Bone mineral density detection method and device based on CT image and storage medium

Abstract

The invention discloses a bone mineral density detection method and device based on a CT image and a storage medium, and relates to the technical field of medical image data processing. The method comprises the following steps: firstly, acquiring a scanning data sequence of a plain-scan CT covering a target vertebral body; and delineating the region of interest of the cancellous bone on a plurality of parallel layers of the same target vertebral body to reduce the influence of interference factors. Linear transformation is carried out according to the scanning data sequence, the HU value of each pixel in the region of interest is converted pixel by pixel, and the CT mean value of the target vertebral body is obtained based on the HU values. And then determining a relative bone mineral density index of the target vertebral body based on the CT mean value by taking a preset reference value as a benchmark. And finally, based on the relative bone mineral density index, performing bone mineral density grading according to a set threshold value to obtain an evaluation result corresponding to each target vertebral body. Therefore, the bone mineral density detection method is simple in process, low in cost and universal.

Description

Technical Field

[0001] This application relates to the field of medical image data processing technology, specifically to a method, device, and storage medium for bone density detection based on CT images. Background Technology

[0002] Osteoporosis is a systemic metabolic bone disease characterized by decreased bone mass and deterioration of bone microstructure. It significantly increases the risk of fragility fractures, leading to increased pain, functional impairment, disability, and even mortality, thus having a significant impact on public health and the quality of life of the elderly. Therefore, establishing repeatable and quantifiable bone mineral density assessment methods for early screening, fracture risk assessment, and treatment follow-up is a crucial foundation for the standardized management of osteoporosis.

[0003] Currently, among the techniques used in clinical practice for quantitative bone mineral density assessment, quantitative computed tomography (QCT) can output quantitative results such as volumetric bone mineral density (vBMD), which has the advantage of reflecting changes in vertebral cancellous bone mineral density. Therefore, it is often used as one of the commonly used techniques for quantitative bone mineral density assessment and research control. However, there are still objective limitations in the widespread application of QCT: its quantitative results usually rely on hydroxyapatite calibration phantoms or equivalent calibration procedures, and require dedicated quantitative analysis software to calculate parameters and determine grading. The overall process has high requirements for equipment configuration, operating procedures, and quality control, resulting in relatively high costs and implementation barriers, leading to limited adoption rates in primary care or resource-constrained institutions. At the same time, many plain CT images obtained in routine clinical pathways are often not configured with phantoms or have undergone standardized quantitative analysis according to the QCT process, making it difficult to use them directly for batch and stable quantitative bone mineral density assessment, thus resulting in the underutilization of existing imaging resources.

[0004] On the other hand, opportunistic bone mineral density (BMD) assessment methods based on plain computed tomography (CT) images have the potential value of reusing existing images without additional scans. However, existing methods face core technical challenges: while Hounsfield units (HUs) in plain CT images are related to bone density, HUs are significantly affected by multiple factors such as the scanning equipment model, tube voltage, slice thickness, reconstruction algorithm / kernel, and iterative reconstruction parameters. Under different equipment or parameter conditions, the same BMD level may correspond to different HU distributions, resulting in systematic drift. Therefore, directly using fixed HU thresholds or simple empirical thresholds for grading often lacks consistency and reproducibility across equipment and centers, limiting the clinical application of plain CT BMD assessment methods. Summary of the Invention

[0005] The purpose of this application is to provide a bone density detection method, device, and storage medium based on CT images, in order to solve the problems of traditional bone density detection methods being complex, costly, and lacking universality.

[0006] To achieve the above objectives, the first aspect of this application provides a bone density detection method based on CT images, comprising: Acquire the scan data sequence of plain CT covering the target vertebral body; Regions of interest in cancellous bone are delineated in multiple parallel planes of the same target vertebral body, wherein the parallel planes are multiple axial planes of the target vertebral body and are matched with the direction of the scanning plane; A linear transformation is performed on the scan data sequence to calculate the HU value of each pixel in the region of interest pixel by pixel, and the average CT value of the target vertebra is obtained based on the HU value. Using a preset reference value as a benchmark, the relative bone mineral density index of the target vertebra is determined based on the mean CT value. The preset reference value is a reference value for measuring the deviation of the bone mineral density index from the benchmark level. Based on the relative bone mineral density index, bone mineral density is graded according to a set threshold to obtain the evaluation result for each target vertebra. The set threshold is a grading threshold determined based on the reference standard. The evaluation result includes the mean CT value of the target vertebra, the relative bone mineral density index, and the grading result. The relative bone mineral density index satisfies the following formula: ; in, The relative bone mineral density index, The mean CT value is... This is the preset reference value.

[0007] A second aspect of this application provides a bone density detection device based on CT images, comprising: The acquisition module is used to acquire the scan data sequence of plain CT scans covering the target vertebral body; The delineation module is used to delineate regions of interest in cancellous bone on multiple parallel planes of the same target vertebra, wherein the parallel planes are perpendicular to the long axis of the target vertebra. The transformation module is used to perform linear transformation according to the scan data sequence, calculate the HU value of each pixel in the region of interest pixel by pixel, and obtain the CT mean value of the target vertebra based on the HU value; The determination module is used to determine the relative bone mineral density index of the target vertebra based on the mean CT value, with a preset reference value as a benchmark. The preset reference value is a reference value for measuring the deviation of the bone mineral density index from the benchmark level. An evaluation module is used to perform bone mineral density grading based on the relative bone mineral density index and according to a set threshold to obtain an evaluation result for each target vertebra. The set threshold is a grading threshold determined based on the reference standard. The evaluation result includes the mean CT value of the target vertebra, the relative bone mineral density index, and the grading result. The relative bone mineral density index satisfies the following formula: ; in, The relative bone mineral density index, The mean CT value is... This is the preset reference value.

[0008] A third aspect of this application provides a computer-readable storage medium storing a program that can be loaded by a processor and executed by the above-described CT image-based bone density detection method.

[0009] The beneficial effects of this application are: This application directly reuses routine plain CT scan data sequences covering the target vertebral body for bone mineral density (BMD) detection. Regions of interest (ROIs) of cancellous bone are delineated on multiple parallel planes of the same target vertebral body, reducing interference from cortical bone, venous plexuses, sclerotic lesions, and metal artifacts. By introducing preset reference values ​​and a standardized relative BMD calculation formula, the mean CT value of the target vertebral body is converted into a dimensionless relative BMD index. This offsets systematic biases in HU values ​​caused by different equipment and scanning parameters, ensuring consistency and comparability of BMD assessment results across equipment and centers. Setting thresholds based on grading critical values ​​of reference standards aligns grading logic with reference standards, improving the accuracy and clinical acceptance of BMD grading. The assessment results can directly provide reference for physicians. The assessment results integrate the mean CT value, relative BMD index, and grading results, preserving original quantitative data while providing intuitive grading conclusions. This facilitates physicians' ability to trace the examination process, assess the degree of bone loss, and provides standardized quantitative evidence for monitoring subsequent treatment effects. Therefore, the BMD detection method of this application is simple, low-cost, and universally applicable.

[0010] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0011] Figure 1 This is a schematic diagram illustrating an application scenario of a bone density detection method based on CT images provided in this application embodiment; Figure 2This is a schematic flowchart of a bone density detection method based on CT images provided in an embodiment of this application; Figure 3 This is a schematic diagram of a bone density detection device based on CT images provided in an embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified. Details are set forth in the following description for illustrative purposes. It should be understood that those skilled in the art will recognize that this application can be implemented without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but rather to be consistent with the broadest scope of the principles and features disclosed herein.

[0014] The CT image-based bone density detection method in this embodiment is applied to a CT image-based bone density detection device, and the CT image-based bone density detection method is installed in an electronic device. For example... Figure 1 As shown, Figure 1 This is a schematic diagram of an application scenario for a bone density detection method based on CT images provided in this application embodiment. The application scenario of the bone density detection method based on CT images in this application embodiment includes an electronic device 110 for the bone density detection method based on CT images. The electronic device 110 integrates a bone density detection device based on CT images to run a computer-readable storage medium corresponding to the bone density detection method based on CT images, so as to execute the steps of the bone density detection method based on CT images.

[0015] Understandable Figure 1The electronic devices in the application scenarios of the bone density detection method based on CT images, or the devices contained in the electronic devices, do not constitute a limitation on the embodiments of this application. That is, the number or type of devices in the application scenarios of the bone density detection method based on CT images, or the number or type of devices contained in each device, do not affect the overall implementation of the technical solution in the embodiments of this application, and can all be considered as equivalent substitutions or derivatives of the technical solutions claimed in the embodiments of this application.

[0016] In this application embodiment, the electronic device 110 can be an independent device, or a device network or device cluster composed of devices. For example, the electronic device 110 described in this application embodiment includes, but is not limited to, a computer, a network host, a single network device, a set of multiple network devices, or a cloud device composed of multiple devices. Among them, the cloud device is composed of a large number of computers or network devices based on cloud computing.

[0017] Those skilled in the art will understand that Figure 1 The application scenarios shown are merely one application scenario corresponding to the technical solution of this application, and do not constitute a limitation on the application scenarios of the technical solution of this application. Other application scenarios may include more than one application scenario. Figure 1 The number of more or fewer electronic devices shown, or the network connections of electronic devices, for example Figure 1 Only one electronic device is shown in the diagram. It is understood that the scenario of the CT image-based bone density detection method may also include one or more other electronic devices, which are not specifically limited here. The electronic device 110 may also include a memory and a processor. The memory is used to store information related to the CT image-based bone density detection method.

[0018] Furthermore, in the application scenario of the CT image-based bone density detection method in this application embodiment, the electronic device 110 may be equipped with a display device, or the electronic device 110 may not have a display device but may be communicatively connected to an external display device 120. The display device 120 is used to output the results of the CT image-based bone density detection method executed in the electronic device. The electronic device 110 can access the background database 130. The background database 130 may be the local storage of the electronic device 110 or a cloud database located in the cloud. The background database 130 stores information related to the CT image-based bone density detection method.

