An osteoporosis evaluation method based on pelvic CT plain scan images

Abstract

This invention relates to the field of osteoporosis detection technology and discloses a method for osteoporosis assessment based on pelvic CT plain scan images. First, DICOM format images are read, specifically pelvic CT images are read using DICOM to obtain pixel data and metadata. A U-Net model is applied for pixel-level segmentation to generate a bone tissue mask, and morphological operations are used to optimize the segmentation boundaries and reduce artifact interference. Next, bone structure features are extracted, multi-dimensional features are quantified, and the average gray value and standard deviation of the bone region are calculated to reflect bone density distribution. Texture features are also extracted, including contrast, energy, and correlation parameters extracted based on the gray-level co-occurrence matrix to quantify the trabecular microstructure. A machine learning model is constructed and evaluated, and the osteoporosis is diagnosed based on the lumbar spine quantitative CT bone mineral density diagnostic criteria. This invention achieves two goals at once: improving the utilization efficiency of medical resources and expanding the application scenarios of osteoporosis assessment.

Description

Technical Field

[0001] This invention relates to the field of osteoporosis detection technology, specifically to an osteoporosis assessment method based on pelvic CT plain scan images. Background Technology

[0002] Traditional CT images of the pelvic bone region contain various tissues and organs, such as muscles, fat, and intestines. The gray values ​​of these tissues overlap to some extent with those of bone tissue, making it difficult to accurately segment the bone region directly from the image. Traditional methods struggle to effectively distinguish bone tissue from surrounding tissues with similar gray values, thus affecting subsequent osteoporosis assessments.

[0003] In actual clinical practice, pelvic CT images acquired have different resolutions. Images with different resolutions have differences in feature extraction and analysis. If they are not processed uniformly, it will affect the accuracy and consistency of bone region segmentation.

[0004] The limitations of single features: relying solely on a single type of feature (such as grayscale features only or texture features only) makes it difficult to comprehensively and accurately reflect the health status of bones. For example, grayscale features can only reflect the general situation of bone density, but cannot reflect the microstructural information of bone trabeculae; while texture features can capture the microstructure of bone trabeculae, they lack the quantification of the overall distribution of bone density.

[0005] Extracting a large number of features presents challenges such as high computational complexity and low efficiency. Furthermore, some features may be redundant or correlated, leading to an imbalance between feature extraction efficiency and accuracy, which negatively impacts the accuracy of the assessment. Handling complex feature relationships is also problematic, as the extracted features exhibit intricate nonlinear relationships that traditional linear models struggle to accurately capture, resulting in inaccurate bone density predictions.

[0006] Limitations of invasive examinations: Traditional osteoporosis assessment methods, such as bone biopsy, are invasive examinations that can cause pain and risks to patients. They are also complex to perform and costly, making them unsuitable for large-scale screening and regular follow-up.

[0007] Therefore, to address the above issues, a method for assessing osteoporosis based on pelvic CT plain scan images is needed. Summary of the Invention

[0008] The purpose of this invention is to provide a method for osteoporosis assessment based on pelvic CT plain scan images. This invention allows for osteoporosis assessment while the patient undergoes a pelvic CT examination, achieving two goals at once, improving the efficiency of medical resource utilization, and expanding the application scenarios of osteoporosis assessment.

[0009] This invention is implemented as follows:

[0010] This invention provides a method for assessing osteoporosis based on pelvic CT plain scan images, specifically implemented according to the following steps:

[0011] S1: Read DICOM format images. Specifically, use a DICOM reading library (such as the pydicom library in Python) to read pelvic CT images and obtain the image's pixel data and metadata. The metadata includes pixel spacing (Spacing) and grayscale range (HU value). Follow these steps:

[0012] S1.1: Resampling to uniform resolution, assuming the pixel spacing of the original image is...

[0013] The target resolution is 1mm×1mm×1mm; according to the resampling principle, for each new pixel in the image... The pixel value at the corresponding position (x, y, z) in the original image is obtained through an interpolation algorithm (such as bilinear interpolation); as shown in the following formula:

[0014] ;

[0015] ;

[0016] ;

[0017] S1.2: Let the original HU value range be... The optimal display range for the target bone tissue is For each pixel in the image, its transformed HU value As shown in the following formula:

[0018]

[0019] in, The original HU value, The value is the value after transformation and normalized to the interval [0, 1].

