Method for measuring mineral density of human tissue

Through the deep learning network model, the corresponding relationship between CT value and density value is learned, and the problems of manual intervention, assumptions and equipment differences in CT value measurement are solved, and efficient and accurate tissue density calculation is achieved. It is suitable for a variety of CT devices and provides More comprehensive diagnostic information.

CN119067935AInactive Publication Date: 2024-12-03SHANGHAI JIYING TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411129306.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing CT value measurement methods have manual intervention, simple correspondence assumptions, single-type density calculations, phantom dependence and equipment differences, resulting in inaccurate density measurement and complex operation.

Method used

Using deep learning network model, the organization density data set is constructed based on the CT image data set, and the corresponding relationship between the organization CT value and the density value is learned through the 3D UNet network and the designed loss function, automatically segment the organization area and calculate the density value, which is suitable for Various CT devices.

Benefits of technology

It improves the degree of automation and accuracy of calculations, has a wide range of application, reduces artificial errors, improves computing efficiency and equipment versatility, and provides more comprehensive organizational density information to assist in diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067935A_ABST
    Figure CN119067935A_ABST
Patent Text Reader

Abstract

The invention discloses a human tissue mineral density measurement method, which comprises the following steps: collecting human tissue mineral density data, processing the human tissue mineral density data, and constructing to obtain a human tissue mineral density data set; constructing a deep learning network model, and learning based on the deep learning network model to obtain a corresponding relationship between the CT value and the density value of the human tissue; and segmenting each tissue region according to the corresponding relationship, calculating a bone density value according to the tissue density value of each tissue region, and outputting a visual measurement result. Based on the deep learning technology, the corresponding relation between the CT value and the density value of the human tissue can be automatically learned, manual intervention is not needed, the automation degree of calculation is remarkably improved, personal errors are reduced, therefore, a more accurate calculation result can be provided, the application range is wide, the calculation efficiency is high, and the equipment universality is high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of medical image processing, and in particular relates to a method for measuring the mineral density of human tissue. Background Art

[0002] Computed tomography (CT) is a key imaging technology widely used in clinical diagnosis and research. The CT value, also known as the Hounsfield Unit (HU), is a unit of measurement used to quantify the density of human tissue or organs. The determination of the CT value is based on the attenuation of X-rays after passing through tissues. Since different tissues absorb different amounts of X-rays, the CT value can be used to distinguish and quantify the density of different tissues. However, the CT value is not only affected by factors such as the X-ray tube voltage, CT device characteristics, temperature, respiration, blood flow, and patient size, but also has many complexities, which makes it challenging to directly obtain accurate tissue density values ​​from CT images. Existing solutions mainly segment and threshold the CT image, and then estimate its density value based on the grayscale value of the tissue. This method of calculating the density value of human tissue has the following problems:

[0003] 1. Human intervention: Existing methods usually require human intervention, such as tissue segmentation and threshold setting, which is not only time-consuming and labor-intensive, but also may introduce human errors.

[0004] 2. Simple correspondence assumption: Existing methods usually assume that there is a simple correspondence between the grayscale value and density value of the tissue, but this assumption is not always true in practical applications. The relationship between CT value and tissue density is affected by many factors and cannot be simply expressed by a linear relationship, resulting in errors in the calculation results.

[0005] 3. Single type density calculation: Existing methods can usually only calculate the density value of a single type of tissue, such as bone density, but cannot accurately calculate the density values ​​of other types of tissues, such as muscle density.

[0006] 4. Phantom dependence: Existing technologies usually use a phantom with known mineral density values ​​to scan the human body, and regress the relationship between CT value and mineral density through the phantom to calculate the density of human tissue. This method increases the complexity of the operation and the time cost.

[0007] 5. Equipment differences: There may be differences between different CT scanners, resulting in different CT values ​​for the same tissue on different devices, affecting the consistency and comparability of the measurement. Summary of the invention

[0008] In order to solve the above technical problems, the present invention provides a method for measuring the mineral density of human tissue, comprising:

[0009] Collecting human tissue mineral density data, and processing the human tissue mineral density data to construct a human tissue mineral density data set;

[0010] Constructing a deep learning network model, and learning and obtaining the corresponding relationship between the CT value and the density value of human tissue based on the deep learning network model;

[0011] Each tissue region is segmented according to the corresponding relationship, the bone density value is calculated by the tissue density value of each tissue region, and a visual measurement result is output.