[0019] It should be noted that, Figure 1 The application scenario of the CT image-based bone density detection method shown is merely an example. The application scenario of the CT image-based bone density detection method described in the embodiments of this application is to more clearly illustrate the technical solution of the embodiments of this application and does not constitute a limitation on the technical solution provided in the embodiments of this application.

[0020] Based on the application scenarios of the CT image-based bone density detection method described above, an embodiment of the CT image-based bone density detection method is proposed. A detailed description is provided below with reference to the accompanying drawings.

[0021] Figure 2 This is a schematic flowchart illustrating a bone density detection method based on CT images provided in an embodiment of this application. Figure 2 As shown, the bone density detection method based on CT images can be executed by the processor in the aforementioned electronic device 110, and steps 201-205 are described in detail below.

[0022] Step 201: Obtain the scan data sequence of plain CT covering the target vertebral body.

[0023] The target vertebra is the core object of bone mineral density (BMD) testing. For example, it is clinically recognized that the L1-L3 lumbar vertebrae or the T6 thoracic vertebrae have abundant cancellous bone, are sensitive to changes in BMD, and are less affected by spinal degeneration; therefore, the L1-L3 lumbar vertebrae can be preferentially selected. In practice, other vertebrae can be included based on clinical needs. Plain CT refers to a computed tomography (CT) scan that does not require the injection of contrast agents and uses X-rays to penetrate the human body for tomographic scanning.

[0024] This application embodiment does not require additional targeted scanning and can directly reuse the scanning data sequence of routine plain CT scans in forest farms, reducing patient radiation exposure and waste of medical resources caused by repeated scanning, and lowering detection costs. The scanning data sequence refers to the continuous image data set generated after a CT scan, typically stored in the DICOM (Digital Imaging and Communications in Medicine) standard format, and may include core data such as original pixel grayscale values ​​and scanning parameters (tube voltage, slice thickness, reconstruction kernel, etc.).

[0025] Step 202: Delineate the region of interest for cancellous bone at multiple parallel planes of the same target vertebra.

[0026] Parallel planes refer to multiple axial planes of the target vertebral body. Axial planes are tomographic planes aligned with the CT scan plane, representing the standard presentation perspective in clinical CT imaging. They clearly display the bone tissue distribution in the cross-section of the target vertebral body; therefore, the axial planes match the scanning plane's direction. In other words, parallel planes are multiple tomographic planes perpendicular to the long axis of the target vertebral body (i.e., the direction of the line connecting the superior and inferior endplates), and these planes are parallel to each other to ensure consistent scanning perspective. The Region of Interest (ROI) of cancellous bone is a specific area delineated from the cross-sectional image of the target vertebral body that contains only cancellous bone tissue. It is necessary to exclude interfering areas such as cortical bone, vertebral endplates, paravertebral soft tissues, and calcifications. This is the core area for bone mineral density quantification.

[0027] Standardized layer selection and ROI delineation logic can reduce subjective errors from manual layer selection or ROI delineation, improving consistency in testing across different operators and cases. Focusing on the cancellous bone region eliminates interference from non-target tissues such as cortical bone, venous plexuses, sclerotic lesions, and metal artifacts, ensuring that subsequent HU value conversion and mean calculation only reflect cancellous bone density characteristics, thus improving detection accuracy. Multiple parallel layers cover key areas of vertebral cancellous bone, comprehensively characterizing the bone density distribution of the target vertebra and reducing random errors caused by single-layer sampling.

[0028] Step 203: Perform linear transformation according to the scan data sequence, calculate the HU value of each pixel in the region of interest pixel by pixel, and obtain the CT mean value of the target vertebra based on the HU value.

[0029] Linear transformation refers to the linear calculation process of converting the original grayscale values ​​of pixels in scan data into HU values ​​based on the standardized calibration parameters of CT equipment, in order to restore the degree of X-ray attenuation of the tissue corresponding to the pixel. The HU value is the standard unit for quantifying the degree of X-ray attenuation of tissue in CT images. It is based on the attenuation coefficient of water (HU value of water = 0). The HU value of bone tissue is usually positive, and the higher the value, the greater the tissue density.

[0030] In one example, normalized calibration parameters can be read from the header of a DICOM format scan data sequence. These parameters, such as the rescale slope (S) and rescale intercept (K) parameters, can be written by the CT equipment after factory calibration to restore the true attenuation characteristics of the tissue. Next, the HU value for each ROI pixel is calculated based on a linear transformation formula.

[0031] The converted set of HU values ​​can be further denoised (e.g., by median filtering or nonlocal mean filtering). This denoising process must ensure that the core quantitative information of bone mineral density is not altered (i.e., the overall distribution characteristics of HU values ​​are preserved), and isolated outliers caused by scanning noise should be eliminated. The mean CT value of the target vertebra is then calculated based on the denoised HU values. For example, the mean HU value for each parallel ROI is calculated separately, and then the arithmetic mean of the mean HU values ​​for multiple ROIs is calculated, which is used as the final mean CT value of the target vertebra.

[0032] Linear transformation based on standard calibration parameters of the scan data sequence ensures the accuracy and standardization of HU value conversion, restores the true density characteristics of cancellous bone, and reduces quantification deviations caused by equipment parameters affecting the original grayscale values. Denoising processing and mean aggregation of multiple parallel slices effectively reduce scanning noise and random errors from single-slice sampling, improving the stability and reliability of CT mean values ​​and providing high-quality basic data for subsequent relative bone mineral density calculations.

[0033] Step 204: Using a preset reference value as a benchmark, determine the relative bone density index of the target vertebra based on the CT mean. The preset reference value is a reference value for measuring the deviation of the bone density index from the benchmark level.

[0034] The preset reference value refers to the distribution characteristics of HU values ​​of vertebral cancellous bone in a population with baseline bone mass in the quantitative CT clinical reference standard, used to eliminate equipment and parameter deviations. The core quantitative indicator of the traditional reference standard is the volumetric bone mineral density (vBMD), which directly reflects the absolute quantitative index of bone mineral content in vertebral cancellous bone. The measurement of vBMD must rely on a hydroxyapatite calibration phantom (scanned simultaneously with the patient). Through dedicated quantitative analysis software, the HU value of cancellous bone in the CT image is compared with the known density calibration curve of the phantom to calculate the bone mineral mass per unit volume in three-dimensional space. This method is costly, complex, and difficult to popularize. In contrast, the quantitative indicator of the embodiment of this application is the relative bone mineral density (cBMD), which is a dimensionless index that quantifies the degree to which the mean CT value of the target vertebra deviates from the baseline bone mass level based on the preset reference value. It is expressed as a percentage, with the closer the value is to 0, the closer the bone density is to the baseline level, and the larger the absolute value of the negative value, the more severe the bone loss. cBMD requires no phantom calibration and can be calculated directly based on conventional plain CT data, adapting to different CT equipment and scanning parameter scenarios.

[0035] In one example, the relative bone mineral density index can satisfy the following formula: ; in, This is a relative bone mineral density index. The mean CT value is... These are preset reference values. After calculation, range validation (e.g., excluding outliers outside the clinically reasonable range) can be performed to ensure the validity of the results. Outliers should prompt for re-examination of the data or ROI delineation.

[0036] By employing a dimensionless design, cBMD effectively eliminates systematic biases caused by different devices and scanning data, enabling cross-device and cross-center comparability and overcoming the bottleneck of traditional HU values ​​for directly assessing bone mineral density. It intuitively quantifies the degree to which bone mass deviates from the baseline level, making it easier for clinicians to understand than absolute HU values, facilitating rapid assessment of bone mineral density status, and providing clear quantitative evidence for grading evaluation.

[0037] Step 205: Based on the relative bone mineral density index, perform bone mineral density grading according to the set threshold to obtain the evaluation results corresponding to each target vertebra.

[0038] The threshold values ​​are based on grading cutoffs determined by reference standards. These cBMD grading boundary values, derived through a linear regression model and validated by Receiver Operating Characteristic Curve Analysis (ROC), correspond one-to-one with bone mineral density grades. Quantitative CT (QCT) reference standards refer to the measurement of vertebral cancellous bone volume mineral density using a dedicated calibration phantom and quantitative analysis software. Based on clinical standards, these standards can be categorized into baseline (normal), osteopenia, and osteoporosis, and possess clinical acceptance.

[0039] The primary grading threshold, aligned with the grading logic of the reference standard, is invoked to clearly define the cBMD intervals for bone mass. The calculated cBMD is compared with the set thresholds to determine the grading result of the target vertebral body's bone mineral density level, and then the evaluation results are integrated. The evaluation result refers to the sum of bone mineral density quantitative data and grading conclusions in the reported data, which may include the target vertebral body's mean CT value (raw quantitative data), relative bone mineral density index (relative quantitative data), and grading result (qualitative conclusion), meeting the needs of clinical diagnosis and follow-up. Furthermore, within the osteoporosis state interval, subdivision thresholds can be set according to reasonable gradients to further classify mild, severe, and extreme osteoporosis.

[0040] Setting thresholds and precisely aligning them with reference standards ensures the clinical acceptance and diagnostic accuracy of grading results. Assessment conclusions can directly provide diagnostic references for physicians without additional cross-validation, improving clinical application efficiency. Standardizing the grading logic reduces differences in grading standards across different medical institutions, enabling a unified interpretation of grading results. Complete assessment results retain original quantitative data for easy traceability while providing relative quantitative indicators and qualitative grading conclusions, balancing the accuracy and intuitiveness of clinical diagnosis, and providing standardized quantitative comparisons for subsequent treatment effect monitoring.

