under the current internal reference region constraint. This data still maintains the original three-dimensional spatial positional relationship, but the numerical meaning changes from the original CT value to the equivalent bone mineral density value.

[0042] Specifically, in step 21, the middle layer is first determined based on the spatial span of the vertebral body contour in the head-to-tail direction corresponding to each target vertebra. Then, the middle layer and its adjacent layers above and below it are selected. The middle layer refers to the layer located in the middle position of the target vertebral body in the head-to-tail direction, used to represent the central layer of the target vertebral body. Subsequently, within the above-mentioned layer range, the paravertebral muscle region is determined by the soft tissue regions adjacent to both sides of the vertebral body contour, and the fat region is determined by the continuous adipose soft tissue region on the periphery of the vertebral body contour, serving as the internal reference area. Here, the paravertebral muscle region refers to the muscle tissue region located on the left and right sides of the vertebral body contour, spatially adjacent to the vertebral body and continuously distributed; the fat region refers to the adipose soft tissue region located on the periphery of the vertebral body contour, continuous in grayscale and spatial distribution. Through the above processing, the internal reference area maintains a close correspondence with the target vertebral body at the same layer level. The reference information called in subsequent conversion no longer depends on the external calibration phantom, but comes directly from the current image data itself.

[0043] In step 22, based on the internal reference area, vascular areas, calcified areas, and residual artifact areas are removed from the paravertebral muscle and fat regions. Vascular areas refer to tissue areas within the soft tissue region that exhibit tubular or locally bright distributions and affect the stability of the average CT value; calcified areas refer to abnormally high-density small areas within the soft tissue region; and residual artifact areas refer to areas that, even after the removal of obvious metallic artifacts in the previous stage, still disturb the local grayscale distribution. After completing the above purification process, the sum of the original CT values ​​of all voxels in the paravertebral muscle region is calculated and divided by the total number of voxels in that region to obtain the average CT value of the muscle region. Similarly, the sum of the original CT values ​​of all voxels in the fat region is calculated and divided by the total number of voxels in that region to obtain the average CT value of the fat region. The total number of voxels refers to the number of all effective voxels in the corresponding region. After this step, the originally dispersed soft tissue grayscale information is consolidated into two stable reference values: one representing the local muscle tissue grayscale level, and the other representing the local fat tissue grayscale level, thus providing a direct basis for subsequent conversion of the original CT value.

[0044] In step 23, for each target vertebral voxel, the difference between the original CT value and the average CT value of the fat region is first determined as the first difference, and the difference between the average CT value of the muscle region and the average CT value of the fat region is determined as the second difference. Then, the conversion factor is determined by dividing the first difference by the second difference. That is, the average CT value of the fat region is first used as a lower limit reference, and the interval between the average CT values ​​of the muscle region and the fat region is used as the conversion scale to determine the relative position of the original CT value of the target voxel within this interval. Then, the difference between the preset reference density corresponding to the muscle region and the preset reference density corresponding to the fat region is multiplied by the conversion factor, and then added to the preset reference density corresponding to the fat region to obtain the equivalent bone mineral density value. Here, the first difference represents the degree of deviation of the target voxel relative to the average CT value of the fat region, the second difference represents the reference span between the local muscle region and the local fat region, and the conversion factor represents the relative proportion of the target voxel within this reference span. The equivalent bone mineral density value can be expressed as: ,in, Indicates the first Voxels within the target vertebra The equivalent bone mineral density value, Voxel representation The original CT values, Indicates the first Average CT value of the muscle region corresponding to each target vertebra. Indicates the first Average CT value of the fat region corresponding to each target vertebral body This indicates the preset reference density corresponding to the muscle region. This represents a preset reference density corresponding to the fat region. Through this processing, the original CT values ​​are uniformly mapped to a reference density range jointly defined by the muscle and fat regions.

[0045] It should be noted that the preset reference density here is not recalculated from the current image, but is a pre-set density baseline value used to convert the image grayscale information into equivalent bone density information that can be directly statistically analyzed later.

[0046] In step 24, the equivalent bone mineral density (BMD) values ​​are backfilled into the corresponding vertebral body contour area according to the spatial location of voxels within each target vertebra, resulting in equivalent BMD data for each target vertebra. Backfilling here refers to replacing the original CT value positions corresponding to voxels with equivalent BMD values ​​without altering the original spatial indexing relationship, thus forming BMD representation data that maintains the original three-dimensional structural relationship. Subsequently, the average CT value of the muscle region, the average CT value of the fat region, and the preset reference density are associated and stored with the equivalent BMD data for each target vertebra. This associated storage means establishing a correspondence between the aforementioned reference values ​​and the equivalent BMD data of the corresponding target vertebra, enabling the tracing of the reference basis used for the target vertebra during subsequent zonal statistics, dispersion calculations, and result backtracking.

[0047] Through the above implementation process, this embodiment first establishes a reference for soft tissues in the same layer and adjacent tissues around the target vertebral body, then converts the original CT values ​​into equivalent bone mineral density values, and forms equivalent bone mineral density volume data that corresponds one-to-one with the main contour of the vertebral body. This preserves the spatial structural relationship of the three-dimensional image data and ensures that the subsequent bone mineral density statistics of the cancellous bone core layer, shell layer, and various regions are based on a unified numerical foundation. For the automatic bone mineral density measurement and osteoporosis analysis process involved in this invention, this processing step can standardize the local image grayscale information into density data that can be directly used in calculations, reduce the impact of grayscale fluctuations under different scanning conditions, and make it easier to maintain consistency in subsequent regional layered statistics and structural degradation determination.

[0048] In one embodiment of the present invention, a cancellous bone core layer region and a shell layer region are constructed based on the equivalent bone mineral density volumetric data corresponding to each target vertebra, and the cancellous bone core layer region is further divided into sub-regions. This part follows the aforementioned unfolding of the vertebral body main outline, standard voxel side length, and equivalent bone mineral density volumetric data. By establishing stable regional boundaries in a unified three-dimensional space, subsequent statistical analyses of overall core layer bone density, overall shell layer bone density, and directional density loss have a clear spatial basis.

[0049] Here, the shell region refers to the outer layer area located on the periphery of the vertebral body contour, preserved after inward dissection; the cancellous bone core layer region refers to the central region obtained after further removing the interference areas of the superior and inferior endplates based on the internal candidate region; the internal candidate region refers to the internal region within the vertebral body contour area, preserved after inward dissection along the outer surface; and the geometric center refers to the central location used for quadrant division within each axial plane. Through this processing, the equivalent bone mineral density volumetric data is no longer just a set of voxel-level values, but is organized into regionalized data that can directly support hierarchical and zonal statistics.

[0050] Specifically, in step 31, the average height of the vertebral body is first determined based on the spatial span of the vertebral body contour in the cephalothorax direction corresponding to each target vertebral body. The average vertebral height refers to the height range of the target vertebral body between the cephalothorax and caudal ends, reflecting the actual scale of the current target vertebral body in three-dimensional space. Subsequently, the shell peeling thickness is determined based on the standard voxel side length and a preset minimum thickness, and the endplate removal height is determined based on a preset ratio of the average vertebral height and a preset maximum height. Here, the shell peeling thickness refers to the peeling distance used when performing equidistant peeling along the outer surface of the vertebral body contour inwards; the endplate removal height refers to the height range removed from the cephalothorax and caudal ends respectively when constructing the cancellous bone nucleus layer region subsequently.