[0020] S2: Perform bone region segmentation and mask generation, specifically through adaptive threshold segmentation and deep learning segmentation; combine adaptive radial line method to locate bone regions (such as hip bone and sacrum), apply U-Net model for pixel-level segmentation, generate bone tissue mask, and optimize segmentation boundary through morphological operations; reduce artifact interference.

[0021] Follow these steps:

[0022] S2.1: Adaptive radial line method is used to locate bone regions. Specifically, multiple radial lines are emitted from the center of the image. Along each radial line, the boundary of the bone region is determined based on the sudden change in pixel value (such as a sudden change from the HU value range of soft tissue to the HU value range of bone tissue). This allows the location of bone regions such as the hip bone and sacrum to be located.

[0023] S2.2: Let the input image be I, and the bone tissue mask obtained after segmentation by the U-Net model be... ,in 0 represents non-bone tissue, and 1 represents bone tissue;

[0024] S2.3: Perform morphological operations to optimize the segmentation boundary. Let the structuring element be SE (usually a small circular or square matrix that fills small holes in the bone tissue mask, smoothing the segmentation boundary and reducing artifact interference). The calculation formula is as follows:

[0025]

[0026] in, This indicates an expansion operation. This indicates a corrosion operation. This is the optimized bone tissue mask.

[0027] S3: Perform bone structure feature extraction, multi-dimensional feature quantization, calculate the average gray value and standard deviation of the bone region to reflect bone density distribution; and texture features, including extracting contrast, energy, and correlation parameters based on the gray-level co-occurrence matrix to quantify the microstructure of bone trabeculae; specifically, perform the following steps:

[0028] S3.1: First, calculate the grayscale features and the average grayscale value of the bone region. As shown in the following formula:

[0029]

[0030] in, This represents the total number of pixels within the bone region. For the first in the bone region The HU value of each pixel;

[0031] S3.2: Calculate the standard deviation of the bone region The specific formula is as follows:

[0032]

[0033] S3.3: Calculate texture features based on the Gray-Level Co-occurrence Matrix (GLCM), and construct the GLCM. Let the number of gray levels in the image be L, and the GLCM be... It is an L×L matrix, where d is the distance between pixels. The direction between pixels (usually taken as) , Indicates direction Above, the probability of occurrence of two pixels with a distance of d, one with grayscale value i and the other with grayscale value j, is used to extract the contrast, as shown in the following formula:

[0034]

[0035] in, For contrast;

[0036] The specific formula for energy extraction is as follows:

[0037]

[0038] in, For energy;

[0039] The correlation is extracted using the following formula:

[0040]

[0041] in, For correlation, .

[0042] S4: To build and evaluate machine learning models, feature fusion and model training are performed first. Gray level, texture, contrast, energy, and correlation parameters are integrated into feature vectors to build a regression model to predict bone density T-values. During model training, the training set and test set are divided in a 7:3 ratio, and five-fold cross-validation is performed.

[0043] Follow these steps:

[0044] S4.1: Feature fusion and model training, combining grayscale features , Texture features, along with contrast, energy, and correlation parameters (such as the area and perimeter of extracted bone regions), are integrated into a single feature vector. , where n is the total number of features;

[0045] S4.2: Construct a regression model using the LGBM algorithm, and let the model parameters be... The model's prediction function is The goal is to learn the optimal parameters. This minimizes the error between the predicted value and the actual bone mineral density (T) value; the mean squared error (MSE) is used as the loss function. The specific formula is as follows:

[0046]

[0047] in, The number of training samples For the first The true bone mineral density T-value of each sample For the first Feature vectors of each sample;

[0048] S4.3: Divide the dataset into training and test sets in a 7:3 ratio and perform five-fold cross-validation. This involves further dividing the training set into five subsets, using four subsets for training each time, and using the remaining subset for validation. Repeat this process five times and take the average performance metric MSE. Evaluate the model's performance.

[0049] S5: Determine the bone mineral density (BMD) value based on the predicted value. Let the predicted BMD value from step S4 be BMD (unit: mg / cm³). Determine the value according to the diagnostic criteria for osteoporosis using quantitative CT bone mineral density analysis of the lumbar spine. According to these criteria, BMD ≤ 80 mg / cm³ indicates osteoporosis, BMD > 80 mg / cm³ but < 120 mg / cm³ indicates osteopenia, and BMD ≥ 120 mg / cm³ indicates normal bone mass. The specific formula is as follows:

[0050] .