[0041] In step 201, the plain CT data in this embodiment may include non-historical plain CT data sequences and historical plain CT data sequences. Non-historical plain CT data sequences refer to a collection of newly acquired plain CT images for the patient based on the clinical needs of this bone density test, without prior medical data archiving, and directly used for this test analysis. Historical plain CT data refers to previous plain CT scan data that has been archived in a medical image storage system (Picture Archiving and Communication Systems, PACS), and can be reused for this bone density test without additional scanning. Data acquisition can be based on clinical testing instructions. After determining the target vertebral body scanning range, a plain CT scan is completed according to standard clinical parameters, and a DICOM format scan data sequence is generated in response to the scan completion instruction. Historical plain CT data can be searched from the PACS using patient identification, examination date, and target vertebral body as search keywords to filter historical plain CT DICOM sequences covering the target vertebral body. These will be described in detail below.

[0042] If the scan data sequence of a plain CT scan is not a historical plain CT scan data sequence, the target vertebral body's scanning range is determined and the scanning parameters are configured based on the plain scan examination instruction. The plain scan examination instruction is a medical instruction issued by a clinician based on the patient's diagnostic and treatment needs (such as osteoporosis screening, investigation of the cause of low back pain, etc.). It clearly includes core information such as the examination site (i.e., the target vertebral body, such as the lumbar vertebrae L1-L3) and the examination type (plain scan, without the injection of contrast agent), and serves as the basis for the CT equipment to perform the scan.

[0043] The scan range is set to a scanning interval that completely covers the target vertebral body, determined by the plain scan command. For example, it can be set to "include the upper and lower endplates of the target vertebral body and the area extending 5-10mm laterally," ensuring that no target vertebral body is missed and reducing the interference of excessive redundant tissue data with subsequent analysis. The scan parameters are the routine clinical parameters used by the CT equipment when performing a plain scan. These may include tube voltage (e.g., 100-120kV), tube current (e.g., 80-120mA), scan slice thickness (e.g., 1-2mm), and reconstruction nucleus (e.g., soft tissue reconstruction nucleus). The values ​​of the scan parameters follow routine clinical diagnosis and treatment guidelines, and no additional customized parameters are required, making them compatible with existing equipment in primary healthcare institutions.

[0044] In response to the scan completion command, a scan data sequence including pixel grayscale values ​​and scan parameters is generated. After the CT equipment completes the scan according to the set scan parameters, it converts the acquired raw data into a continuous sequence file in DICOM format. Each file corresponds to a slice and contains complete information such as the original pixel grayscale values ​​of that slice, the scan parameters, and patient identification.

[0045] Next, the scanned data sequences undergo quality verification, which includes sequence integrity, complete coverage of the scan range, and the degree of artifact interference. Quality verification screens newly acquired scanned data sequences for validity, eliminating invalid data caused by equipment malfunction, operational errors, or poor patient cooperation. This is a crucial preliminary step to ensure the accuracy of subsequent bone mineral density quantification. In one example, the sequence number of the DICOM file can be used to confirm that the tomographic sequence is continuous without gaps, and that the file is free of loss or format errors. Through image preview, it is manually confirmed that the scan range completely includes the upper and lower endplates of the target vertebral body, without any truncation of the target area. At the same time, metal artifacts, motion artifacts, and respiratory artifacts are screened. If the area of ​​the cancellous bone region of the target vertebral body covered by artifacts is less than a set proportion, such as 5%, it can be judged as qualified. Conversely, if the artifacts severely interfere with the target area, it can be judged as unqualified and a rescan is required.

[0046] For validated scan data sequences, they can be categorized and stored according to the correspondence between patient ID, examination date, and target vertebra, for example, Patient ID-20250101-L1-L3. Then, the target information is extracted from the header file of the scan data sequence. The header file refers to the core parameters contained in the header of the DICOM format scan data sequence that are crucial for subsequent bone mineral density calculations. These parameters may include scaling slope parameters, scaling intercept parameters, scan slice thickness, tube voltage, reconstruction algorithm, etc., which are the basis for accurate conversion of pixel grayscale values ​​to HU values. Through three-dimensional associative categorized storage, it can be ensured that each set of data accurately corresponds to a specific patient, examination time, and target vertebra, complying with medical data privacy protection and compliance requirements, while also facilitating subsequent clinical follow-up and traceability.

[0047] If the plain CT scan data sequence is a historical plain CT data sequence, the search can be performed using patient identification, examination date, and target vertebral body as keywords to filter the scan data sequence corresponding to the target vertebral body covered by the set scan range. Keyword search is used to accurately locate the target historical data. Patient identification ensures that the data belongs to a unique patient, examination date limits the time range of the data, and target vertebral body can filter sequences containing the object being examined, thus achieving fast and accurate data retrieval.

[0048] The raw parameter information is extracted from the scan data sequence, retaining patient identification and examination date. The raw parameter information consists of bone mineral density quantification-related parameters contained in historical plain CT data sequences, consistent with the target information in the header file of non-historical plain CT data sequences. In one example, a data parsing tool can extract all raw parameter information from the retrieved historical plain CT data sequences, retaining the necessary calibration and scan parameters, while filtering and retaining metadata such as patient identification and examination date, and removing redundant information irrelevant to the current bone mineral density test.

[0049] Establish a correlation between historical plain CT data sequences and current medical records, and classify and store them according to patient identification, examination date, and the corresponding target vertebral body. The correlation includes establishing a mapping relationship between historical plain CT data and current medical records through patient identification, so that historical bone mineral density data can be directly compared with the current test results, providing data support for assessing the trend of changes in the patient's bone mass.

[0050] In this embodiment, multiple parallel planes refer to three characteristic tomographic planes perpendicular to and parallel to the long axis of the target vertebral body. These can include a first plane, a second plane, and a third plane of the target vertebral body, used to comprehensively characterize the density distribution of the cancellous bone in the target vertebral body. The first plane is located at the first height of the target vertebral body, the second plane at the second height, and the third plane at the third height. The first, second, and third heights are standardized positioning coordinates defined based on the overall height of the target vertebral body, with the first height being less than the second height, and the second height being less than the third height. For example, the first height can correspond to 1 / 4 of the overall height of the target vertebral body (upper plane), the second height can correspond to 1 / 2 of the overall height of the target vertebral body (middle plane, midpoint of the vertebral body), and the third height can correspond to 3 / 4 of the overall height of the target vertebral body (lower plane). The three heights increase sequentially along the long axis of the target vertebral body, ensuring that the selection of planes is standardized and consistent.

[0051] In step 202, the target delineation method for each region of interest (ROI) in cancellous bone can be determined based on the mode selection command. The mode selection command is a command issued by the operator based on the clinical scenario, image quality, or operational needs, and can be used to specify the delineation method for the ROI, adapting to different image quality and operator proficiency requirements. The target delineation method is the specific way the ROI is drawn, which can include automatic template matching, semi-automatic assistance, and manual operation, flexibly switching according to actual needs, balancing efficiency and accuracy. The ROI of cancellous bone only includes a specific area of ​​cancellous bone tissue, which is the analysis area for subsequent HU value conversion and bone mineral density quantification.

[0052] The system offers several key features: Automatic template matching is suitable for routine cases with good image quality and regular vertebral body morphology. The system uses a pre-trained vertebral cancellous bone template to automatically match the vertebral body contour at the current image level, quickly generating a Region of Interest (ROI). Semi-automatic assistance is suitable for cases with slight variations in vertebral body morphology or minor artifacts. The system automatically generates an initial ROI contour, which the operator can then fine-tune by dragging and dropping nodes to improve accuracy. Manual operation is suitable for special cases with poor image quality, severe artifacts, or vertebral deformities. The operator manually draws the ROI contour directly on the image, ensuring that interfering areas are avoided.

[0053] Next, the region of interest is delineated according to the target delineation method, and interfering tissues or pixels are excluded during the delineation process, such as cortical bone (high-density hard bone at the edge of the target vertebral body), venous plexus (a densely vascularized area within the target vertebral body), osteophyte sclerosis foci (abnormally high-density bone areas), metal artifacts (image distortion areas caused by in vivo metal implants), and extreme outlier pixels (abnormal pixels with HU values ​​that significantly deviate from the normal range of cancellous bone).

[0054] In one example, tissue-level exclusion can be initially screened using grayscale thresholding, automatically avoiding interfering tissues such as cortical bone and venous plexuses. Abnormal area exclusion can be achieved through image feature recognition, manually or automatically marking and excluding areas corresponding to osteophyte sclerosis and metal artifacts. Pixel-level exclusion can perform HU value statistics on pixels within a defined area, removing pixels with extreme outliers.

[0055] Specifically, the region of interest (ROI) must be at least 2 pixels away from the edge of the cortical bone, and the area of ​​the ROI in each layer must be at least 80 mm². The set pixel is a standardized threshold set to minimize the safe distance from the cortical bone during ROI delineation. The set area is the minimum area required to ensure statistical representativeness of the ROI. Specific values ​​can be adaptively adjusted based on the pixel pitch of the CT equipment. For example, a distance of ≥2 pixels from the edge of the cortical bone and an ROI area of ​​≥80 mm² per layer. 2 This ensures that the number of selected pixels is statistically representative, reducing quantization deviation caused by insufficient sampling.

[0056] This application's embodiments locate three parallel planes at 1 / 4, 1 / 2, and 3 / 4 of the overall height of the target vertebral body. This reduces subjective errors from manual plane selection, ensures consistent sampling locations across different cases and operators, and improves the comparability of test results. The three delineation methods cover various clinical scenarios, including routine and special cases, ensuring both delineation efficiency for routine cases and accuracy for special cases, thus enhancing the method's clinical applicability. Interference factors are eliminated from three dimensions: tissue, region, and pixel. Combined with distance and area control standards, this ensures ROI purity and reduces interference from non-cancellous bone tissue or abnormal pixels on the calculation of the mean HU value, significantly improving the accuracy of bone mineral density quantification results.