[0012] Preferably, the process of collecting human tissue mineral density data, processing the human tissue mineral density data, and constructing a human tissue mineral density data set includes:

[0013] Perform CT scans on humans and phantoms of different ages, genders, and health conditions simultaneously to collect human CT images;

[0014] Performing standardization and denoising on the human body CT image, and then marking different tissue types in the human body CT image;

[0015] Obtaining a linear relationship between HU value and mineral density based on the mineral density data regression of the phantom, and converting the human body CT image into a density map using the linear relationship between HU value and mineral density;

[0016] A human tissue mineral density dataset of paired human CT images and density maps is constructed based on the human CT images and density maps;

[0017] Wherein, the standardization process includes adjusting the resolution and adjusting the HU value range;

[0018] Such tissue types include bone, muscle, fat, intervertebral disc and ligaments.

[0019] Preferably, the deep learning network model is constructed, and the process of learning and obtaining the corresponding relationship between the CT value and the density value of human tissue based on the deep learning network model includes:

[0020] Improve the network architecture of 3D UNet to obtain an improved 3D UNet network;

[0021] Designing a loss function, and constructing an initial deep learning network model based on the improved 3D UNet network and the designed loss function;

[0022] The initial deep learning network model is trained by using human CT image data to obtain the deep learning network model;

[0023] The human body CT image is input into the deep learning network model, a density map is obtained as output, and the corresponding relationship between the human body tissue CT value and the density value is obtained.

[0024] Preferably, the process of improving the network architecture of 3D UNet includes:

[0025] Based on the network architecture of 3D UNet, the output layer uses a 1x1x1 convolution layer to map the feature map to the density value, and outputs a density map with the same spatial resolution as the input;

[0026] For the encoder part, we use 3D ResNet blocks instead of ordinary convolution blocks and add Squeeze-and-Excitation blocks after each downsampling step.

[0027] For the decoder part, transposed convolution is used for upsampling instead of simple interpolation to make the upsampling process learnable, and dense connection blocks are added after each upsampling step to enhance feature reuse;

[0028] For skip connections, an attention gating mechanism is used to optimize the skip connections so that the network pays more attention to relevant areas.

[0029] Preferably, the process of designing the loss function includes:

[0030] The model is fully optimized by combining the loss functions, where:

[0031] Mean Square Error:

[0032] L_mse=1 / N*Σ(Y_true-Y_pred)^2

[0033] Where N is the total number of voxels, Y_true and Y_pred represent the true density distribution map and the predicted density distribution map respectively;

[0034] Structural Similarity Loss:

[0035] L_ssim=1-SSIM_3D(Y_true,Y_pred)

[0036] Among them, SSIM_3D is the similarity between the calculated predicted density distribution map (Y_pred) and the true density distribution map (Y_true);

[0037] Gradient Difference Loss:

[0038]

[0039] in, represents a 3D gradient operator;

[0040] Total loss:

[0041] L_total=α*L_mse+β*L_ssim+γ*L_grad

[0042] Among them; α, β, γ represent the weights for balancing different loss terms.

[0043] Preferably, the initial deep learning network model is trained by human CT image data, and the process of obtaining the deep learning network model includes:

[0044] The initial deep learning network model is trained in stages, followed by learning rate scheduling and data augmentation;

[0045] The process of staged training includes:

[0046] Phase 1: Preliminary training using only MSE loss;

[0047] The second stage: introduce SSIM and gradient loss to fine-tune the network;

[0048] The learning rate scheduling uses a cosine annealing learning rate with a warm-up period; the formula expression is:

[0049] lr=lr_max*0.5*(1+cos(π*t / T))

[0050] Where t is the current step number and T is the total step number;

[0051] The data enhancement includes random rotation, random scaling, random noise addition, and elastic deformation.

[0052] Preferably, the process of segmenting each tissue region according to the corresponding relationship includes:

[0053] Input the original CT image, segment each tissue area in the CT image through the deep learning network model, and output the tissue segmentation mask;

[0054] Morphological operations are then applied to refine the segmentation results, remove small isolated regions, and identify major structures using connected component analysis.