[0051] In one specific embodiment, the shell peeling thickness is the larger of 1.5 times the standard voxel side length and 2 mm. That is, when the standard voxel side length is small, the shell peeling thickness is not less than 2 mm; when the standard voxel side length increases, the shell peeling thickness increases accordingly. The endplate removal height is the smaller of 15% of the average vertebral body height and 5 mm. That is, when the target vertebral body height is small, the endplate removal height is determined proportionally to the average vertebral body height; when the target vertebral body height is large, the endplate removal height does not exceed five millimeters. Through the above settings, the regional construction parameters correspond to both the voxel scale and the vertebral body's own scale, thereby ensuring consistency in the construction boundaries of the subsequent internal candidate region, shell region, and cancellous bone core layer region.

[0052] In step 32, based on the shell peeling thickness, equidistant peeling is performed inward along the outer surface of the vertebral body contour corresponding to each target vertebra, obtaining the internal candidate region. Here, equidistant peeling refers to shrinking the boundary inward at a consistent distance along the vertebral body contour boundary, thereby forming a new internal region. Subsequently, the internal candidate region is subtracted within the vertebral body contour corresponding to each target vertebra, obtaining the shell region. Through this process, the vertebral body contour is decomposed into two levels, with the outer layer representing the shell region and the inner layer representing the internal candidate region. This not only separates the peripheral region of the vertebral body from the overall vertebral body but also provides a foundation for the next step of constructing the cancellous bone nucleus layer.

[0053] It should be noted that the shell region here is not the entire cortical bone in an anatomical sense, but rather the peripheral layered region obtained according to the regional construction rules of this invention, the boundary of which comes from the spatial difference between the main outline of the vertebral body and the internal candidate region.

[0054] In step 33, based on the endplate removal height, the upper endplate interference area is removed from the cephalic end of the internal candidate area, and the lower endplate interference area is removed from the caudal end of the internal candidate area. The remaining area after removing the upper and lower endplate interference areas yields the cancellous bone nucleus layer. Here, the upper and lower endplate interference areas refer to the areas located at the cephalic and caudal ends of the vertebral body, which are easily affected by the edge structure of the endplate and local high density. By further trimming along the cephalic and caudal direction based on the internal candidate area, the remaining area can be more concentrated in the more stable cancellous bone distribution in the middle of the vertebral body. In other words, step 32 yields the inner candidate area, and step 33 further removes the edge influence near the endplate, finally obtaining the cancellous bone nucleus layer suitable for subsequent bone density statistics. For example, when there is a significant edge density change near the cephalic and caudal ends of a vertebral body, this step will exclude that part from the cancellous bone nucleus layer.

[0055] In step 34, based on the cancellous bone nucleus layer, the bone is divided into upper, middle, and lower layers along the head-to-tail direction, and into anterior and posterior columns along the anterior-posterior direction. Within each axial layer, four quadrants are further divided using the geometric center of the cancellous bone nucleus layer as the dividing center. Here, the upper, middle, and lower layers are regions obtained after dividing the cancellous bone nucleus layer from head-to-tail; the anterior and posterior columns are regions obtained after dividing the cancellous bone nucleus layer from anterior-posterior direction; and the four quadrants are local regions obtained by further dividing each axial layer according to the geometric center. Through this zoning process, subsequent analysis can be performed on the bone mineral density of the cancellous bone nucleus layer and shell layer at the overall level, as well as on the differences in vertical, anterior-posterior, and local distributions at the directional level. This preserves the overall structural information of the target vertebral body and allows for more detailed extraction of regional bone mineral density variations.

[0056] Through the above implementation process, the equivalent bone mineral density data corresponding to each target vertebra is further organized into shell region, cancellous bone core region, and multi-level partitioning results. After this processing, clear, stable, and consistent spatial boundaries can be used when performing statistical analysis on overall core layer bone mineral density, overall shell layer bone mineral density, head-to-tail density misalignment, and anteroposterior density misalignment. For this invention, this step not only clarifies the internal structural relationships of the vertebrae in the three-dimensional images but also establishes regional statistics on a unified spatial layering basis, thereby making it easier to maintain consistency in subsequent analysis results.

[0057] In one embodiment of the present invention, based on the results of the cancellous bone core layer, shell layer, and partitions, the overall core layer bone mineral density, overall shell layer bone mineral density, head-to-tail density loss, and anteroposterior density loss are determined. This part expands upon the aforementioned equivalent bone mineral density data, the cancellous bone core layer, shell layer, and the structural results of each partition. By performing stratified and partitioned statistical analysis on the equivalent bone mineral density data, quantitative results that can reflect the overall density level and directional distribution differences of the target vertebral body are further generated.

[0058] The overall core layer bone mineral density (BMD) here refers to the average result obtained by uniformly statistically analyzing the equivalent BMD values ​​corresponding to all voxels within the cancellous bone core layer. The overall shell layer BMD refers to the average result obtained by uniformly statistically analyzing the equivalent BMD values ​​corresponding to all voxels within the shell layer. The corresponding region BMD refers to the average result obtained by statistically analyzing the upper, middle, lower, anterior column, posterior column, and four quadrant regions respectively. The cephalocaudal density discrepancy and anteroposterior density discrepancy are relative differences further formed on the basis of the above-mentioned regional BMD, used to indicate whether the density distribution of the target vertebral body is balanced in different directions.

[0059] In step 41, based on the equivalent bone mineral density (BMD) data corresponding to each target vertebra, the sum of the equivalent BMD values ​​of all voxels within the cancellous bone nucleus layer is calculated and divided by the total number of voxels within the cancellous bone nucleus layer to obtain the overall nucleus layer BMD. Here, the total number of voxels refers to the number of all effective voxels within the cancellous bone nucleus layer; the sum of equivalent BMD values ​​refers to the cumulative result of the equivalent BMD values ​​corresponding to all voxels within the cancellous bone nucleus layer.

[0060] The above statistics can be used to consolidate the voxel-level density information that was originally scattered in the cancellous bone core layer into a single overall density value, providing a unified reference for subsequent directional difference calculations.

[0061] In step 42, based on the equivalent bone mineral density (BMD) data corresponding to each target vertebra, the sum of the equivalent BMD values ​​of all voxels within the shell region is calculated and divided by the total number of voxels in the shell region to obtain the overall shell BMD. The statistical method here is consistent with step 41, but the statistical boundary is changed from the cancellous bone core layer to the shell region. That is, the overall core layer BMD represents the average density state of the cancellous bone region inside the vertebral body, while the overall shell layer BMD represents the average density state of the peripheral region of the vertebral body. By statistically analyzing the two types of regions separately, the density distribution of the inner and outer layers can be distinguished from the overall vertebral body, laying the foundation for subsequent shell-core differential analysis.