[0051] Furthermore, the present invention provides an osteoporosis assessment system based on pelvic CT plain scan images, including an image acquisition and preprocessing module, which interfaces with the hospital's existing CT scanning equipment to acquire CT plain scan image data of the patient's pelvic region in real time.

[0052] The bone region segmentation module employs an adaptive radiative method for preliminary localization. It radiates rays outward from the image center, statistically analyzing the grayscale changes along each ray. Based on the grayscale characteristics of bone tissue in CT images, it determines the extent of the bone region. A trained U-Net model then performs pixel-level fine segmentation on the pre-localized image, generating an accurate bone tissue mask. The module further optimizes the segmentation results generated by the U-Net model, such as removing small isolated regions and filling holes, to further improve the accuracy of bone region segmentation and ensure the reliability of subsequent feature extraction.

[0053] The feature extraction module performs grayscale feature extraction and calculates the average grayscale value and standard deviation of the bone region. The average grayscale value reflects the overall density of bone tissue, while the standard deviation reflects the uniformity of bone density distribution, providing a basic quantitative indicator for osteoporosis assessment.

[0054] The bone mineral density prediction and osteoporosis assessment module normalizes or standardizes extracted features, unifying features of different dimensions to the same scale range. Based on the preprocessed feature data, it predicts the patient's bone mineral density (T-score) and compares the predicted T-score with a standard threshold according to the diagnostic criteria for lumbar spine quantitative CT bone mineral density, determining whether the patient has osteoporosis and its severity (e.g., mild, moderate, severe). This provides clinicians with a clear diagnostic result.

[0055] Furthermore, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the osteoporosis assessment method based on pelvic CT plain scan images as described above.

[0056] Furthermore, the present invention provides a computer-storable medium storing a computer program, wherein when the program is executed, it sequentially executes any one of the above-described methods for assessing osteoporosis based on pelvic CT plain scan images.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] 1. This invention directly reflects bone density distribution by calculating the average gray value and standard deviation of bone regions. The average gray value is related to the bone mineral content, while the standard deviation reflects the uniformity of bone density distribution, providing a fundamental and crucial quantitative indicator for osteoporosis assessment. Based on parameters such as contrast, energy, and correlation extracted from the gray-level co-occurrence matrix, the microstructure of trabeculae can be quantified. Changes in trabecular microstructure are one of the important characteristics of osteoporosis. These texture features can capture information such as the arrangement, spacing, and connectivity of trabeculae, thereby providing a more comprehensive assessment of bone health.

[0059] 2. The adaptive radial line method of this invention can initially locate bone regions, providing a more accurate segmentation range for deep learning models and reducing the processing range and computational load of the models. The U-Net model, with its powerful pixel-level segmentation capabilities, performs fine segmentation of bone tissue, generating accurate bone tissue masks. This combination of methods fully leverages the rapid localization advantage of adaptive methods and the high-precision segmentation advantage of deep learning models, improving the accuracy of bone region segmentation and thus enhancing the accuracy of subsequent feature extraction and osteoporosis assessment.

[0060] 3. This invention resamples pelvic CT images of different resolutions to a uniform resolution of 1mm×1mm×1mm, simplifying the subsequent image processing and analysis process, avoiding calculation errors and complexity caused by inconsistent resolutions, and improving the efficiency of the entire evaluation process. Window transformation is used to adjust the HU value to the optimal display range of bone tissue and normalize it to the [0,1] interval, highlighting the characteristics of bone tissue, reducing interference from irrelevant information, and making subsequent segmentation and feature extraction more efficient. Attached Figure Description

[0061] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0062] Figure 1 This is a flowchart of the method of the present invention;

[0063] Figure 2 This is a system module structure diagram of the present invention;

[0064] Figure 3 This is a segmented image of the sacral region of the present invention;

[0065] Figure 4 This is a region segmentation image of the iliac bone according to the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to describe selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Please see Figures 1-4 This invention provides a method for assessing osteoporosis based on pelvic CT plain scan images, specifically implemented according to the following steps:

[0068] S1: Read DICOM format images. Specifically, use a DICOM reading library (such as the pydicom library in Python) to read pelvic CT images and obtain the image's pixel data and metadata. The metadata includes pixel spacing (Spacing) and grayscale range (HU value). Follow these steps:

[0069] S1.1: Resampling to uniform resolution, assuming the pixel spacing of the original image is...