[0055] Preferably, the process of calculating the bone density value through the tissue density value of each tissue area includes:

[0056] Extract tissue density values ​​for each tissue region, and calculate bone density values ​​based on the extracted tissue density values;

[0057] The process of extracting tissue density values ​​for each tissue region includes:

[0058] The density map is combined with the tissue segmentation mask to extract the density values ​​of the main bone structures and calculate the statistical characteristics of the bone structures;

[0059] The process of calculating the bone density value according to the extracted tissue density value comprises:

[0060] Bone density was calculated for major skeletal structures as well as T-scores and Z-scores based on age-, sex-, and race-specific reference databases.

[0061] Preferably, the statistical characteristics of the bone structure include average density, median density, standard deviation of density, and volume-weighted average density;

[0062] The formula for calculating bone density for the main bone structures is:

[0063] BMD=Σ(Vi*Di) / ΣVi

[0064] Among them, Vi is the volume of the i-th voxel, and Di is its density value;

[0065] Based on age-, sex-, and race-specific reference databases, the formulas for calculating T scores and Z scores are:

[0066] T score = (measured BMD-mean BMD of young reference population) / standard deviation of BMD of young reference population;

[0067] Z score = (measured BMD - average BMD of the reference population of the same age) / standard deviation of BMD of the reference population of the same age.

[0068] Preferably, the process of outputting the visual measurement results includes:

[0069] Generate a bone density report, where the bone density report includes: BMD values ​​of each major bone part,

[0070] T-scores and Z-scores, as well as color coding to visually display density distribution.

[0071] Compared with the prior art, the present invention has the following advantages and technical effects:

[0072] (1) High degree of automation: The present invention adopts a deep learning method, which can automatically learn the correspondence between the CT value and the density value of human tissue without human intervention, significantly improving the degree of automation of the calculation and reducing human errors.

[0073] (2) More accurate calculation results: The present invention constructs a human CT data set with a phantom, and uses the known phantom density and the measured phantom HU value to convert the HU value of human tissue into a density value. The deep learning model is trained based on a large amount of data and can accurately convert the HU value into a density value, avoiding the calculation error that may be caused by the assumption that there is a linear relationship between the tissue gray value and the density value in the traditional method, thereby providing a more accurate calculation result.

[0074] (3) Wide scope of application: The present invention can not only calculate bone density values, but also calculate other types of tissue density values, such as muscle density values, etc. It has a wider scope of application and can meet the needs of different types of medical image analysis.

[0075] (4) High computational efficiency: The present invention significantly improves computational efficiency when calculating human tissue density values ​​based on deep learning technology. Compared with traditional methods, this solution can greatly save time and computing resources and improve overall processing speed.

[0076] (5) Strong equipment versatility: The present invention has good generalization ability by training on data collected on a variety of CT devices, and can adapt to the differences between different CT scanners. This versatility makes the measurement results between different medical institutions more comparable, which is conducive to multi-center research and remote diagnosis.

[0077] (6) Significant clinical value: The present invention can assist doctors in making more precise diagnoses and treatment decisions by providing more accurate and comprehensive tissue density information. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0079] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0080] Figure 2 Schematic diagram of the architecture of the deep learning network model of an embodiment of the present invention. DETAILED DESCRIPTION

[0081] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0082] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0083] Embodiment 1

[0084] like Figure 1-2 As shown, this embodiment provides a method for measuring the mineral density of human tissue, comprising:

[0085] Collect human tissue mineral density data, process the human tissue mineral density data, and construct a human tissue mineral density data set;

[0086] Construct a deep learning network model, and learn the corresponding relationship between the CT value and density value of human tissue based on the deep learning network model;

[0087] Each tissue area is segmented according to the corresponding relationship, the bone density value is calculated through the tissue density value of each tissue area, and the visual measurement results are output.

[0088] Furthermore, the human tissue mineral density data is collected and processed, and the process of constructing a human tissue mineral density data set includes:

[0089] CT scans are performed simultaneously on humans and phantoms of different ages, genders, and health conditions, and human CT images are collected to ensure sample diversity.

[0090] The human CT images are standardized and denoised to improve image quality. Then different tissue types in the human CT images are labeled to ensure data accuracy.