[0062] In step 43, based on the equivalent bone mineral density (BMD) data corresponding to each target vertebra, the sum of the equivalent BMD values ​​of all voxels in the upper, middle, lower, anterior column, posterior column, and four quadrant regions is calculated and divided by the total number of voxels in the corresponding region to obtain the BMD of each region. Here, the four quadrants refer to four local regions formed within each axial plane, with the geometric center of the cancellous bone core layer as the dividing center. After this step, the same target vertebra no longer corresponds to only one overall core layer BMD, but further corresponds to multiple directional regional BMDs. This processing preserves the overall density level of the vertebral body while also representing the distribution differences between local regions. For example, when the BMD of the upper region of a target vertebra is significantly lower than that of the lower region, this difference will be directly reflected in the regional statistical results of this step.

[0063] In step 44, the absolute value of the difference between the upper and lower bone mineral density regions is determined as the cephalothorax density difference. This cephalothorax density difference is then divided by the overall core bone mineral density to obtain the cephalothorax density metric. Similarly, the absolute value of the difference between the anterior and posterior column bone mineral density regions is determined as the anteroposterior density difference. This anteroposterior density metric is then divided by the overall core bone mineral density to obtain the anteroposterior density metric. Here, the absolute value of the difference is taken first to represent only the degree of difference without distinguishing which side is higher. Using the overall core bone mineral density as a normalization benchmark ensures that the directional differences between different target vertebrae are on a uniform scale.

[0064] In other words, when the overall nuclear bone mineral density of the target vertebral body is high but the local differences are small, the loss of mass remains low; when the local differences increase, the loss of mass increases accordingly. Through the above processing, the originally dispersed regional bone mineral density results are further transformed into directional distribution parameters that can be directly used in subsequent analyses.

[0065] Through the above implementation process, this embodiment further organizes the equivalent bone mineral density data into overall core layer bone mineral density, overall shell layer bone mineral density, bone mineral density of each corresponding region, and directional density loss measure based on the established region boundaries. This maintains a clear hierarchy of target vertebral body density statistics and extracts the distribution variations in different directions from the voxel-level information. Subsequently, when determining the results of low-density voxel sets, low-density connectivity disruption, shell-core loss measure, and single vertebral body analysis, these results can be directly invoked, thus establishing automatic bone mineral density measurement and osteoporosis analysis on a consistent and continuous regional statistical chain.

[0066] In one embodiment of the present invention, the low-density voxel set, low-density connectivity disruption, and shell-nucleus mismatch are determined based on the dispersion of the overall core layer bone density and the equivalent bone density within the cancellous bone core layer. This part is based on the aforementioned equivalent bone density volume data, statistical results of the cancellous bone core layer, shell layer, and overall core layer and shell layer bone densities, and mainly addresses whether there is dispersion, aggregation, and interlayer imbalance in the density distribution within the target vertebral body. Here, the low-density voxel set refers to the set of voxels below a threshold within the cancellous bone core layer; the low-density connectivity disruption refers to the proportion of low-density voxels that have formed spatial connectivity aggregation; and the shell-nucleus mismatch represents the relative density deviation between the shell layer and the cancellous bone core layer.

[0067] In step 51, based on the equivalent bone mineral density (BMD) data corresponding to each target vertebra, the difference between the equivalent BMD values ​​of all voxels and the overall core layer BMD is statistically analyzed within the cancellous bone core layer. Specifically, the difference between the equivalent BMD value of each voxel and the overall core layer BMD is first squared, then all squared results are summed, averaged using the total number of voxels within the cancellous bone core layer, and finally the square root of the average is taken to obtain the core layer dispersion. Here, the core layer dispersion is a uniform representation of the density fluctuation range within the cancellous bone core layer. The core layer dispersion can be expressed as: ,in, Indicates the first The degree of dispersion of the nuclear layer corresponding to each target vertebral body Indicates the first Voxels within the target vertebra The equivalent bone mineral density value, Indicates the first The overall nucleus bone density corresponding to each target vertebral body Indicates the first The cancellous bone nucleus layer corresponding to each target vertebral body The total number of voxels within the cancellous bone core layer indicates that if the core layer dispersion is small, the density distribution within that region is relatively concentrated; if the core layer dispersion is large, it indicates that the local density fluctuations are more obvious. Through the above processing, the voxel-level differences originally dispersed within the cancellous bone core layer are merged into a single discrete parameter that can be directly accessed.

[0068] In step 52, a preset discrete threshold coefficient is first obtained, and then this coefficient is combined with the nuclear layer dispersion to form the low-density voxel threshold corresponding to the current target vertebra. The preset discrete threshold coefficient refers to a pre-set proportional parameter used to control the influence of the dispersion on the threshold adjustment range. Subsequently, all voxels with equivalent bone mineral density values ​​lower than the low-density voxel threshold are extracted in the cancellous bone core layer region to obtain a low-density voxel set. That is to say, this step does not directly use a fixed density value as a screening criterion, but rather forms a low-density screening threshold corresponding to the density distribution of the current vertebra based on the overall nuclear layer bone density and nuclear layer dispersion of the current target vertebra. This can maintain the consistency of processing between different target vertebrae and avoid the deviation caused by a single fixed threshold. For example, when the overall nuclear layer bone density of a target vertebra is low and the internal fluctuation is large, the low-density voxel threshold will be adjusted accordingly, and the subsequently extracted low-density voxel set will be more consistent with the actual distribution of the current vertebra.

[0069] In step 53, the low-density voxel set is partitioned into connected components according to a preset three-dimensional adjacency relationship, resulting in each low-density connected component. The preset three-dimensional adjacency relationship refers to the adjacency rules used in three-dimensional volume data to determine whether voxels are connected to each other. After partitioning the connected components, the total number of voxels in each low-density connected component is summed and divided by the total number of voxels in the cancellous bone core layer to obtain the low-density connectivity disruption amount. The low-density connectivity disruption amount can be expressed as: in, Indicates the first The amount of low-density connectivity disruption corresponding to each target vertebral body represents the total proportion of low-density connected domains within the cancellous bone nucleus layer. Indicates the first The first target vertebrae corresponding to the first A low-density connected domain, Indicates the first The total number of voxels within a low-density connected region; through this processing, the originally scattered low-density voxels are further transformed into a representation of the degree of spatial clustering. That is, the low-density connectivity disruption quantity considers not only the quantity of low-density voxels, but also whether these voxels form a continuous cluster in space. If there are many low-density voxels but they are scattered among each other, the quantity is relatively low; if low-density voxels form a large connected region, the quantity is relatively high.

[0070] In step 54, the overall shell bone density is subtracted from the overall nucleus bone density to obtain the shell-nucleus density difference. This difference is then divided by the overall shell bone density to obtain the shell-nucleus mismatch. The shell-nucleus density difference refers to the difference in average density between the shell region and the cancellous bone nucleus region; the shell-nucleus mismatch is the result of normalizing this difference based on the shell region density benchmark. This process allows for comparison of the shell-nucleus density deviations between different target vertebrae on a unified scale.

[0071] It should be noted that the shell-nucleus mismatch does not simply indicate that the bone mineral density in the shell region is higher than that in the nucleus region, but rather indicates the degree of relative deviation between the two, thus providing directly usable interlaminar difference parameters for subsequent single vertebral body analysis.