[0057] In step 203, the header file corresponding to the scan data sequence can be read first, and the slope parameter and intercept parameter can be extracted from the header file. The header file is the file part of the scan data sequence used to store metadata. It can contain core data such as calibration parameters, equipment information, and scanning parameters during the CT scan process. It does not directly store image pixel information and is the basis for accurate conversion of raw grayscale values ​​to HU values. The slope parameter, also known as the scaling slope parameter, is a standardized parameter written into the header file after the CT equipment is calibrated at the factory. It is used to correct the linear scaling deviation of the X-ray attenuation signal, ensuring that grayscale values ​​under different equipment and different scanning parameters can be converted into a unified standard density quantization value. The intercept parameter is the scaling intercept parameter, a standardized calibration parameter used in conjunction with the slope parameter. It is used to correct the signal baseline deviation, ensuring that density quantization is performed based on the attenuation coefficient of water (HU value of water = 0).

[0058] Then, based on the slope parameter, intercept parameter, and the original grayscale value of each pixel within the region of interest (ROI), the HU value is calculated pixel-by-pixel within each ROI. The original grayscale value refers to the raw numerical value obtained by analog-to-digital conversion of the signal received by the detector after X-rays penetrate human tissue during a CT scan. It directly reflects the degree of tissue attenuation of X-rays, but it has not undergone standardized calibration and cannot be directly compared between different devices. The HU value is the standard unit for characterizing the degree of X-ray attenuation in CT images and is also a core indicator for quantifying bone tissue density; a higher value indicates greater tissue density. For each pixel within the ROI, its original grayscale value is extracted and substituted into the standardized conversion formula to calculate the HU value.

[0059] In one example, the formula for calculating the HU value for each pixel satisfies: ; in, The original grayscale value. For slope parameter, This is the intercept parameter.

[0060] After the conversion, a HU value matrix for each ROI layer can be automatically generated. The matrix dimensions perfectly match the pixel distribution of the ROI, facilitating subsequent denoising and mean calculation. Next, denoising is performed on the HU values. Denoising refers to the process of eliminating random fluctuations in HU values ​​caused by equipment noise, electronic interference, and other factors during the scanning process. The goal is to preserve the true distribution characteristics of cancellous bone density and remove isolated abnormal pixel values. Denoising can include median filtering or nonlocal mean filtering. Median filtering is a nonlinear filtering algorithm that replaces a pixel value with the median of all pixel values ​​in its neighborhood. This effectively removes salt-and-pepper noise (isolated high / low abnormal pixels) and preserves the true density characteristics of bone tissue edges, meeting the denoising requirements of cancellous bone HU values. Nonlocal mean filtering is a similarity-weighted filtering algorithm that searches for pixel blocks similar to the target pixel in the entire image and assigns them weights for averaging. This can eliminate noise while preserving image details (such as subtle density differences within cancellous bone), making it suitable for CT images with relatively uniform noise distribution.

[0061] For multiple parallel slices of the same target vertebra, the mean HU value of the region of interest in each parallel slice is calculated. The average of the mean HU values ​​of the regions of interest in multiple parallel slices is taken as the mean CT value of the target vertebra.

[0062] The mean HU value refers to the arithmetic mean of all valid HU values ​​within a single parallel ROI, reflecting the local average density of the cancellous bone at that level. The average of the HU values ​​from multiple levels is the mean CT value of the target vertebral body, reflecting the overall average density of the cancellous bone in the entire target vertebral body. This is the core foundational data for subsequent cBMD calculations, directly characterizing the overall density level of the cancellous bone in the target vertebral body. The mean HU value of each level, the mean CT value of the target vertebral body, and the corresponding ROI information and patient information are linked and stored for easy data traceability and anomaly detection.

[0063] This application's embodiments eliminate device and parameter deviations through standardization. Pixel-by-pixel conversion is performed based on the standardized slope and intercept parameters in the DICOM header file, ensuring that the original grayscale values ​​under different CT devices and scanning parameters are converted into a unified standard HU value. This lays the foundation for comparability of results across devices and centers, solving the core problem that traditional original grayscale values ​​cannot be directly compared.

[0064] The preset reference value in step 204 is the core benchmark for calculating cBMD, and its accuracy and robustness directly determine the quantitative reliability and cross-scenario comparability of cBMD. Therefore, the preset reference value in this embodiment can be determined based on a dual strategy of the normal anchoring method and the optimal consistency method. This anchors the bone mineral density benchmark level of the normal population while ensuring that the grading results based on the reference value are highly consistent with the reference standard. The normal anchoring method is a method for determining the benchmark value based on the bone mineral density distribution characteristics of the normal population. The core logic is to anchor the population that has been determined to be in the "bone mass benchmark state" by the reference standard, and use the robust distribution range of their mean CT values ​​as the basic candidate range of the reference value to ensure that the reference value closely matches the true normal bone mineral density level. The optimal consistency method is a method for optimizing the reference value based on clinical diagnostic consistency. The core logic is to screen out the reference value range with the optimal diagnostic consistency by verifying the degree of matching between the grading results corresponding to different candidate reference values ​​and the reference standard, thus ensuring the clinical suitability of the reference value.

[0065] For the normal anchor point method, firstly, CT image samples with quantitative CT labels are collected, covering different scanning equipment, different scanning parameters, and different populations, to construct a sample set. Quantitative CT labels can include volumetric bone mineral density (BMD) indicators and the corresponding grading results. CT image samples with quantitative CT labels refer to samples that simultaneously contain conventional plain CT image data and quantitative CT test result labels. The label information is clear and authoritative, providing a reference standard for the calibration and verification of reference values.

[0066] From the sample set, samples that were determined to be at the baseline bone mass status by quantitative CT were selected as the baseline sample subset. The baseline bone mass status refers to the state of normal bone mass as determined by reference standards, i.e., a vBMD value higher than the critical threshold for bone loss; it is the core sample type for anchoring normal bone mineral density levels. The baseline sample subset is the set of samples selected from the sample set that were determined to be at the baseline bone mass status by quantitative CT, used to analyze the CT mean distribution characteristics of the normal population.

[0067] For each sample in the baseline sample subset, the mean CT value of the target vertebra is determined. The mean CT value refers to the mean CT value of the target vertebra in each sample in the baseline sample subset, and the calculation logic is consistent with step 203 of this application. Next, the distribution data of the mean CT values ​​are statistically analyzed, and the distribution center trend and the robust median interval across the center are extracted. The robust median interval is used as the first candidate interval of the preset reference value. The first candidate interval is the candidate interval obtained based on the normal anchor point method. The robust median interval refers to the median value interval (such as the 25%-75% quantile interval) that covers more than 50% of the core samples in the distribution of the mean CT values ​​after removing extreme outliers. This interval is less affected by extreme values ​​and can more stably represent the concentrated distribution range of the mean CT values ​​of the normal population.

[0068] For the consistency-optimal method, the three-category result of quantitative CT is first used as the reference standard. A preset range of candidate values ​​and a step size are then set. The three-category result refers to the bone mineral density grading result of the reference standard, which can include baseline bone mass, osteopenia, and osteoporosis. For example, combining clinical experience with the range of the first candidate interval, a preset range of candidate values ​​is set (e.g., if the first candidate interval is [420, 460] HU, the candidate value range can be expanded to [410, 470] HU). A step size for the candidate values ​​is then set (e.g., 5 HU) to ensure a fine traversal of the candidate value range.

[0069] For each candidate value within the candidate value range, the relative bone mineral density (cBMD) of each sample in the sample set is calculated and classified to obtain the classification result. For example, for each candidate reference value within the candidate value range, all samples in the sample set are traversed, and the relative bone mineral density of each sample is calculated according to the formula for calculating the relative bone mineral density. Then, based on a preset classification threshold aligned with the reference standard, the cBMD of each sample is classified into three categories to obtain the classification result based on the current candidate reference value.

[0070] Next, the accuracy and Cohen's Kappa coefficient of the grading results compared to the reference standard are calculated. Cohen's Kappa is a statistical indicator that measures the consistency between two diagnostic methods (or grading standards), ranging from -1 to 1. For example, a Kappa value closer to 1 indicates stronger consistency; Kappa ≥ 0.75 indicates good consistency, and Kappa ≥ 0.85 indicates excellent consistency; it is a core indicator for assessing clinical diagnostic consistency. For each candidate reference value, the grading result is compared with the sample's reference standard grading result, and two core consistency indicators are calculated. One is the accuracy (number of correctly graded samples / total number of samples × 100%), which measures the overall grading match. The other is the Cohen's Kappa coefficient, which measures the consistency of the grading results (excluding the influence of random matching).

[0071] Plot the curves of accuracy and Coen Kappa coefficient as a function of candidate values, identify stable peak plateau intervals in the curves, and use these peak plateau intervals as the second candidate intervals for the preset reference values. The second candidate intervals are selected based on the optimal consistency method. A peak plateau interval refers to the interval in the curves of accuracy and Coen Kappa coefficient as a function of candidate reference values ​​where the indicator values ​​remain at a high level (e.g., accuracy ≥ 90%, Kappa ≥ 0.8) and fluctuate little. Reference values ​​within this interval ensure high consistency between the grading results and the reference standard. Specifically, plot the curves of the two indicators with the candidate reference values ​​on the horizontal axis and accuracy and Coen Kappa coefficient on the vertical axis respectively; identify the intervals with high stability in both areas, i.e., peak plateau intervals with accuracy ≥ 90% and Coen Kappa coefficient ≥ 0. (indicator value fluctuations ≤ 5% within this interval). This peak plateau interval is determined as the second candidate interval for the preset reference values, ensuring that reference values ​​within this interval guarantee high consistency between the grading results and the reference standard.