[0091] The phantom has a known mineral density, and the linear relationship between the HU value and the mineral density is obtained based on the regression of the mineral density data of the phantom. The linear relationship between the HU value and the mineral density is used to convert the human CT image into a density map;

[0092] A human tissue mineral density dataset based on paired human CT images and density maps;

[0093] Among them, the standardization process includes adjusting the resolution and adjusting the HU value range;

[0094] Tissue types include bone, muscle, fat, discs, and ligaments.

[0095] Furthermore, if Figure 2As shown, a deep learning network model is constructed, and the process of learning and obtaining the corresponding relationship between the CT value and the density value of human tissue based on the deep learning network model includes:

[0096] Improve the network architecture of 3D UNet to obtain an improved 3D UNet network;

[0097] Design a loss function and obtain an initial deep learning network model based on the improved 3D UNet network and the designed loss function;

[0098] The initial deep learning network model is trained by using human CT image data to obtain a deep learning network model;

[0099] The human CT image is input into the deep learning network model, the density map is output, and the corresponding relationship between the CT value and density value of human tissue is obtained.

[0100] Furthermore, the process of improving the network architecture of 3D UNet includes:

[0101] Based on the network architecture of 3D UNet, the output layer uses a 1x1x1 convolutional layer to map the feature map to the density value, and outputs a density map with the same spatial resolution as the input. One or more of the following improvements can be made to the architecture of 3D U-net:

[0102] For the encoder part, 3D ResNet blocks (ResNetb lock) are used to replace ordinary convolution blocks to enhance feature extraction capabilities. Squeeze-and-Excitation (SE) blocks are added after each downsampling step to enhance feature correlation between channels;

[0103] For the decoder part, transposed convolution is used for upsampling instead of simple interpolation to make the upsampling process learnable, and a densely connected block is added after each upsampling step to enhance feature reuse;

[0104] For skip connections, the attention gating mechanism (Attention Gates) is used to optimize the skip connections so that the network pays more attention to relevant areas.

[0105] Furthermore, the process of designing the loss function includes:

[0106] The model is fully optimized by combining the loss functions, where:

[0107] Mean Square Error:

[0108] L_mse=1 / N*Σ(Y_true-Y_pred)^2

[0109] Where N is the total number of voxels, Y_true and Y_pred represent the true density distribution map and the predicted density distribution map respectively;

[0110] Structural Similarity Loss:

[0111] L_ssim=1-SSIM_3D(Y_true,Y_pred)

[0112] Among them, SSIM_3D is the similarity between the calculated predicted density distribution map (Y_pred) and the true density distribution map (Y_true). When Y_true and Y_pred are exactly the same, SSIM_3D = 1. When Y_true and Y_pred are very different, SSIM_3D is close to -1.

[0113] Gradient Difference Loss:

[0114]

[0115] in, represents a 3D gradient operator;

[0116] Total loss:

[0117] L_total=α*L_mse+β*L_ssim+γ*L_grad

[0118] Among them; α, β, γ represent the weights for balancing different loss terms.

[0119] Furthermore, the initial deep learning network model is trained by human CT image data, and the process of obtaining the deep learning network model includes:

[0120] The initial deep learning network model is trained in stages, followed by learning rate scheduling and data augmentation;

[0121] The phased training process includes:

[0122] Phase 1: Preliminary training using only MSE loss;

[0123] The second stage: introduce SSIM and gradient loss to fine-tune the network;

[0124] The learning rate schedule uses a cosine annealing learning rate with a warm-up period; the formula is:

[0125] lr=lr_max*0.5*(1+cos(π*t / T))

[0126] Where t is the current step number and T is the total step number;

[0127] Data augmentation includes random rotation (in the range of [-15°, 15°]), random scaling (in the range of [0.9, 1.1]), random noise addition (Gaussian noise, σ in the range of [0, 0.05]), and elastic deformation (maximum displacement is 5% of the original voxel size).

[0128] Furthermore, the process of segmenting each tissue region according to the corresponding relationship includes:

[0129] Input the original CT image, segment each tissue area in the CT image through the deep learning network model, and output the tissue segmentation mask;

[0130] Morphological operations, such as dilation and erosion, are then applied to refine the segmentation results, remove small isolated regions, and identify major structures using connected component analysis.