[0072] Through the above implementation process, this embodiment, based on the statistical analysis of overall core layer bone mineral density and overall shell layer bone mineral density, further categorizes the density distribution within the target vertebral body into four types of results: core layer dispersion, low-density voxel aggregation, low-density connectivity disruption, and shell-core misalignment. After this processing, subsequent steps in determining the load-bearing capacity of weak sections and judging the results of single vertebral body analysis can not only access the average density information but also simultaneously access the local low-density aggregation and shell-core layer deviation, thus making the data chain between automatic bone mineral density measurement and osteoporosis analysis more complete.

[0073] In one embodiment of the present invention, axial sections are extracted along the head-to-tail direction of each target vertebra, and the load-bearing capacity of the weak section is determined based on the density-area quantity and density-inertia quantity of each axial section. This part builds upon the aforementioned equivalent bone mineral density data, vertebral body contour, overall core layer bone mineral density, directional density loss, and structural degradation results. The key processing point is to further transform the density distribution inside the target vertebra into load-bearing information at the section level. Here, the axial section refers to the transverse section extracted layer by layer along the head-to-tail direction of the target vertebra and located within the vertebral body contour; the standard voxel corresponding section area refers to the area of ​​a single standard voxel on the axial section; the density-area quantity refers to the result of accumulating the equivalent bone mineral density values ​​of all voxels within the section with their corresponding areas; the density-inertia quantity refers to the result of accumulating the distribution of the equivalent bone mineral density values ​​relative to the geometric center of the section with the area information; and the load-bearing capacity of the weak section refers to the load-bearing result corresponding to the section with the lowest load-bearing capacity among all axial sections.

[0074] Specifically, in step 61, all axial sections located within the main contour of the vertebral body are extracted along the head-to-tail direction of each target vertebra. The main contour of the vertebral body is the boundary of the vertebral body retained after the removal of accessory bony structures in the previous steps; therefore, the axial sections extracted in this step are all limited to the effective range of the target vertebral body. Subsequently, the product of the standard voxel side length and the standard voxel side length is determined as the cross-sectional area corresponding to the standard voxel. That is, under the current unified voxel scale, each voxel occupies a fixed area in the axial plane and can directly participate in subsequent cross-sectional statistics. Further, the geometric center of each cross-section is determined in all axial cross-sections, and the maximum distance from the geometric center of the cross-section to the cross-sectional boundary is determined as the maximum cross-sectional radius. The maximum cross-sectional radius here is not a random value for a particular local cross-section, but rather the maximum result obtained after a unified comparison of the geometric scales of all axial cross-sections, thus providing a consistent dimensional benchmark for subsequent load-bearing capacity calculations.

[0075] In step 62, for each axial section, the equivalent bone mineral density value of each voxel within the section is multiplied by the corresponding cross-sectional area of ​​the standard voxel, and then all products are summed to obtain the density-area quantity of each axial section. The density-area quantity can be expressed as: ,in, Indicates the first The first target vertebrae corresponding to the first Density area of ​​a single axial cross section Indicates the first The first target vertebrae corresponding to the first Axial section Indicates the first Voxels within the target vertebra The equivalent bone mineral density value, This represents the cross-sectional area corresponding to a standard voxel. The density area quantity represents the cumulative result of the combined effect of bone density distribution and cross-sectional area within the current axial cross-section; after this processing, the density scale differences between different cross-sections can be represented in a unified manner.

[0076] For example, when a cross-sectional area is large and the internal equivalent bone mineral density level is high, its density area will also increase accordingly; when the cross-sectional area is small or the local density is low, the corresponding density area will be lower.

[0077] In step 63, after obtaining the density-area quantity of each axial section, the equivalent bone density value of each voxel, the square of the distance from the voxel to the geometric center of the axial section, and the corresponding cross-sectional area of ​​the standard voxel are multiplied for each axial section, and all products are summed to obtain the density inertia quantity. The density inertia quantity can be expressed as: ,in, Indicates the first The first target vertebrae corresponding to the first Density inertia of a single axial section Voxel representation To the The distance between the geometric centers of each axial section This represents the cross-sectional area corresponding to a standard voxel. The density inertia quantity represents the distribution of bone density within the current axial cross-section relative to the geometric center of the cross-section. The farther the voxel distance, the larger the corresponding squared distance term, and the greater its influence on the density inertia quantity, thus representing the degree of spatial distribution of bone density within the cross-section.

[0078] Subsequently, the product of density area and a first preset weight is determined as the first load-bearing term, and the product of density inertia and a second preset weight is divided by the square of the maximum cross-sectional radius to determine the second load-bearing term. The sum of the first and second load-bearing terms is then determined as the cross-sectional load-bearing capacity. The cross-sectional load-bearing capacity can be expressed as: ,in, Indicates the first The first target vertebrae corresponding to the first Section bearing capacity of each axial section Indicates the first preset weight. Indicates the first The first target vertebrae corresponding to the first Density area of ​​a single axial cross section This indicates the second preset weight. Indicates the first The first target vertebrae corresponding to the first Density inertia of a single axial section Indicates the first The maximum cross-sectional radius of each target vertebra; the cross-sectional load capacity represents the overall load-bearing level of the current axial cross-section. The first preset weight and the second preset weight are preset parameters used to adjust the influence of cross-sectional area factors and spatial distribution factors. Through the above processing, the cross-sectional load capacity can reflect both the overall density scale of the cross-section and the degree of density expansion relative to the geometric center, thus making the cross-sectional load capacity representation more complete.

[0079] In step 64, the load-bearing capacity of all cross-sections corresponding to each target vertebra is compared, and the minimum value is extracted to obtain the load-bearing capacity of the weakest section. This weakest section load-bearing capacity is not the average of all cross-section load-bearing capacities, but rather specifically corresponds to the axial section with the lowest load-bearing capacity. In other words, this step focuses not on the overall load-bearing capacity of the target vertebra, but on the load-bearing capacity of its weakest point.

[0080] For example, when most sections of a target vertebra have a high load-bearing capacity, but a certain local section has the lowest load-bearing capacity due to low density or deviation in distribution, this step extracts the lowest value, rather than performing equalization processing on all sections.

[0081] Through the above implementation process, this embodiment further organizes the dispersed equivalent bone mineral density information within the target vertebral body into load-bearing results at the axial cross-sectional level, and finally extracts the load-bearing capacity of the weakest cross-section. After this processing, when determining the results of single vertebral body analysis, it is possible not only to call up the overall core layer bone mineral density, directional density loss, and structural degradation, but also to simultaneously call up the load-bearing results of the weakest cross-section of the target vertebral body. This maintains the consistency between the automatic bone mineral density measurement chain and the regional analysis chain, and also makes the osteoporosis analysis closer to the actual distribution state inside the target vertebral body.