[0072] Next, the intersection of the first and second candidate intervals is taken as the effective candidate interval for the preset reference value. A target value is selected from the effective candidate interval as the preset reference value. The effective candidate interval refers to the intersection of the first candidate interval (determined by the normal anchor point method) and the second candidate interval (determined by the consistency optimization method), which not only conforms to the bone density distribution of the normal population but also ensures consistency in clinical diagnosis, representing a reliable range for screening the final preset reference value. The preferred approach for selecting the target value from the effective candidate interval as the final preset reference value is to choose the median of the effective candidate interval. The median combines stability and representativeness, minimizing the potential bias caused by interval boundary values. Preferably, through experiments, the preset reference value in this embodiment can be 440 HU. For example, grid scanning on the test set shows that the accuracy and Kern-Kappa accuracy plateau around 430-440 HU and approach their peak. Specifically, at the preset reference value of 430 HU, the accuracy is approximately 0.756 and Kern-Kappa approximately 0.496, and at the preset reference value of 440 HU, the accuracy is approximately 0.744 and Kern-Kappa approximately 0.460. This indicates that 440 HU is within the optimal range and the performance is stable, which facilitates cross-center standardization.

[0073] This application employs a normal anchor point method to screen bone mineral density baseline samples, using the robust median interval of their CT mean as a basis to ensure that the preset reference value accurately reflects the bone mineral density level of the normal population, avoiding quantitative bias caused by the reference value deviating from the actual clinical benchmark. The consistency optimization method is used to verify the degree of matching between the reference value and the reference standard in terms of grading, selecting the peak plateau interval as a second candidate interval to ensure that the cBMD grading results based on this reference value are highly consistent with the reference standard, thus improving the clinical diagnostic value of the method. The effective candidate interval is the intersection of the two methods, excluding both reference values ​​that closely fit the normal distribution but have poor grading consistency and reference values ​​that have high grading consistency but deviate from the normal distribution. Simultaneously, the sample set covers multiple devices, multiple parameters, and multiple populations, making the finally determined preset reference value highly robust and adaptable to conventional CT equipment in different medical institutions, achieving cross-device and cross-center result comparability.

[0074] In step 205, a first and a second set threshold can be set to classify bone mineral density (BMD) indicators. The first set threshold is a critical value used to distinguish between baseline BMD and osteopenia. Its value is based on the critical value between normal and osteopenia in the reference standard, determined by back-calculation through a linear regression model and verified by ROC analysis. The second set threshold is a critical value used to distinguish between osteopenia and osteoporosis. Its value is based on the critical value between osteopenia and osteoporosis in the reference standard, determined by back-calculation through a linear regression model and verified by ROC analysis.

[0075] If the relative bone mineral density (RMD) is greater than a first set threshold, the grading result is determined to be the baseline bone mineral density state. If the RMD is less than or equal to the first set threshold but greater than a second set threshold, the grading result is determined to be a state of decreased bone mineral density. If the RMD is less than the second set threshold, the grading result is determined to be a state of osteoporosis. Then, the mean CT value of the target vertebra, the RMD, and the grading result are correlated and integrated to obtain the assessment result for each target vertebra.

[0076] In one example, the first set threshold may be -20%, and the second set threshold may be -50%. Therefore, cBMD > -20% can be determined as the bone density reference state. -50% < cBMD ≤ -20% can be determined as the osteopenia state. cBMD ≥ -50% is determined as osteoporosis. For clinical management, it can be further divided: -70% ≤ cBMD < -50% is mild osteoporosis, -90% ≤ cBMD < -70% is moderate osteoporosis, and cBMD ≤ -90% is severe osteoporosis. To meet the needs of clinical follow-up and stratified management, within the osteoporosis range (≤ -50%), light / medium / severe stratification is set at approximately 20% gradients; this equidistant stratification matches the measurement error level, with stable grading and no change in the alignment relationship with the main demarcation of quantitative CT. It should be noted that the equivalent adjustment within the ranges of (-65% to -75%) and (-85% to -95%) is still within the protection scope of the present invention.

[0077] The grading determination of the embodiments of the present application is based on clear numerical thresholds, with clear logic and unified standards. The set thresholds are derived from the transformation and verification of reference standards, ensuring that the grading results are highly consistent with clinically recognized reference standards, and having extremely high accuracy and clinical recognition. The final evaluation results integrate quantitative indicators and qualitative conclusions. Doctors can quickly obtain the bone density level and the status of the target vertebra without complex professional interpretation, significantly improving the convenience and efficiency of clinical applications. Since cBMD eliminates the influence of equipment and parameters, and the grading thresholds are standardized, the evaluation results of the embodiments of the present application are highly comparable among different hospitals and different CT devices, and are very suitable for multi-center clinical research and osteoporosis screening of large-scale populations.

[0078] The determination of the first set threshold and the second set threshold of the embodiments of the present application can be achieved through the following methods.

[0079] First, a linear regression model was established on the clinical development set to correlate quantitative CT-detected volumetric bone mineral density (vBMD) with the HU value of the target vertebral body. The clinical development set is a labeled collection of clinical samples from multiple centers, using multiple devices, and involving multiple populations. Each sample simultaneously possesses both vBMD data from quantitative CT and HU value data from conventional plain CT scans. This data is used to build the model, back-calculate thresholds, and verify consistency, forming the data foundation for threshold determination. The linear regression model characterizes the quantitative correlation between vBMD from quantitative CT and HU values ​​from conventional CT. For each sample in the clinical development set, the HU value of the target vertebral body was extracted as the independent variable X, and the vBMD value from quantitative CT of the same target vertebral body was extracted as the dependent variable Y. (X, Y) variable pairs were established for each sample to ensure precise matching between variables. Then, the least squares method was used to perform a linear retrospective fitting of the variable pairs, establishing a linear regression model between vBMD and HU values. The reliability of the model was verified using a goodness-of-fit index, ensuring a strong linear relationship between vBMD and HU values, providing a reliable data foundation for subsequent threshold back-calculation.

[0080] Then, based on a linear regression model, the first HU value corresponding to the bone mass baseline threshold and the second HU value corresponding to the osteoporosis status threshold in the reference standards were respectively derived. The bone mass baseline threshold is the vBMD cutoff value in the reference standards that defines the bone mass baseline status and bone loss (e.g., the clinically routine standard is 120 mg / cm²). 3 The bone mass baseline (vBMD) is defined as follows: vBMD > 1 HU indicates a bone mass baseline state, while vBMD ≤ 1 HU indicates a non-bone mass baseline state. The osteoporosis threshold is the vBMD cutoff value used in reference standards to define the difference between bone loss and osteoporosis (e.g., the clinically accepted standard is 80 mg / cm²). 3 That is, vBMD≤2HU indicates osteoporosis, and vBMD>2HU indicates non-osteoporosis.

[0081] The first HU value was obtained by back-calculation using a linear regression model. It represents the HU value of a conventional CT scan corresponding to the reference standard's bone mass baseline threshold, and is the result of transforming the reference standard threshold from the vBMD domain to the HU domain. Similarly, the second HU value was also obtained by back-calculation using a linear regression model. It represents the HU value of a conventional CT scan corresponding to the reference standard's osteoporosis threshold, and is also the result of cross-domain transformation of the reference standard threshold.

[0082] Next, the first and second HU values ​​are substituted into the relative bone mineral density (cBMD) calculation formula, and normalized to transform them into the quantification domain of the relative bone mineral density index, thus determining the primary grading threshold corresponding to the grading logic of the reference standard. Normalization is the process of substituting the HU value form of the threshold into the relative bone mineral density calculation formula to convert it into a dimensionless cBMD threshold. Its core purpose is to eliminate systematic biases caused by different equipment and scanning parameters, making the threshold adaptable to cross-scenario applications. The primary grading threshold refers to the transformed cBMD candidate threshold corresponding to the grading logic of the reference standard, including the first primary grading threshold corresponding to the bone mass baseline threshold and the second primary grading threshold corresponding to the osteoporosis status threshold. The logical rationality of the two primary grading thresholds is judged based on clinical experience; the first primary grading threshold must be greater than the second primary grading threshold. If this condition is not met, the process returns to verifying the linear regression model and the reverse calculation process.

[0083] Finally, using relative bone mineral density as a continuous variable, ROC analysis was performed on a first-class classification task based on baseline and non-baseline bone mass states, and a second-class classification task based on non-osteoporotic and osteoporotic states, respectively, to verify whether the target cut-off points corresponding to the primary grading thresholds matched the judgment logic of the reference standards. The first-class classification task refers to the binary classification between baseline and non-baseline bone mass states, used to verify the discriminative effectiveness of the first primary grading threshold. The second-class classification task refers to the binary classification between non-osteoporotic and osteoporotic states, used to verify the discriminative effectiveness of the second primary grading threshold.

[0084] In one example, the cBMD value of all samples in the clinical development set can be used as a continuous variable, with the reference standard's bone mass baseline status or non-bone mass baseline status as labels, to construct a first classification task and plot an ROC curve. The target cut-off point corresponding to the first major grading threshold is located, and the diagnostic performance indicators for this cut-off point can be calculated, including accuracy, true positive rate (sensitivity in identifying bone mass baseline status), false positive rate (probability of misclassification as bone mass baseline status), and Coencapsulation coefficient. Similarly, using the cBMD value as a continuous variable, with the reference standard's non-osteoporotic status or osteoporotic status as labels, a second classification task is constructed, and an ROC curve is plotted. The target cut-off point corresponding to the second major grading threshold is located, and the corresponding accuracy, true positive rate (sensitivity in identifying osteoporotic status), false positive rate, and Coencapsulation coefficient are calculated. Consistency can be determined based on accuracy and the Coencapsulation coefficient; for example, the diagnostic performance qualification standard can be set as accuracy ≥ 90% and Coencapsulation coefficient ≥ 0.8. If the target cut-off points corresponding to both primary grading thresholds meet the qualification criteria, and the differentiation results show no systematic deviation from the reference standard (e.g., no batch misjudgments), then the primary grading thresholds are considered to match the reference standard's judgment logic. Otherwise, if they do not meet the criteria, the linear regression model parameters are adjusted or the clinical development set sample size is increased, and the above steps are repeated.