[0070] The target resolution is 1mm×1mm×1mm; according to the resampling principle, for each new pixel in the image... The pixel value at the corresponding position (x, y, z) in the original image is obtained using the bilinear interpolation algorithm, as shown in the following formula:

[0071] ;

[0072] ;

[0073] ;

[0074] S1.2: Let the original HU value range be... The optimal display range for the target bone tissue is For each pixel in the image, its transformed HU value As shown in the following formula:

[0075]

[0076] in, The original HU value, The value is the value after transformation and normalized to the interval [0, 1].

[0077] S2: Perform bone region segmentation and mask generation, specifically through adaptive threshold segmentation and deep learning segmentation; combine adaptive radial line method to locate bone regions (such as hip bone, sacrum), apply the U-Net model for pixel-level segmentation, generate bone tissue masks, and optimize segmentation boundaries through morphological operations; reduce artifact interference; specifically, follow these steps:

[0078] S2.1: Adaptive radial line method is used to locate bone regions. Specifically, multiple radial lines are emitted from the center of the image. Along each radial line, the boundary of the bone region is determined by the sudden change of the pixel value from the soft tissue HU value range to the bone tissue HU value range, thereby locating the position of bone regions such as the hip bone and sacrum.

[0079] S2.2: Let the input image be I, and the bone tissue mask obtained after segmentation by the U-Net model be... ,in 0 represents non-bone tissue, and 1 represents bone tissue;

[0080] S2.3: Perform morphological operations to optimize the segmentation boundary. Let the structuring element be SE, typically a small circular or square matrix, to fill small holes in the bone tissue mask, smooth the segmentation boundary, and reduce artifact interference. The calculation formula is as follows:

[0081]

[0082] in, This indicates an expansion operation. This indicates a corrosion operation. This is the optimized bone tissue mask.

[0083] S3: Perform bone structure feature extraction, multi-dimensional feature quantization, calculate the average gray value and standard deviation of the bone region to reflect bone density distribution; and texture features, including extracting contrast, energy, and correlation parameters based on the gray-level co-occurrence matrix to quantify the microstructure of bone trabeculae; specifically, perform the following steps:

[0084] S3.1: First, calculate the grayscale features and the average grayscale value of the bone region. As shown in the following formula:

[0085]

[0086] in, This represents the total number of pixels within the bone region. For the first in the bone region The HU value of each pixel;

[0087] S3.2: Calculate the standard deviation of the bone region The specific formula is as follows:

[0088]

[0089] S3.3: Calculate texture features based on the Gray-Level Co-occurrence Matrix (GLCM), and construct the GLCM. Let the number of gray levels in the image be L, and the GLCM be... It is an L×L matrix, where d is the distance between pixels. The direction between pixels is usually taken as... , Indicates direction Above, the probability of occurrence of two pixels with a distance of d, one with grayscale value i and the other with grayscale value j, is used to extract the contrast, as shown in the following formula:

[0090]

[0091] in, For contrast;

[0092] The specific formula for energy extraction is as follows:

[0093]

[0094] in, For energy;

[0095] The correlation is extracted using the following formula:

[0096]

[0097] in, For correlation, .

[0098] S4: Machine learning model construction and evaluation. First, feature fusion and model training are performed, integrating grayscale, texture, contrast, energy, and correlation parameters into a feature vector to build a regression model that predicts bone mineral density T-values. During model training, the training and test sets are divided in a 7:3 ratio, and five-fold cross-validation is performed. Specifically, follow these steps:

[0099] S4.1: Feature fusion and model training, combining grayscale features , Texture features, contrast, energy, and correlation parameters, along with extracted bone region area and perimeter, are integrated into a single feature vector. , where n is the total number of features;

[0100] In this embodiment, the feature vector set X includes grayscale features. , Texture features and contrast ,energy , A dataset containing a set of correlation parameters;

[0101] S4.2: Construct a regression model using the LGBM algorithm, and let the model parameters be... The model's prediction function is The goal is to learn the optimal parameters. This minimizes the error between the predicted value and the actual bone mineral density (T) value; the mean squared error (MSE) is used as the loss function. The specific formula is as follows:

[0102]

[0103] in, The number of training samples For the first The true bone mineral density T-value of each sample For the first Feature vectors of each sample;

[0104] S4.3: Divide the dataset into training and test sets in a 7:3 ratio and perform five-fold cross-validation, that is, further divide the training set into 5 subsets, use 4 subsets for training each time, and use the remaining subset for validation, repeat 5 times, and take the average performance metrics MSE and R² to evaluate the model performance.