[0082] In one embodiment of the present invention, the single-vertebral analysis results corresponding to each target vertebra are determined based on the overall core layer bone density, the density imbalance in the head-to-tail direction, the density imbalance in the anterior-posterior direction, the low-density connectivity disruption, the shell-core imbalance, and the weak cross-sectional bearing capacity. This part is based on the aforementioned density statistics, regional imbalance results, low-density aggregation results, and weak cross-sectional bearing capacity results. By unifying the multiple types of quantitative results obtained, a unified decision chain is formed to generate an analysis conclusion for a single target vertebra.

[0083] The core layer bone mineral density threshold here refers to a preset limit used to determine whether the overall core layer bone mineral density is too low; the cephalocaudal density threshold and the anteroposterior density threshold refer to preset limits used to determine whether there is significant uneven density distribution in the vertical and anteroposterior directions of the target vertebral body; the low-density connectivity disruption threshold refers to a preset limit used to determine whether the overall proportion of low-density connected regions is too high; the shell-nucleus imbalance threshold refers to a preset limit used to determine whether the density deviation between the shell region and the cancellous bone-nucleus layer region exceeds the normal range; and the load-bearing capacity threshold refers to a preset limit used to determine whether the load-bearing capacity of the weakest section of the target vertebral body is too low. These thresholds can be pre-defined based on vertebral body level, sample statistical results, or preset analysis standards to ensure that subsequent judgments maintain a consistent standard across different target vertebrae.

[0084] Specifically, the following steps are taken: first, core layer bone mineral density thresholds, cephalocaudal density thresholds, anteroposterior density thresholds, low-density connectivity disruption thresholds, putamen imbalance thresholds, and load-bearing capacity thresholds are obtained. Then, the overall core layer bone mineral density is compared with these core layer bone mineral density thresholds. When the overall core layer bone mineral density is lower than the core layer bone mineral density threshold, this is identified as evidence of low density. This low-density evidence indicates that the target vertebral body has experienced a decrease in density at the overall cancellous bone core layer level; it is not a localized, single-point abnormality, but rather a general decrease in density surrounding the cancellous bone core layer region.

[0085] Based on this, the density imbalance in the cephalothorax direction is compared with the cephalothorax density threshold, and the density imbalance in the anteroposterior direction is compared with the anteroposterior density threshold. When the density imbalance in the cephalothorax direction is higher than the cephalothorax density threshold, or the density imbalance in the anteroposterior direction is higher than the anteroposterior density threshold, this situation is identified as evidence of directional imbalance. This evidence of directional imbalance indicates that the density distribution within the target vertebral body is no longer relatively uniform, but rather shows differences exceeding a preset range between upper and lower layers or anterior and posterior zones. In other words, even if the overall core layer bone density has not yet reached its minimum level, as long as the directional distribution has significantly deviated, it will be included in the subsequent assessment in this step.

[0086] Next, the amount of low-density connectivity disruption was compared to a low-density connectivity disruption threshold, and the amount of shell-nuclear misalignment was compared to a shell-nuclear imbalance threshold. When the amount of low-density connectivity disruption exceeded the low-density connectivity disruption threshold, or the amount of shell-nuclear misalignment exceeded the shell-nuclear imbalance threshold, this was identified as evidence of structural damage. This evidence of structural damage encompasses both large-scale connectivity anomalies caused by low-density voxel aggregation within the cancellous bone core layer and interlaminar density deviations between the shell and cancellous bone core layers.

[0087] Subsequently, the load-bearing capacity of the weak section is compared with a load-bearing capacity threshold. When the load-bearing capacity of the weak section is lower than the load-bearing capacity threshold, this situation is identified as evidence of insufficient load-bearing capacity. Here, evidence of insufficient load-bearing capacity indicates that the overall load-bearing capacity of the weakest section in the target cone has dropped below a preset range. This result does not depend on a single voxel or a single region, but is based on the comprehensive processing of cross-sectional layer density area and density inertia, thus forming a complementary relationship with the aforementioned overall density and structural distribution results.

[0088] After the above evidence extraction is completed, the single vertebral body analysis results are determined according to the preset evidence judgment order. In one embodiment, when low-density evidence, structural damage evidence, and insufficient load-bearing evidence are all valid, the target vertebral body can be identified as structurally fragile and enhanced; when low-density evidence is valid, and at least one of the directional imbalance evidence and structural damage evidence is valid, the target vertebral body can be identified as osteoporotic; when only low-density evidence is valid, the target vertebral body can be identified as hypoplastic; when none of the above evidence meets the corresponding judgment conditions, the target vertebral body can be identified as normal. Through this sequential judgment, the evidence formed in the previous step will be used in the next step, thus maintaining a clear source relationship and stable judgment logic for the single vertebral body analysis results.

[0089] Through the above implementation process, this embodiment integrates overall core layer bone density, directional distribution anomalies, local low-density aggregation, shell-core layer deviation, and the load-bearing state of the weakest section into a single vertebral body analysis chain. This ensures that the analysis results of the target vertebra no longer rely on a single indicator but are based on a comprehensive judgment of multi-dimensional, homogeneous data. After this processing, the single vertebral body analysis results can reflect not only the overall density changes of the target vertebra but also the changes in internal spatial distribution and structural state, thus enabling subsequent summary analysis at the subject level to be based on more complete single vertebral body results.

[0090] In one embodiment of the present invention, based on the integrity index corresponding to each target vertebra, the overall nuclear layer bone density corresponding to each target vertebra is weighted and summarized, and combined with the low-density connectivity disruption, weak cross-sectional load-bearing capacity, and single vertebra analysis results, the osteoporosis analysis results at the subject level are determined; this part unifies the analysis results of multiple target vertebrae into a subject-level conclusion. The subject-level average nuclear layer bone density refers to the overall density result formed by weighting and summing based on the integrity index corresponding to each target vertebra; the subject-level low-density connectivity disruption refers to the result with the highest degree of low-density connectivity disruption among all target vertebrae; the subject-level weak cross-sectional load-bearing capacity refers to the result with the lowest load-bearing level among all target vertebrae; abnormal vertebrae refer to target vertebrae whose single vertebra analysis results show osteoporosis or structural fragility and enhancement.

[0091] In step 71, the integrity index of each target vertebra is first summed. Then, the integrity index of each target vertebra is divided by the sum to obtain the weighting coefficient for each target vertebra. That is, the higher the integrity index, the greater the weight of the corresponding target vertebra in the subsequent aggregation; the lower the integrity index, the smaller its influence in the aggregation. Subsequently, the weighting coefficient of each target vertebra is multiplied by the overall nuclear bone density of each target vertebra, and all products are summed to obtain the average nuclear bone density at the subject level. Through this processing, the overall nuclear bone density of multiple target vertebrae is consolidated into a unified subject-level density result, which simultaneously reflects the vertebral density level and the usability of the imaging.