[0085] If the target cut-off point corresponding to the primary grading threshold matches the judgment logic of the reference standard, the ROC analysis validation is successful. Then, a first set threshold and a second set threshold are determined based on the primary grading threshold. For example, the first primary grading threshold is determined as the first set threshold, and the second primary grading threshold is determined as the second set threshold. The two determined set thresholds are stored in the system parameter library and can be directly called upon in subsequent clinical applications for bone mineral density grading determination in step 205.

[0086] Traditional bone mineral density (CMD) testing methods often result in unreliable results due to structural abnormalities. In order to reduce the deviation in CMD quantification caused by abnormal cancellous bone structure from the source and ensure that the final output of cBMD and grading results has real clinical reference value, before step 204 of this application, the standard deviation of the HU value of the pixels in the region of interest can be calculated and compared with the set standard deviation to determine whether there is an abnormality in density distribution.

[0087] The standard deviation of the HU value refers to the quantitative index of the dispersion of all effective HU values ​​within each parallel layer of the same target vertebral body after noise reduction in step 203, expressed in HU. A larger standard deviation indicates a more uneven density distribution of cancellous bone within the ROI, potentially indicating structural abnormalities or interfering factors. Conversely, a smaller standard deviation indicates a more uniform density distribution and more stable quantification results. The standard deviation is set as a critical threshold based on the HU value distribution characteristics of cancellous bone in a normal population. It is used to determine whether the HU value distribution is abnormal. Its value is obtained through statistical analysis of a large sample of normal bone mass baseline data and serves as the basis for distinguishing between uniform and abnormal bone density distribution.

[0088] If the detected standard deviation exceeds the set standard deviation, an abnormal bone mineral density distribution is determined, and a prompt message indicating that the bone mineral density test results need to be reviewed is output. A structural integrity check is then performed on the plain CT scan of the target vertebra. The prompt message is used to remind the operator to re-examine the bone mineral density testing process and image data to investigate the cause of any abnormalities. The structural integrity check involves morphological analysis of the plain CT images of the target vertebra to identify any abnormalities that could damage the normal structure of cancellous bone and cause distortion in bone mineral density quantification. This serves as the morphological basis for determining the applicability of the bone mineral density test results.

[0089] If any of the following conditions are detected in the plain CT image: compression fracture, postoperative internal fixation, or tumor erosion leading to destruction of cancellous bone structure, the bone mineral density test result of the target vertebra is marked as inapplicable, and the reason for inapplicability is explained.

[0090] Compression fractures refer to vertebral body lesions caused by external force or osteoporosis, resulting in decreased vertebral height and fragmentation of cancellous bone. This can cause a sharp increase in local bone mineral density (HU) values ​​(due to trabecular compression and density) or extremely uneven distribution, severely affecting the accuracy of bone mineral density quantification. Postoperative internal fixation refers to the implantation of metal fixation devices (such as screws, plates, and fusion cages) after vertebral body surgery (such as fracture reduction and fixation, intervertebral disc replacement, etc.). These devices produce strong metal artifacts on CT images, interfering with the detection of HU values ​​in the surrounding cancellous bone. Simultaneously, the bone structure in the internal fixation area is already destroyed, making it impossible to reflect the true bone mineral density level. Tumor erosion refers to tumor tissue invading the vertebral cancellous bone, leading to trabecular destruction, dissolution, or abnormal proliferation, resulting in irregular and abnormal local HU value distribution (such as areas of extremely low or extremely high density), making it impossible to reflect the true bone mineral density using conventional quantitative methods.

[0091] For example, a vertebral body height that is ≥20% lower than that of an adjacent normal vertebral body, or a wedge-shaped or flattened vertebral body with a high-density compression band visible in the cancellous bone area, is considered a compression fracture. The presence of metallic fixation in the vertebral body area, accompanied by metallic artifacts (such as radial high-density stripes), or bone defects corresponding to the surgical incision, indicates postoperative internal fixation. Irregular low-density destructive areas, osteophyte formations, or sclerotic foci within the vertebral cancellous bone, or the presence of a paravertebral soft tissue mass, indicate tumor erosion.

[0092] If any structural abnormality is found, the system marks the bone mineral density (BMD) test result of the target vertebra as "not applicable" and automatically generates a standardized explanation for inapplicability (e.g., "Compression fracture exists in the target vertebra, cancellous bone structure is destroyed, and the BMD quantification result is unreliable," or "Postoperative internal fixation is visible in the target vertebra, and metal artifacts interfere with HU value detection"). Marking the result as "not applicable" means that when the aforementioned cancellous bone structure destruction is confirmed, the BMD quantification result of the target vertebra is deemed to have no clinical reference value, and subsequent relative BMD calculations will not be performed. The reason for inapplicability is also noted to reduce erroneous result output.

[0093] If no structural abnormalities are found, the operator needs to further verify the previous steps (such as whether the ROI delineation includes interfering tissue, whether the denoising parameters are reasonable, and whether there are artifacts in the scan), correct the problems, and recalculate the standard deviation of the HU value. If the problem still cannot be resolved after verification, it can be manually marked as "further evaluation based on clinical history is required," and the doctor can be prompted to determine whether to continue the test based on the patient's clinical information.

[0094] This application's embodiments ensure the reliability of quantitative results by pre-emptively eliminating abnormal interference. By verifying the standard deviation of the HU value, cases with abnormal cancellous bone mineral density distribution are identified in advance. Combined with structural integrity checks to pinpoint the root cause, this reduces deviations in bone mineral density quantification caused by structural damage such as compression fractures, internal fixation, and tumor erosion, eliminating unreliable test results at the source and improving the safety of clinical diagnosis. Automatic verification, early warning information generation, and standardized labeling of inapplicable reasons eliminate the need for manual investigation by operators, reducing workload. Simultaneously, standardized explanations of the reasons facilitate doctors' quick understanding of the limitations of the test results, reducing the over-interpretation of erroneous data.

[0095] The following example uses actual data obtained from the bone mineral density assessment method based on the embodiments of this application.

[0096] 1. Case-level consistency example.

[0097] Female, 69 years old, with a history of thyroid tumor. QCT: L1–L3; vBMD = 55 / 51 / 49 mg / cm² 3 (Osteoporosis). The three-slice ROIs obtained from plain CT scans have the following HU (upper / middle / lower): L1: 58 / 78 / 72; L2: 66 / 68 / 64; L3: 64 / 60 / 70. CT_mean ≈ 69.33 / 66.00 / 64.67 HU.

[0098] Substituting the values ​​into the formula cBMD=(CT_mean-440) / 440×100%, we can obtain cBMD≈-84.24% / -85.00% / -85.26% for the target vertebral bodies L1–L3, all of which are "osteoporosis", consistent with the classification of quantitative CT.

[0099] 2. Multicenter statistics.

[0100] Total number of cases N = 600, divided into development / testing in a 7:3 ratio. The following metrics are reported on the plain scan test set (n = 164): Tri-class classification accuracy ≈ 0.744, Cohen's κ ≈ 0.460; cBMD AUC for "QCT ≥ 120" ≈ 0.903, and for "QCT ≥ 80" AUC ≈ 0.880. Bland–Altman mean difference (when vBMD_pred is obtained by regression) ≈ -1.13 mg / cm². 3 The 95% consensus threshold is approximately [-20.74, +18.48] mg / cm³. 3 Regarding repeatability, the average CV% of the simulated two-evaluator cBMD was approximately 7.85%, and the ICC(2,1) was approximately 0.995.

[0101] Subgroup stability: The three-class classification accuracy and Cohen's κ remain within acceptable range for different brands, kV, and reconstruction kernels. For example: Philips (n=62) accuracy ≈ 0.774, κ ≈ 0.522; GE (n=53) accuracy ≈ 0.755, κ ≈ 0.506; Siemens (n=49) accuracy ≈ 0.694, κ ≈ 0.306; reconstruction kernel standard group (n=93) accuracy ≈ 0.806, κ ≈ 0.590.

[0102] This application's embodiments, relying solely on plain CT scans without using phantoms or dedicated quantitative CT software, construct cross-device comparable cBMD indices through "fixed reference points + percentage normalization," and achieve directly usable stratified management with a unified grading system aligned with clinical thresholds for quantitative CT. The process is simple, with comprehensive quality control and documentation, facilitating rapid departmental adoption and opportunistic reuse of historical plain CT images.

[0103] Figure 3 This is a schematic diagram of a CT image-based bone density detection device provided in an embodiment of this application. The CT image-based bone density detection device 300 may include an acquisition module 301, a delineation module 302, a transformation module 303, a determination module 304, and an evaluation module 305.

[0104] The acquisition module 301 is used to acquire the scan data sequence of plain CT covering the target vertebral body.

[0105] The delineation module 302 is used to delineate the region of interest of cancellous bone in multiple parallel layers of the same target vertebra, with the parallel layers perpendicular to the long axis of the target vertebra.

[0106] The transformation module 303 is used to perform linear transformation according to the scan data sequence, calculate the HU value of each pixel in the region of interest pixel by pixel, and obtain the CT mean value of the target vertebra based on the HU value.

[0107] The determination module 304 is used to determine the relative bone mineral density index of the target vertebra based on the mean CT value, with a preset reference value as the benchmark. The preset reference value is a reference value for measuring the deviation of the bone mineral density index from the benchmark level.