[0131] Furthermore, the process of calculating the bone density value through the tissue density value of each tissue area includes:

[0132] Extract tissue density values ​​for each tissue region, and calculate bone density values ​​based on the extracted tissue density values;

[0133] The process of extracting tissue density values ​​for each tissue region includes:

[0134] The density map is combined with the tissue segmentation mask to extract the density values ​​of the main bone structures and calculate the statistical characteristics of the bone structures;

[0135] The process of calculating bone density values ​​based on extracted tissue density values ​​includes:

[0136] Bone density was calculated for major skeletal structures as well as T-scores and Z-scores based on age-, sex-, and race-specific reference databases.

[0137] Furthermore, the statistical characteristics of bone structure include mean density, median density, standard deviation of density, and volume-weighted mean density;

[0138] For major bone structures, such as lumbar vertebrae T11-L4, the formula for calculating bone density is:

[0139] BMD=Σ(Vi*Di) / ΣVi

[0140] Among them, Vi is the volume of the i-th voxel, and Di is its density value;

[0141] Based on age-, sex-, and race-specific reference databases, the formulas for calculating T scores and Z scores are:

[0142] T score = (measured BMD-mean BMD of young reference population) / standard deviation of BMD of young reference population;

[0143] Z score = (measured BMD - average BMD of the reference population of the same age) / standard deviation of BMD of the reference population of the same age.

[0144] Furthermore, the process of outputting the visual measurement results includes:

[0145] Generate a bone density report, which includes: BMD values ​​of major bone parts,

[0146] T-scores and Z-scores, as well as color coding to visually display density distribution.

[0147] The present invention provides a more accurate and efficient method for calculating tissue density by improving the existing technology and introducing a deep learning method. Through an automated calculation method, human intervention is reduced, calculation efficiency is improved, and the risk of human error is reduced.

[0148] The present invention improves the accuracy of calculation by adopting a more complex and accurate model, and can realize the calculation of density values ​​of various types of human tissues, such as muscle density values, etc., thereby providing more comprehensive diagnostic information and research data.

[0149] Based on the generalization ability of the deep learning model, the present invention can adapt to the differences between different CT devices, improve the consistency and comparability of measurement results, and thus improve the versatility of the equipment.

[0150] By providing more accurate and comprehensive tissue density information, the present invention can assist doctors in making more precise diagnosis and treatment decisions, such as early diagnosis of osteoporosis and assessment of muscle atrophy. For example, in the early diagnosis of osteoporosis, this solution can detect subtle changes that may be overlooked by traditional methods, which helps with early intervention and treatment, thereby enhancing the value of clinical applications.

[0151] Through the above improvements, the present invention is superior to the prior art in terms of automation level, calculation accuracy, application scope and calculation efficiency, and has obvious technical advantages and practical value.

[0152] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for measuring the mineral density of human tissue, characterized in that: include: Collecting human tissue mineral density data, and processing the human tissue mineral density data to construct a human tissue mineral density data set; Constructing a deep learning network model, and learning and obtaining the corresponding relationship between the CT value and the density value of human tissue based on the deep learning network model; Each tissue region is segmented according to the corresponding relationship, the bone density value is calculated by the tissue density value of each tissue region, and a visual measurement result is output.

2. The method for measuring the mineral density of human tissue according to claim 1, characterized in that: The process of collecting human tissue mineral density data, processing the human tissue mineral density data, and constructing a human tissue mineral density data set includes: Perform CT scans on humans and phantoms of different ages, genders, and health conditions simultaneously to collect human CT images; Performing standardization and denoising on the human body CT image, and then marking different tissue types in the human body CT image; Obtaining a linear relationship between HU value and mineral density based on the mineral density data regression of the phantom, and converting the human body CT image into a density map using the linear relationship between HU value and mineral density; A human tissue mineral density dataset of paired human CT images and density maps is constructed based on the human CT images and density maps; Wherein, the standardization process includes adjusting the resolution and adjusting the HU value range; Such tissue types include bone, muscle, fat, intervertebral disc and ligaments.

3. The method for measuring the mineral density of human tissue according to claim 1, characterized in that: The process of constructing the deep learning network model and learning and obtaining the corresponding relationship between the CT value and the density value of human tissue based on the deep learning network model includes: Improve the network architecture of 3D UNet to obtain an improved 3D UNet network; Designing a loss function, and constructing an initial deep learning network model based on the improved 3D UNet network and the designed loss function; The initial deep learning network model is trained by using human CT image data to obtain the deep learning network model; The human body CT image is input into the deep learning network model, a density map is obtained as output, and the corresponding relationship between the human body tissue CT value and the density value is obtained.