[0092] In step 72, the maximum value of low-density connectivity disruption and the minimum value of weak section bearing capacity corresponding to each target vertebra are extracted to obtain the low-density connectivity disruption at the test object level and the weak section bearing capacity at the test object level, respectively. The former represents the degree of connectivity disruption with the most prominent structural abnormality in the test object, and the latter represents the minimum bearing capacity of the weakest section in the test object. Subsequently, target vertebrae with osteoporosis or enhanced structural fragility in the single vertebral analysis results are identified as abnormal vertebrae, and the number of abnormal vertebrae is counted; at the same time, the number of target vertebrae with osteopenia in the single vertebral analysis results is counted to obtain the number of osteopenias. Through the above processing, the single vertebral results are further organized into two test object-level quantity results, one representing the distribution of higher-level abnormalities and the other representing the distribution of osteopenias. For example, if two of the three target vertebrae are identified as osteoporosis, the number of abnormal vertebrae is two; if the other target vertebra is identified as osteopenia, the number of osteopenias is one.

[0093] In step 73, the threshold values ​​for low-density connectivity disruption and weak cross-section bearing capacity at the subject level are first obtained. When the number of abnormal vertebrae is not less than two, the subject-level osteoporosis analysis result is directly determined to be osteoporosis. This indicates that multiple target vertebrae within the subject have simultaneously reached a high abnormality level, and further supplementary judgment based on other conditions is no longer necessary. If the number of abnormal vertebrae is equal to one, the subject-level low-density connectivity disruption and weak cross-section bearing capacity are compared. When the subject-level low-density connectivity disruption is higher than the subject-level low-density connectivity disruption threshold, and the subject-level weak cross-section bearing capacity is lower than the subject-level weak cross-section bearing capacity threshold, the subject-level osteoporosis analysis result is also determined to be osteoporosis. In other words, even with only one abnormal vertebra, it is still necessary to combine the overall structural damage level and minimum bearing capacity of the subject for joint judgment, thereby avoiding overly broad or narrow conclusions based solely on a single abnormal result.

[0094] In step 74, the core layer bone mineral density threshold for the subject-level osteoporosis is obtained. If the subject-level osteoporosis determination criteria described in step 73 are not met, the next level of judgment is performed based on the subject-level average core layer bone mineral density and the number of bone loss cases. When the subject-level average core layer bone mineral density is lower than the subject-level core layer bone mineral density threshold, or the number of bone loss cases is not less than one, the subject-level osteoporosis analysis result is determined to be bone loss; other cases are determined to be normal. The judgment order here is to first determine whether osteoporosis has been reached, then determine whether it belongs to bone loss, and finally classify the remaining cases as normal. Through this sequential processing, a clear correspondence is maintained between the subject-level conclusion and the previous single vertebral body analysis results, and it also makes it easier to trace back to the specific target vertebra and specific quantitative parameters of the overall results.

[0095] Through the above implementation process, this embodiment integrates the integrity index, overall core layer bone density, low-density connectivity disruption, weak section bearing capacity, and single vertebral body analysis results into the subject-level analysis chain. This ensures that the final result is not simply a list of conclusions for multiple target vertebrae, but rather a holistic analysis result obtained after weighted aggregation, maximum / minimum value extraction, and stratified determination. This maintains the continuity of data generated in the preceding steps and ensures that the final osteoporosis analysis results are consistent with the image quality, density level, and structural state of each target vertebra.

[0096] This invention provides an automated system for measuring bone mineral density and analyzing osteoporosis from CT images, comprising:

[0097] The data preprocessing module is used to acquire CT body data corresponding to the target subject, process it to obtain candidate analysis body data, and determine the target vertebral body set and the integrity index corresponding to each target vertebral body;

[0098] The reference conversion module is used to extract the internal reference area for each target vertebra, determine the average CT value of the muscle region and the average CT value of the fat region; based on the average CT value of the muscle region, the average CT value of the fat region and the preset reference density, the original CT value of the voxels in each target vertebra is converted to obtain the equivalent bone mineral density data corresponding to each target vertebra.

[0099] The partitioning module is used to construct the cancellous bone core layer and shell layer based on the equivalent bone mineral density data corresponding to each target vertebra, and to partition the cancellous bone core layer.

[0100] The density statistics module is used to determine the overall core layer bone density, overall shell layer bone density, head-to-tail density loss, and anteroposterior density loss based on the results of the cancellous bone core layer, shell layer, and regional divisions.

[0101] The degradation analysis module is used to determine the low-density voxel set, the amount of low-density connectivity disruption, and the shell-core loss based on the dispersion of the overall core layer bone density and the equivalent bone density in the cancellous bone core layer region.

[0102] The load-bearing determination module is used to extract axial sections along the head-to-tail direction of each target vertebra and determine the load-bearing capacity of weak sections based on the density area and density inertia of each axial section; and determine the single vertebra analysis results corresponding to each target vertebra based on the overall core layer bone density, density loss in the head-to-tail direction, density loss in the anterior-posterior direction, low-density connectivity disruption, shell-core loss, and load-bearing capacity of weak sections.

[0103] The results summary module is used to perform weighted summation of the overall nuclear layer bone density corresponding to each target vertebra based on the integrity index of each target vertebra, and combine the low-density connectivity disruption, weak cross-sectional load-bearing capacity and single vertebra analysis results to determine the osteoporosis analysis results of the tested object.

[0104] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain a formula that is closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0105] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present embodiments, all of which are within the protection scope of the present embodiments.

Claims

1. An automatic method for measuring bone mineral density and analyzing osteoporosis from CT images, characterized in that, Includes the following steps: Step 1: Obtain CT body data corresponding to the target subject, process the data to obtain candidate analysis body data, and determine the target vertebral body set and the integrity index corresponding to each target vertebral body; Step 2: Extract the internal reference area for each target vertebral body, and determine the average CT value of the muscle region and the average CT value of the fat region; based on the average CT value of the muscle region, the average CT value of the fat region and the preset reference density, convert the original CT value of the voxels in each target vertebral body to obtain the equivalent bone mineral density data corresponding to each target vertebral body. Step 3: Construct the cancellous bone core layer and shell layer based on the equivalent bone mineral density data corresponding to each target vertebra, and divide the cancellous bone core layer into subdivisions; Step 4: Based on the results of the cancellous bone core layer, shell layer and partition, determine the overall core layer bone density, overall shell layer bone density, head-to-tail density loss and anteroposterior density loss. Step 5: Based on the dispersion of the overall core layer bone density and the equivalent bone density in the cancellous bone core layer region, determine the low-density voxel set, the low-density connectivity disruption amount, and the shell-core loss amount. Step 6: Extract axial sections along the head-to-tail direction of each target vertebra, and determine the load-bearing capacity of the weak section based on the density area and density inertia of each axial section; determine the single vertebra analysis results corresponding to each target vertebra based on the overall core layer bone density, density loss in the head-to-tail direction, density loss in the anterior-posterior direction, low-density connectivity disruption, shell-core loss, and load-bearing capacity of the weak section. Step 7: Based on the integrity index corresponding to each target vertebra, perform weighted summation on the overall nuclear layer bone density corresponding to each target vertebra, and combine the low-density connectivity disruption, weak cross-sectional load-bearing capacity and single vertebra analysis results to determine the osteoporosis analysis results of the tested object.