[0105] In this embodiment, the average performance index (MSE) is trained as follows:

[0106]

[0107] in, : No. Validation set index, For the first The number of samples in the validation set. , This represents the total number of training samples; For the first The true bone mineral density T-value of each sample; For the first Folding model for feature vectors The predicted value, For the first The minimum loss model parameters obtained after training;

[0108] in As shown in the following formula:

[0109]

[0110] in, For the first The training set for folding;

[0111] The average MSE is calculated as follows:

[0112]

[0113] Five-fold cross-validation Calculate as follows:

[0114] Single-fold validation set :

[0115]

[0116] in, , The mean true bone mineral density of the k-th fold validation set;

[0117] average As shown in the following formula:

[0118]

[0119] In step S4.3, the data is first partitioned: the total training set (accounting for 70% of the original data) is divided into 5 mutually exclusive subsets. The number of samples in each subset is equal. Then cross-validate the loop: for each :

[0120] by Train the model using the total training set / Dk (4 / 5 data points) and optimize the parameters. (Minimize the training set MSE); on the validation set Calculation on (1 / 5 of the data) and .

[0121] Final performance evaluation, using 50% of the results. and The average value is used as the final performance indicator. In this invention, and The average values ​​are defined by professionals in this field, and this invention does not impose any specific limitations on them.

[0122] S5:LGBM Model The regression output is taken as input feature vector X, and the predicted bone mineral density (BMD) value is output based on the input feature vector. The judgment is then made based on the predicted BMD value, according to the prediction model trained in step S4.3. The predicted bone mineral density value is BMD (unit: mg / cm3), which is determined according to the diagnostic criteria for osteoporosis using quantitative CT bone mineral density of the lumbar spine. According to the diagnostic criteria for osteoporosis using quantitative CT bone mineral density of the lumbar spine, BMD ≤ 80 mg / cm3 indicates osteoporosis, BMD > 80 mg / cm3 and < 120 mg / cm3 indicates osteopenia, and BMD ≥ 120 mg / cm3 indicates normal bone mass, as shown in the following formula:

[0123] .

[0124] In this embodiment, the specific principle is as follows: Step S1 performs data preprocessing, followed by step S2 bone segmentation; in step S1, the input data includes DICOM raw pixel data and metadata (pixel spacing). Output: Resampled image (resolution 1mm×1mm×1mm), normalized HU value ([0,1] interval), specifically obtained through resampling and coordinate transformation formula. The new coordinates are mapped to the original physical space to ensure isotropic resolution, providing a unified scale basis for subsequent segmentation.

[0125] Then, the HU value is normalized and piecewise linearly transformed to enhance the contrast between bone tissue and surrounding tissue, thereby improving the feature extraction capability of the U-Net segmentation network.

[0126] In this embodiment, during bone segmentation in step S2, the input data is resampled, the HU value is normalized, and then feature extraction is performed in step S3, outputting optimized bone tissue mask Mop (0 / 1 binary map) data. In step S2, the adaptive radiometric method locates the hip / sacral boundary through HU value mutations, generating the initial mask Mraw. Then, morphological operations of dilation and erosion are used to fill voids within the mask and smooth edges, reducing the impact of CT artifacts such as metallic textures on segmentation.

[0127] In this embodiment, the mask Mop serves as a spatial filter, and in S3, feature calculation is performed only on the bone region of the pixels within the mask to exclude interference from non-bone tissue.

[0128] Feature extraction is performed in S3, followed by model training in S4. In this embodiment, the inputs are the bone tissue mask Mop and the original HU value, and the outputs are data and multidimensional feature vectors. In steps S3-S4, the grayscale features are... Average gray level reflects the overall bone mineral density level. Standard deviation quantifies the dispersion of bone mineral density distribution, and the two are directly related to trabecular bone density.

[0129] Furthermore, in terms of texture features, GLCM-based contrast Co, energy En, and correlation Corr quantify the spatial arrangement patterns of trabecular microstructures, such as anisotropy and homogeneity.

[0130] The feature vector X integrates grayscale and texture features to form a high-dimensional representation, which serves as the input to the LGBM regression model in S4.