[0108] The assessment module 305 is used to perform bone density grading based on relative bone mineral density index and according to a set threshold to obtain the assessment result for each target vertebra. The set threshold is a grading critical value determined based on a reference standard. The assessment result includes the mean CT value of the target vertebra, relative bone mineral density index, and grading result.

[0109] The relative bone mineral density index satisfies the following formula: ; in, This is a relative bone mineral density index. The mean CT value is... This is a preset reference value.

[0110] In this embodiment of the application, the acquisition module 301 may include a configuration unit, a generation unit, a verification unit, a first extraction unit, a filtering unit, a second extraction unit, and an association unit.

[0111] The configuration unit is used to determine the set scanning range of the target vertebra and configure the set scanning parameters based on the plain scan examination instructions if the scan data sequence of the plain CT scan is not a historical plain CT scan data sequence.

[0112] The generation unit is used to generate a scan data sequence including pixel grayscale values ​​and scan parameters in response to a scan completion command.

[0113] The verification unit is used to perform quality verification on the scanned data sequence. The quality verification includes sequence integrity, complete coverage of the scan range, and the degree of artifact impact.

[0114] The first extraction unit is used to classify and store the qualified scan data sequences according to the patient identifier, examination date and the correspondence of the target vertebra, and to extract the target information from the header file of the scan data sequence.

[0115] The filtering unit is used to search for the target vertebrae covered by the set scanning range if the scan data sequence of the plain CT scan is a historical plain CT scan data sequence, using the patient identifier, examination date and target vertebrae as keywords.

[0116] The second extraction unit is used to extract the original parameter information from the scan data sequence and retain the patient identification and examination date.

[0117] The association unit is used to establish the association between historical plain CT data sequences and current medical data, and to classify and store them according to the correspondence between patient identifier, examination date and target vertebra.

[0118] In the embodiments of this application, the multiple parallel layers may include a first layer, a second layer, and a third layer of the target vertebral body. The first layer is located at a first height of the target vertebral body, the second layer is located at a second height of the target vertebral body, and the third layer is located at a third height of the target vertebral body. The first height is less than the second height, and the second height is less than the third height.

[0119] The selection module 302 may include a first selection unit and an exclusion unit.

[0120] The first selection unit is used to determine the target delineation method for each region of interest in cancellous bone based on the mode selection instruction. The target delineation methods include automatic template matching, semi-automatic assistance, and manual operation.

[0121] The exclusion unit is used to delineate the region of interest according to the target delineation method, and excludes cortical bone, venous plexus, osteosclerotic foci, metal artifacts and extreme outlier pixels during the delineation process.

[0122] Among them, the region of interest is greater than or equal to a set number of pixels from the edge of the cortical bone, and the area of ​​the region of interest in each layer is greater than or equal to a set area.

[0123] In this embodiment, the transformation module 303 may include a reading unit, a conversion unit, a noise reduction unit, a first calculation unit, and a mean value unit.

[0124] The reading unit is used to read the header file corresponding to the scanned data sequence and extract the slope and intercept parameters from the header file.

[0125] The conversion unit is used to perform pixel-by-pixel conversion within each region of interest based on the slope parameter, intercept parameter, and the original grayscale value of each pixel within the region of interest, to obtain the HU value corresponding to each pixel.

[0126] The denoising unit is used to denoise the HU value. The denoising process includes denoising using median filtering or nonlocal mean filtering.

[0127] The first calculation unit is used to calculate the mean HU value of the region of interest for each of the multiple parallel layers of the same target vertebra.

[0128] The mean value unit is used to average the HU mean values ​​of multiple parallel layers of the region of interest as the CT mean value of the target vertebra.

[0129] The formula for calculating the HU value for each pixel satisfies: ; in, The original grayscale value. For slope parameter, This is the intercept parameter.

[0130] In this embodiment, the CT image-based bone density detection device 300 may further include a first preset module. This first preset module may include a construction unit, a second selection unit, a second calculation unit, a first recognition unit, a setting unit, a third calculation unit, a fourth calculation unit, a second recognition unit, a third recognition unit, and a third selection unit.

[0131] The construction unit is used to collect CT image samples with quantitative CT tags covering different scanning devices, different scanning parameters, and different populations, and to construct a sample set. The quantitative CT tags include volumetric bone mineral density index and the corresponding grading results of bone mineral density.

[0132] The second selection unit is used to select samples from the sample set that are determined to be in the baseline state of bone mass by quantitative CT as the baseline sample subset.

[0133] The second calculation unit is used to determine the mean CT value of the target vertebra for each sample in the baseline sample subset.

[0134] The first identification unit is used to statistically analyze the distribution data of the mean CT of the sample, extract the distribution center trend and the robust median interval across the center, and use the robust median interval as the first candidate interval of the preset reference value.

[0135] The setting unit is used to set the candidate value range and step size of the preset reference value based on the three-class results of quantitative CT. The three-class results include bone mass at the baseline state, bone loss, and osteoporosis.

[0136] The third calculation unit is used to calculate the relative bone density index of each sample in the sample set for the candidate values ​​within the candidate value range and to classify them to obtain the classification results.

[0137] The fourth calculation unit is used to calculate the accuracy of the grading results against the reference standard and the Cohn-Kappa coefficient.

[0138] The second identification unit is used to plot the curves of accuracy and Cohen-Kappa coefficient as a function of candidate values, identify the peak plateau intervals with stable performance in the curves, and use the peak plateau intervals as the second candidate intervals of the preset reference values.

[0139] The third identification unit is used to take the intersection of the first candidate interval and the second candidate interval as the valid candidate interval of the preset reference value.

[0140] The third selection unit is used to select a target value from the valid candidate interval as a preset reference value.

[0141] In this embodiment of the application, the evaluation unit 305 may include a first determination unit, a second determination unit, a third determination unit, and an integration unit.

[0142] The first determination unit is used to determine the grading result as the baseline state of bone mineral density if the relative bone mineral density index is greater than the first set threshold.

[0143] The second determination unit is used to determine the grading result as a state of decreased bone mineral density if the relative bone mineral density index is less than or equal to the first set threshold and greater than the second set threshold.

[0144] The third determination unit is used to determine the osteoporosis status if the relative bone mineral density index is less than or equal to the second set threshold.

[0145] The integration unit is used to correlate and integrate the mean CT value, relative bone mineral density index and grading results of the target vertebral body to obtain the evaluation result corresponding to each target vertebral body.

[0146] In this embodiment, the CT image-based bone density detection device 300 may further include a second preset module. This second preset module may include an establishment unit, a reverse calculation unit, a conversion unit, an analysis unit, and a threshold determination unit.

[0147] The unit is used to establish a linear regression model of the volumetric bone mineral density index detected by quantitative CT and the HU value of the target vertebral body on the clinical development set.

[0148] The reverse calculation unit is used to reverse calculate the first HU value corresponding to the bone mass baseline state threshold and the second HU value corresponding to the osteoporosis state threshold in the reference standard based on the linear regression model.

[0149] The conversion unit is used to substitute the first HU value and the second HU value into the calculation formula of relative bone mineral density, and convert them to the quantification domain of relative bone mineral density index through normalization processing, and determine the main grading threshold corresponding to the grading logic of the reference standard.

[0150] The analysis unit is used to perform ROC analysis on the first classification task formed by the relative bone mineral density index as a continuous variable, and on the second classification task formed by the non-osteoporotic state and the osteoporotic state, respectively, to verify whether the target cut-off point corresponding to the main grading threshold matches the judgment logic of the reference standard.

[0151] The threshold determination unit is used to determine the first set threshold and the second set threshold based on the main hierarchical threshold if the target tangent point corresponding to the main hierarchical threshold matches the judgment logic of the reference standard.

[0152] The CT image-based bone density detection device 300 may also include a verification module. This verification module may include a fifth calculation unit, a first prompting unit, and a second prompting unit.

[0153] The fifth calculation unit is used to calculate the standard deviation of the HU values ​​of pixels in the region of interest and compare the standard deviation with the set standard deviation.

[0154] The first prompting unit is used to output a prompt message indicating that the bone density test results need to be reviewed if the detected standard deviation is greater than the set standard deviation, and to perform a structural integrity check on the plain CT scan of the target vertebra.

[0155] The second prompting unit is used to mark the bone mineral density test result of the target vertebra as inapplicable if any of the following conditions are detected in the plain CT image: compression fracture, postoperative internal fixation, or tumor erosion leading to destruction of cancellous bone structure, and to explain the reason for inapplicability.

[0156] This application also provides a computer-readable storage medium storing a program that can be loaded by a processor and executed by any of the CT image-based bone density detection methods in this application.

[0157] The clinical significance of the embodiments in this application includes the following aspects.

[0158] This application embodiment, without adding additional scans, relying on calibration phantoms and dedicated quantitative software, fully exploits the objective grayscale information of vertebral cancellous bone in plain CT images, and constructs a system based on fixed reference values ​​( Based on the relative bone mineral density index (cBMD) and its unified grading threshold system, this application enables standardized assessment and hierarchical management of bone mass status. Compared with traditional quantitative bone mineral density assessments that rely on QCT (Quick Test-Based Computation) procedures, the embodiments of this application significantly reduce the implementation threshold and overall cost, enabling large-scale application of bone mineral density assessment capabilities in a wider range of medical institutions and clinical pathways, especially suitable for primary care and resource-constrained scenarios.

[0159] Furthermore, this embodiment of the application normalizes the average HU value of the target vertebral body and outputs reproducible grading results, which improves the consistency and comparability of bone mineral density assessment under different equipment models, scanning parameters and reconstruction conditions. This is beneficial for cross-center promotion, long-term follow-up monitoring and efficacy evaluation, and reduces result deviation and duplicate examinations caused by equipment differences.