4. The method for measuring the mineral density of human tissue according to claim 3, characterized in that: The process of improving the network architecture of 3D UNet includes: Based on the network architecture of 3D UNet, the output layer uses a 1x1x1 convolution layer to map the feature map to the density value, and outputs a density map with the same spatial resolution as the input; For the encoder part, we use 3D ResNet blocks instead of ordinary convolution blocks and add Squeeze-and-Excitation blocks after each downsampling step. For the decoder part, transposed convolution is used for upsampling instead of simple interpolation to make the upsampling process learnable, and dense connection blocks are added after each upsampling step to enhance feature reuse; For skip connections, an attention gating mechanism is used to optimize the skip connections so that the network pays more attention to relevant areas.

5. The method for measuring the mineral density of human tissue according to claim 3, characterized in that: The process of designing a loss function includes: The model is fully optimized by combining the loss functions, where: Mean Square Error: L_mse=1 / N*Σ(Y_true-Y_pred)^2 Where N is the total number of voxels, Y_true and Y_pred represent the true density distribution map and the predicted density distribution map respectively; Structural Similarity Loss: L_ssim=1-SSIM_3D(Y_true,Y_pred) Among them, SSIM_3D is the similarity between the calculated predicted density distribution map (Y_pred) and the true density distribution map (Y_true); Gradient Difference Loss: in, represents a 3D gradient operator; Total loss: L_total=α*L_mse+β*L_ssim+γ*L_grad Among them; α, β, γ represent the weights for balancing different loss terms.

6. The method for measuring the mineral density of human tissue according to claim 3, characterized in that: The initial deep learning network model is trained by human CT image data to obtain the deep learning network model, and the process includes: The initial deep learning network model is trained in stages, followed by learning rate scheduling and data augmentation; The process of staged training includes: Phase 1: Preliminary training using only MSE loss; The second stage: introduce SSIM and gradient loss to fine-tune the network; The learning rate scheduling uses a cosine annealing learning rate with a warm-up period; the formula expression is: lr=lr_max*0.5*(1+cos(π*t / T)) Where t is the current step number and T is the total step number; The data enhancement includes random rotation, random scaling, random noise addition, and elastic deformation.

7. The method for measuring the mineral density of human tissue according to claim 1, characterized in that: The process of segmenting each tissue region according to the corresponding relationship includes: Input the original CT image, segment each tissue area in the CT image through the deep learning network model, and output the tissue segmentation mask; Morphological operations are then applied to refine the segmentation results, remove small isolated regions, and identify major structures using connected component analysis.

8. The method for measuring the mineral density of human tissue according to claim 1, characterized in that: The process of calculating bone density values ​​from tissue density values ​​of each tissue area includes: Extract tissue density values ​​for each tissue region, and calculate bone density values ​​based on the extracted tissue density values; The process of extracting tissue density values ​​for each tissue region includes: The density map is combined with the tissue segmentation mask to extract the density values ​​of the main bone structures and calculate the statistical characteristics of the bone structures; The process of calculating the bone density value according to the extracted tissue density value comprises: Bone density was calculated for major skeletal structures as well as T-scores and Z-scores based on age-, sex-, and race-specific reference databases.

9. The method for measuring the mineral density of human tissue according to claim 8, characterized in that: The statistical characteristics of the bone structure include average density, median density, standard deviation of density, and volume-weighted average density; The formula for calculating bone density for the main bone structures is: BMD=Σ(Vi*Di) / ΣVi Among them, Vi is the volume of the i-th voxel, and Di is its density value; Based on age-, sex-, and race-specific reference databases, the formulas for calculating T scores and Z scores are: T score = (measured BMD-mean BMD of young reference population) / standard deviation of BMD of young reference population; Z score = (measured BMD - average BMD of the reference population of the same age) / standard deviation of BMD of the reference population of the same age.

10. The method for measuring the mineral density of human tissue according to claim 1, characterized in that: The process of outputting the visual measurement results includes: Generate a bone density report, wherein the bone density report includes: BMD values, T scores and Z scores of each major bone site, and intuitive display of density distribution using color coding.