[0131] In this embodiment, model training is performed in step S4, and then clinical judgment is made in step S5; the input feature vector X and the true bone mineral density T value (clinical gold standard) are used to output the predicted bone mineral density value BMD.

[0132] In this embodiment, the present invention provides an osteoporosis assessment system based on pelvic CT plain scan images, including an image acquisition and preprocessing module, which interfaces with the hospital's existing CT scanning equipment to acquire CT plain scan image data of the patient's pelvic region in real time.

[0133] The bone region segmentation module employs an adaptive radiative method for preliminary localization. It radiates rays outward from the image center, statistically analyzing the grayscale changes along each ray. Based on the grayscale characteristics of bone tissue in CT images, it determines the extent of the bone region. A trained U-Net model then performs pixel-level fine segmentation on the pre-localized image, generating an accurate bone tissue mask. The module further optimizes the segmentation results generated by the U-Net model, such as removing small isolated regions and filling holes, to further improve the accuracy of bone region segmentation and ensure the reliability of subsequent feature extraction.

[0134] The feature extraction module performs grayscale feature extraction and calculates the average grayscale value and standard deviation of the bone region. The average grayscale value reflects the overall density of bone tissue, while the standard deviation reflects the uniformity of bone density distribution, providing a basic quantitative indicator for osteoporosis assessment.

[0135] The bone mineral density prediction and osteoporosis assessment module normalizes or standardizes extracted features, unifying features of different dimensions to the same scale range. Based on the preprocessed feature data, it predicts the patient's bone mineral density (T-score) and compares the predicted T-score with a standard threshold according to the diagnostic criteria for lumbar spine quantitative CT bone mineral density, determining whether the patient has osteoporosis and its severity (e.g., mild, moderate, severe). This provides clinicians with a clear diagnostic result.

[0136] In this embodiment, as Figures 3-4 The present invention calculates the above method using data from specific embodiments: the original data is shown in Table 1:

[0137] Table 1. Obtaining raw data

[0138]

[0139] Bone region segmentation and mask generation are shown in Table 2:

[0140] Table 2 Bone region segmentation and mask generation data

[0141]

[0142] Bone structure features were extracted, as shown in Table 3:

[0143] Table 3 Bone structure feature extraction data

[0144]

[0145] Calculate texture features, as shown in Table 4:

[0146] Table 4 Texture Feature Data

[0147]

[0148] In this embodiment, machine learning model construction and evaluation are performed, as shown in Table 5:

[0149] Table 5. Machine Learning Model Construction and Evaluation Data

[0150]

[0151] A comprehensive assessment of osteoporosis was conducted, as shown in Table 6;

[0152] Table 6. Comprehensive assessment data for osteoporosis

[0153]

[0154] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations will be apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for assessing osteoporosis based on pelvic CT plain scan images, characterized in that: Follow these steps: S1: Read DICOM format images. Specifically, use the DICOM reading library to read pelvic CT images and obtain the image's pixel data and metadata, including pixel spacing (Spacing) and grayscale (HU) values. S2: Perform bone region segmentation and mask generation, specifically through adaptive threshold segmentation and deep learning segmentation; combine adaptive radii method to locate bone regions, apply U-Net model for pixel-level segmentation, generate bone tissue masks, and optimize segmentation boundaries through morphological operations. S3: Perform bone structure feature extraction, multi-dimensional feature quantification, calculate the average gray value and standard deviation of the bone region to reflect the bone density distribution; and texture features, including extracting contrast, energy and correlation parameters based on the gray-level co-occurrence matrix to quantify the trabecular microstructure. S4: To build and evaluate machine learning models, feature fusion and model training are first performed. Gray values, texture features, contrast, energy, and correlation parameters are integrated into feature vectors to build a regression model to predict bone density T-values. During model training, the training set and test set are divided in a 7:3 ratio, and five-fold cross-validation is performed. S5: Make a judgment based on the predicted bone mineral density value. Let the bone mineral density value predicted in step S4 be BMD. Make a judgment based on the criteria for diagnosing osteoporosis using quantitative CT bone mineral density of the lumbar spine.