[0160] Based on the above characteristics, the embodiments of this application can transform a large amount of conventional CT image resources into quantitative evidence that can be used for opportunistic screening and risk stratification of osteoporosis, providing a more economical, accessible and scalable technical path for fracture risk assessment, treatment decision-making and follow-up management, and has significant clinical application value and industrial transformation prospects.

[0161] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0162] The above examples illustrate this application only to aid understanding and are not intended to limit its scope. Those skilled in the art to which this application pertains can make various simple deductions, modifications, or substitutions based on the ideas presented.

Claims

1. A method for detecting bone mineral density based on CT images, characterized in that, include: Acquire the scan data sequence of plain CT covering the target vertebral body; Regions of interest in cancellous bone are delineated in multiple parallel planes of the same target vertebral body, wherein the parallel planes are multiple axial planes of the target vertebral body and are matched with the direction of the scanning plane; A linear transformation is performed on the scan data sequence to calculate the HU value of each pixel in the region of interest pixel by pixel, and the average CT value of the target vertebra is obtained based on the HU value. Using a preset reference value as a benchmark, the relative bone mineral density index of the target vertebra is determined based on the mean CT value. The preset reference value is a reference value for measuring the deviation of the bone mineral density index from the benchmark level. Based on the relative bone mineral density index, bone mineral density is graded according to a set threshold to obtain the evaluation result for each target vertebra. The set threshold is a grading threshold determined based on the reference standard. The evaluation result includes the mean CT value of the target vertebra, the relative bone mineral density index, and the grading result. The relative bone mineral density index satisfies the following formula: ； in, The relative bone mineral density index, The mean CT value is... This is the preset reference value.

2. The bone density detection method based on CT images according to claim 1, characterized in that, The acquisition of the plain CT scan data sequence covering the target vertebral body includes: If the scan data sequence of the plain CT scan is a non-historical plain CT scan data sequence, the set scan range of the target vertebra is determined based on the plain scan examination command and the set scan parameters are configured. In response to the scan completion command, a scan data sequence including pixel grayscale values ​​and scan parameters is generated; The scanned data sequence is subjected to quality verification, which includes sequence integrity, complete coverage of the scan range, and the degree of artifact impact. The scan data sequences that pass the verification are classified and stored according to the patient identifier, examination date and the correspondence of the target vertebra, and the target information is extracted from the header file of the scan data sequence. If the scan data sequence of the plain CT scan is a historical plain CT scan data sequence, the scan data sequence corresponding to the target vertebra covered by the set scan range is filtered by using the patient identifier, the examination date and the target vertebra as keywords. Extract the original parameter information from the scan data sequence, and retain the patient identifier and the examination date; Establish the association between the historical plain CT data sequence and the current medical data, and classify and store them according to the correspondence between the patient identifier, the examination date, and the target vertebra.

3. The bone density detection method based on CT images according to claim 1, characterized in that, The plurality of parallel layers include a first layer, a second layer and a third layer of the target vertebral body, wherein the first layer is at a first height of the target vertebral body, the second layer is at a second height of the target vertebral body, and the third layer is at a third height of the target vertebral body, wherein the first height is less than the second height and the second height is less than the third height; The process of delineating the region of interest of cancellous bone at multiple parallel levels within the same target vertebral body includes: The target delineation method for each region of interest in the cancellous bone is determined based on the mode selection instruction. The target delineation method includes automatic template matching, semi-automatic assistance, and manual operation. The region of interest is delineated according to the target delineation method, and cortical bone, venous plexus, osteosclerotic foci, metal artifacts and extreme outlier pixels are excluded during the delineation process. The region of interest is located at a distance greater than or equal to a set number of pixels from the edge of the cortical bone, and the area of ​​each region of interest is greater than or equal to a set area.

4. The bone density detection method based on CT images according to claim 1, characterized in that, The step of performing a linear transformation according to the scan data sequence, calculating the HU value of each pixel in the region of interest pixel by pixel, and obtaining the average CT value of the target vertebra based on the HU value includes: Read the header file corresponding to the scanned data sequence, and extract the slope parameter and intercept parameter from the header file; Based on the slope parameter, the intercept parameter, and the original grayscale value of each pixel in the region of interest, the HU value corresponding to each pixel is calculated pixel by pixel in each region of interest. The HU value is denoised, and the denoising process includes denoising using median filtering or nonlocal mean filtering. For multiple parallel layers of the same target vertebra, the mean HU value of the region of interest for each parallel layer is calculated. The average value of the HU mean of the region of interest in multiple parallel layers is taken as the average CT value of the target vertebra. The formula for calculating the HU value corresponding to each pixel satisfies: ； in, The original grayscale value, The slope parameter is... The intercept parameter is denoted as .

5. The bone density detection method based on CT images according to claim 1, characterized in that, Also includes: Collect CT image samples with quantitative CT tags covering different scanning devices, different scanning parameters and different populations to construct a sample set. The quantitative CT tags include volumetric bone mineral density index and the corresponding grading results of bone mineral density. From the sample set, samples that are determined to be in the baseline state of bone mass by quantitative CT are taken as the baseline sample subset; For each sample in the aforementioned subset of benchmark samples, determine the mean CT value of the target vertebral body. The distribution data of the sample CT mean is statistically analyzed, and the robust median intervals with distribution center tendency and cross-center are extracted. The robust median intervals are used as the first candidate intervals of the preset reference values. Using the three-class classification results of quantitative CT as a reference standard, the candidate value range and step size of the preset reference value are set. The three-class classification results include bone mass at the baseline state, bone loss, and osteoporosis. For each candidate value within the range of candidate values, the relative bone mineral density index of each sample in the sample set is calculated and graded to obtain the grading result; Calculate the accuracy and Cohen-Kappa coefficient of the grading results compared to the reference standard; Plot the curves of the accuracy and the Coen Kappa coefficient as a function of the candidate values, identify the peak plateau intervals with stable performance in the curves, and use the peak plateau intervals as the second candidate intervals for the preset reference values; The intersection of the first candidate interval and the second candidate interval is taken as the valid candidate interval of the preset reference value; Select a target value from the valid candidate interval as the preset reference value; The preset reference value is 440HU.

6. The bone density detection method based on CT images according to claim 1, characterized in that, The step of performing bone mineral density grading based on the relative bone mineral density index according to a set threshold to obtain the evaluation result corresponding to each target vertebra includes: If the relative bone mineral density index is greater than the first set threshold, the grading result is determined to be the baseline state of bone mineral density. If the relative bone mineral density index is less than or equal to the first set threshold and greater than the second set threshold, then the grading result is determined to be a state of reduced bone mineral density. If the relative bone mineral density index is less than or equal to the second set threshold, the grading result is determined to be osteoporosis. The mean CT value, relative bone mineral density index, and grading results of the target vertebra are correlated and integrated to obtain the evaluation result corresponding to each target vertebra.

7. The bone density detection method based on CT images according to claim 6, characterized in that, Also includes: In the clinical development set, a linear regression model was established between the volumetric bone mineral density index detected by the quantitative CT and the HU value of the target vertebral body; Based on the linear regression model, the first HU value corresponding to the bone mass benchmark state threshold and the second HU value corresponding to the osteoporosis state threshold in the reference standard are respectively calculated. Substitute the first HU value and the second HU value into the calculation formula of the relative bone mineral density, and transform them into the quantization domain of the relative bone mineral density index through normalization processing to determine the main grading threshold corresponding to the grading logic of the reference standard. Using the relative bone mineral density index as a continuous variable, ROC analysis was performed on the first classification task formed by the bone mass baseline state and the non-bone mass baseline state, and the second classification task formed by the non-osteoporosis state and the osteoporosis state, respectively, to verify whether the target cut point corresponding to the main grading threshold matches the judgment logic of the reference standard. If the target cut point corresponding to the main classification threshold matches the judgment logic of the reference standard, then the first set threshold and the second set threshold are determined based on the main classification threshold. The first set threshold is -20%, and the second set threshold is -50%.

8. The bone density detection method based on CT images according to claim 1, characterized in that, Before the step of determining the relative bone mineral density index of the target vertebra based on the mean CT value and using a preset reference value as a benchmark, the method further includes: Calculate the standard deviation of the HU values ​​of the pixels in the region of interest, and compare the standard deviation with a set standard deviation; If the standard deviation is detected to be greater than the set standard deviation, a prompt message is output indicating that the bone density test result needs to be reviewed, and the structural integrity of the target vertebra is checked by plain CT scan. If any of the following conditions are detected in the plain CT image: compression fracture, postoperative internal fixation, or tumor erosion leading to destruction of cancellous bone structure, the bone density test result of the target vertebra is marked as inapplicable, and the reason for inapplicability is explained.

9. A bone density detection device based on CT images, characterized in that, include: The acquisition module is used to acquire the scan data sequence of plain CT scans covering the target vertebral body; The delineation module is used to delineate regions of interest in cancellous bone on multiple parallel planes of the same target vertebra, wherein the parallel planes are perpendicular to the long axis of the target vertebra. The transformation module is used to perform linear transformation according to the scan data sequence, calculate the HU value of each pixel in the region of interest pixel by pixel, and obtain the CT mean value of the target vertebra based on the HU value; The determination module is used to determine the relative bone mineral density index of the target vertebra based on the mean CT value, with a preset reference value as a benchmark. The preset reference value is a reference value for measuring the deviation of the bone mineral density index from the benchmark level. An evaluation module is used to perform bone mineral density grading based on the relative bone mineral density index and according to a set threshold to obtain an evaluation result for each target vertebra. The set threshold is a grading threshold determined based on the reference standard. The evaluation result includes the mean CT value of the target vertebra, the relative bone mineral density index, and the grading result. The relative bone mineral density index satisfies the following formula: ； in, The relative bone mineral density index, The mean CT value is... This is the preset reference value.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that can be loaded by a processor and executed as described in any one of claims 1 to 8, the method for detecting bone density based on CT images.

Images