2. The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images according to claim 1, characterized in that, Acquire CT body data corresponding to the target subject, process it to obtain candidate analysis body data, and determine the target vertebral body set and the integrity index corresponding to each target vertebral body, including: Step 11: Obtain CT volume data, extract slice thickness, pixel spacing, slice number and spatial orientation information, perform three-dimensional reconstruction, standard voxel resampling, volume extraction, bed board removal and obvious metal artifact removal on the CT volume data, retain the trunk area and the soft tissue area around the vertebral body, and obtain candidate analysis volume data. Step 12: Extract bony regions based on candidate analysis volume data, separate multiple vertebral candidate regions according to connectivity, remove transverse processes, spinous processes and adjacent non-target bony structures from each vertebral candidate region to obtain the main outline of the vertebral body, and establish the head-to-tail sequence of the vertebral body in the direction from head to tail. Step 13: Based on the head-to-tail order of the vertebral bodies, the first to third lumbar vertebral bodies are preferentially identified as the target vertebral body set. When there are missing vertebral bodies from the first to the third lumbar vertebral bodies, continuous vertebral bodies are identified from the twelfth thoracic vertebral body to the fourth lumbar vertebral body. When there is only one vertebral body outline that is neither missing nor truncated, the vertebral body corresponding to that outline is identified as the target vertebral body set. Within the vertebral body outline corresponding to each target vertebral body, the total number of primes, the number of truncated voxels, and the number of residual artifact interference voxels are counted. The ratio of the sum of the number of truncated voxels and the number of residual artifact interference voxels to the total number of primes is used as the reduction factor. The difference between the reduction factor and the total number of primes is determined as the integrity index.

3. The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images according to claim 1, characterized in that, For each target vertebra, an internal reference region is extracted to determine the average CT value of the muscle region and the average CT value of the fat region. Based on the average CT values ​​of the muscle region, the average CT values ​​of the fat region, and the preset reference density, the original CT values ​​of voxels within each target vertebra are converted to obtain the equivalent bone mineral density data corresponding to each target vertebra, including: Step 21: Determine the middle layer based on the spatial span of the vertebral body contour in the head-to-tail direction corresponding to each target vertebral body, and select the middle layer and its adjacent upper and lower layers; within the middle layer and its adjacent upper and lower layers, determine the paravertebral muscle area based on the soft tissue area on both sides of the vertebral body contour, and determine the fat area based on the continuous adipose soft tissue area around the vertebral body contour, as an internal reference area. Step 22: Based on the internal reference area, remove the vascular area, calcified area and residual artifact area in the paravertebral muscle area and fat area; sum the original CT values ​​of all voxels in the paravertebral muscle area and divide by the total number of voxels to obtain the average CT value of the muscle area; sum the original CT values ​​of all voxels in the fat area and divide by the total number of voxels to obtain the average CT value of the fat area. Step 23: For each target vertebral body voxel, the difference between the original CT value and the average CT value of the fat region is taken as the first difference, the difference between the average CT value of the muscle region and the average CT value of the fat region is taken as the second difference, and the ratio of the first difference to the second difference is taken as the conversion factor; the difference between the preset reference density corresponding to the muscle region and the preset reference density corresponding to the fat region is multiplied by the conversion factor, and then added to the preset reference density corresponding to the fat region to obtain the equivalent bone mineral density value; Step 24: According to the spatial position of the voxels in each target vertebra, fill the equivalent bone mineral density value into the corresponding vertebral body contour range to obtain the equivalent bone mineral density data corresponding to each target vertebra; and associate and store the average CT value of the muscle region, the average CT value of the fat region, and the preset reference density with the equivalent bone mineral density data corresponding to each target vertebra.

4. The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images according to claim 1, characterized in that, Based on the equivalent bone mineral density data corresponding to each target vertebra, the cancellous bone core layer and shell layer are constructed, and the cancellous bone core layer is further divided into subdivisions, including: Step 31: Determine the average height of the vertebra based on the spatial span of the vertebral body contour in the head-to-tail direction corresponding to each target vertebra; determine the shell peeling thickness according to the larger value between the standard voxel side length and the preset minimum thickness; determine the endplate removal height according to the smaller value between the preset ratio value of the average height of the vertebra and the preset maximum height. Step 32: Based on the shell peeling thickness, perform equidistant peeling inward along the outer surface of the vertebral body contour corresponding to each target vertebra to obtain the internal candidate region; subtract the internal candidate region within the vertebral body contour corresponding to each target vertebra to obtain the shell region. Step 33: Based on the endplate removal height, remove the upper endplate interference area at the cephalic end of the internal candidate area and remove the lower endplate interference area at the caudal end of the internal candidate area; the remaining area after removing the upper and lower endplate interference areas is used to obtain the cancellous bone core layer area. Step 34: Based on the cancellous bone core layer, divide the upper, middle and lower layers along the head-to-tail direction, divide the anterior column and posterior column along the front-to-back direction, and divide the four quadrants within each axial layer with the geometric center of the cancellous bone core layer as the dividing center.

5. The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images according to claim 1, characterized in that, Based on the results of the cancellous bone core layer, shell layer, and regional divisions, the overall core layer bone mineral density, overall shell layer bone mineral density, cephalocaudal density loss, and anteroposterior density loss were determined, including: Step 41: Based on the equivalent bone mineral density data corresponding to each target vertebra, the total equivalent bone mineral density values ​​of all voxels in the cancellous bone core layer are counted and divided by the total number of voxels in the cancellous bone core layer to obtain the overall core layer bone mineral density. Step 42: Based on the equivalent bone mineral density data corresponding to each target vertebra, the sum of the equivalent bone mineral density values ​​of all voxels in the shell region is calculated and divided by the total number of voxels in the shell region to obtain the overall shell bone mineral density. Step 43: Based on the equivalent bone mineral density data corresponding to each target vertebra, the total equivalent bone mineral density values ​​of all voxels in the upper, middle, lower, anterior column, posterior column and four quadrant regions are calculated and divided by the total number of voxels in the corresponding region to obtain the bone mineral density of each corresponding region. Step 44: The absolute value of the difference between the upper and lower bone mineral density is determined as the cephalothorax density difference, and the cephalothorax density difference is divided by the overall core layer bone mineral density to obtain the cephalothorax density loss; the absolute value of the difference between the anterior and posterior column bone mineral density is determined as the anteroposterior density difference, and the anteroposterior density loss is divided by the overall core layer bone mineral density to obtain the anteroposterior density loss.

6. The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images according to claim 1, characterized in that, Based on the dispersion of overall core layer bone mineral density and equivalent bone mineral density within the cancellous bone core layer, the low-density voxel set, low-density connectivity disruption, and putamen loss were determined, including: Step 51: Based on the equivalent bone mineral density data corresponding to each target vertebra, the sum of squares of the differences between the equivalent bone mineral density values ​​of all voxels in the cancellous bone nucleus layer and the overall nucleus layer bone mineral density is calculated, and then the square root is taken after dividing by the total number of voxels in the cancellous bone nucleus layer to obtain the degree of nucleus layer dispersion. Step 52: Obtain the preset discrete threshold coefficient, use the product of the preset discrete threshold coefficient and the degree of dispersion of the core layer as the reduction amount, subtract the reduction amount from the overall core layer bone density to obtain the low-density voxel threshold; extract voxels with equivalent bone density values ​​lower than the low-density voxel threshold in the cancellous bone core layer area to obtain the low-density voxel set. Step 53: Divide the low-density voxel set into connected domains according to the preset three-dimensional adjacency relationship to obtain each low-density connected domain; count the total number of voxels in each low-density connected domain and divide it by the total number of voxels in the cancellous bone core layer to obtain the low-density connectivity destruction amount. Step 54: Subtract the total shell bone density from the total core bone density to obtain the shell-core density difference, and divide the shell-core density difference by the total shell bone density to obtain the shell-core loss.

7. The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images according to claim 1, characterized in that, Axial sections are extracted along the head-to-tail direction of each target vertebra. The load-bearing capacity of the weak sections is determined based on the density-area quantity and density-inertia quantity of each axial section, including: Step 61: Extract all axial sections located within the main contour of each target vertebra along the head-tail direction. Determine the cross-sectional area corresponding to the standard voxel by the product of the standard voxel side length and the maximum distance from the geometric center of the cross-section to the boundary of the cross-section among all axial sections. Step 62: For each axial section, multiply the equivalent bone density value of each voxel in the section by the corresponding cross-sectional area of ​​the standard voxel, and sum all the products to obtain the density area of ​​each axial section. Step 63: For each axial section, multiply the equivalent bone density value of each voxel, the square of the distance from the voxel to the geometric center of the axial section, and the cross-sectional area corresponding to the standard voxel, respectively, and sum all the products to obtain the density inertia; determine the first load-bearing term by multiplying the density area by the first preset weight, determine the second load-bearing term by dividing the product of the density inertia by the second preset weight by the square of the maximum cross-sectional radius, and determine the cross-sectional load-bearing capacity by summing the first load-bearing term and the second load-bearing term; Step 64: Compare the load-bearing capacity of all sections corresponding to each target vertebra, extract the minimum value, and obtain the load-bearing capacity of the weak section.

8. The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images according to claim 1, characterized in that, Based on overall core bone mineral density, head-to-tail density loss, anteroposterior density loss, low-density connectivity disruption, shell-core loss, and weak cross-sectional load-bearing capacity, the individual vertebral body analysis results for each target vertebra are determined, including: The following thresholds were obtained: core layer bone mineral density threshold, cephalo-coccygeal density threshold, antero-posterior density threshold, low-density connectivity disruption threshold, shell-core imbalance threshold, and load-bearing capacity threshold. Overall core layer bone mineral density below the core layer bone mineral density threshold was identified as evidence of low density. Cephalo-coccygeal density imbalance above the cephalo-coccygeal density threshold or antero-posterior density imbalance above the antero-posterior density threshold was identified as evidence of directional imbalance. Low-density connectivity disruption above the low-density connectivity disruption threshold or shell-core imbalance above the shell-core imbalance threshold was identified as evidence of structural damage. Weak section load-bearing capacity below the load-bearing capacity threshold was identified as evidence of insufficient load-bearing capacity. The single vertebral body analysis results were determined according to a preset evidence judgment order.

9. The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images according to claim 1, characterized in that, Based on the integrity index corresponding to each target vertebra, the overall nuclear layer bone mineral density corresponding to each target vertebra is weighted and summarized. Combined with the low-density connectivity disruption, weak cross-sectional load-bearing capacity, and single vertebra analysis results, the osteoporosis analysis results for the examined subject are determined, including: Step 71: Calculate the sum of the integrity indices corresponding to each target vertebra, and divide the integrity index corresponding to each target vertebra by the sum to obtain the weighting coefficient corresponding to each target vertebra; multiply the weighting coefficient corresponding to each target vertebra by the overall nuclear layer bone density corresponding to each target vertebra, and sum all the products to obtain the average nuclear layer bone density of the tested object. Step 72: Extract the maximum value of low-density connectivity disruption and the minimum value of weak cross-section bearing capacity corresponding to each target vertebra, and obtain the low-density connectivity disruption and weak cross-section bearing capacity at the level of the tested object, respectively; identify the target vertebrae with osteoporosis or structural fragility enhancement in the single vertebral analysis as abnormal vertebrae, and count the number of abnormal vertebrae; count the number of target vertebrae with reduced bone mass in the single vertebral analysis, and obtain the number of reduced bone mass. Step 73: Obtain the low-density connectivity disruption threshold and the weak section bearing capacity threshold at the subject level; when the number of abnormal vertebrae is not less than two, the subject-level osteoporosis analysis result is determined to be osteoporosis; when the number of abnormal vertebrae is equal to one, and the low-density connectivity disruption threshold at the subject level is higher than the low-density connectivity disruption threshold at the subject level, and the weak section bearing capacity threshold at the subject level is lower than the weak section bearing capacity threshold at the subject level, the subject-level osteoporosis analysis result is determined to be osteoporosis. Step 74: Obtain the core layer bone mineral density threshold for the subject-level osteoporosis. If the conditions for determining osteoporosis at the subject-level described in Step 73 are not met, the subject-level osteoporosis analysis result is determined to be osteoporosis if either the average core layer bone mineral density of the subject-level is lower than the core layer bone mineral density threshold for the subject-level or the number of cases of bone loss is not less than one. Otherwise, the result is determined to be normal.

10. An automatic system for measuring bone mineral density and analyzing osteoporosis from CT images, characterized in that, The method for automatic measurement of bone mineral density and osteoporosis analysis of CT images as described in any one of claims 1-9 includes: The data preprocessing module is used to acquire CT body data corresponding to the target subject, process it to obtain candidate analysis body data, and determine the target vertebral body set and the integrity index corresponding to each target vertebral body; The reference conversion module is used to extract the internal reference area for each target vertebra, determine the average CT value of the muscle region and the average CT value of the fat region; based on the average CT value of the muscle region, the average CT value of the fat region and the preset reference density, the original CT value of the voxels in each target vertebra is converted to obtain the equivalent bone mineral density data corresponding to each target vertebra. The partitioning module is used to construct the cancellous bone core layer and shell layer based on the equivalent bone mineral density data corresponding to each target vertebra, and to partition the cancellous bone core layer. The density statistics module is used to determine the overall core layer bone density, overall shell layer bone density, head-to-tail density loss, and anteroposterior density loss based on the results of the cancellous bone core layer, shell layer, and regional divisions. The degradation analysis module is used to determine the low-density voxel set, the amount of low-density connectivity disruption, and the shell-core loss based on the dispersion of the overall core layer bone density and the equivalent bone density in the cancellous bone core layer region. The load-bearing determination module is used to extract axial sections along the head-to-tail direction of each target vertebra and determine the load-bearing capacity of weak sections based on the density area and density inertia of each axial section; and determine the single vertebra analysis results corresponding to each target vertebra based on the overall core layer bone density, density loss in the head-to-tail direction, density loss in the anterior-posterior direction, low-density connectivity disruption, shell-core loss, and load-bearing capacity of weak sections. The results summary module is used to perform weighted summation of the overall nuclear layer bone density corresponding to each target vertebra based on the integrity index of each target vertebra, and combine the low-density connectivity disruption, weak cross-sectional load-bearing capacity and single vertebra analysis results to determine the osteoporosis analysis results of the tested object.

Images