2. The osteoporosis assessment method based on pelvic CT plain scan images according to claim 1, characterized in that: In step S1, the following steps are specifically performed: S1.1: Pixel data acquisition specifically includes resampling to achieve a uniform resolution. Let the pixel spacing of the original image be... The target resolution is 1mm × 1mm × 1mm; for each new pixel in the image The pixel value at the corresponding position (x, y, z) in the original image is obtained through an interpolation algorithm, as shown in the following formula: ； ； ； S1.2: Let the original HU value range be... The optimal display range for the target bone tissue is For each pixel in the image, its transformed HU value As shown in the following formula: ； in, The original HU value, The value is the value after transformation and normalized to the interval [0, 1].

3. The osteoporosis assessment method based on pelvic CT plain scan images according to claim 1, characterized in that: In step S2, the following steps are specifically performed: S2.1: Adaptive radial line method is used to locate the bone region. Specifically, multiple radial lines are emitted from the center of the image. The boundary of the bone region is determined according to the change in pixel value along each radial line, thereby locating the position of the hip bone and sacrum bone regions. S2.2: Let the input image be I, and the bone tissue mask obtained after segmentation by the U-Net model be... ,in 0 represents non-bone tissue, and 1 represents bone tissue; S2.3: Perform morphological operations to optimize the segmentation boundary. Let the structuring element be SE, and the calculation formula is as follows: ； in, This indicates an expansion operation. This indicates a corrosion operation. This is the optimized bone tissue mask.

4. The osteoporosis assessment method based on pelvic CT plain scan images according to claim 1, characterized in that: In step S3, the following steps are specifically performed: S3.1: First, calculate the grayscale features and the average grayscale value of the bone region. As shown in the following formula: ； in, This represents the total number of pixels within the bone region. For the first in the bone region The HU value of each pixel; S3.2: Calculate the standard deviation of the bone region The specific formula is as follows: ； S3.3: Calculate texture features based on the Gray-Level Co-occurrence Matrix (GLCM), and construct the GLCM. Let the number of gray levels in the image be L, and the GLCM be... It is an L×L matrix, where d is the distance between pixels. The direction between pixels, Indicates direction Above, the probability of occurrence of two pixels with a distance of d, one with grayscale value i and the other with grayscale value j, is used to extract the contrast, as shown in the following formula: ； in, For contrast; The specific formula for energy extraction is as follows: ； in, For energy; The correlation is extracted using the following formula: ； in, For correlation, 。 5. The osteoporosis assessment method based on pelvic CT plain scan images according to claim 1, characterized in that: In step S4, the following steps are specifically performed: S4.1: Feature fusion and model training, combining grayscale features , Texture features, along with contrast, energy, and correlation parameters, are integrated into a single feature vector. , where n is the total number of features; S4.2: Construct a regression model using the LGBM algorithm, and let the model parameters be... The model's prediction function is The goal is to learn the optimal parameters. This minimizes the error between the predicted value and the actual bone mineral density T value; the mean squared error (MSE) is used as the loss function. The specific formula is as follows: ； in, The number of training samples For the first The true bone density of each sample value, For the first Feature vectors of each sample; S4.3: Divide the dataset into training and test sets in a 7:3 ratio and perform five-fold cross-validation. This involves further dividing the training set into five subsets, using four subsets for training each time, and using the remaining subset for validation. Repeat this process five times and take the average performance metric MSE. Evaluate the model's performance.

6. The osteoporosis assessment method based on pelvic CT plain scan images according to claim 1, characterized in that: In step S5, osteoporosis is diagnosed based on the quantitative CT bone mineral density (BMD) standard for the lumbar spine: BMD ≤ 80 mg / cm3 indicates osteoporosis, BMD > 80 mg / cm3 but < 120 mg / cm3 indicates osteopenia, and BMD ≥ 120 mg / cm3 indicates normal bone mass, as shown in the following formula: 。 7. An osteoporosis assessment system based on pelvic CT plain scan images, characterized in that: It includes an image acquisition and preprocessing module, which interfaces with the hospital's existing CT scanning equipment to acquire CT plain scan image data of the patient's pelvic area in real time; The bone region segmentation module uses an adaptive radiative method for preliminary localization. It radiates rays outward from the image center, counts the changes in grayscale values ​​on each ray, determines the extent of the bone region based on the grayscale characteristics of bone tissue in CT images, and uses a trained U-Net model to perform pixel-level fine segmentation on the preliminarily localized image to generate an accurate bone tissue mask. The segmentation results generated by the U-Net model are then optimized. The feature extraction module performs grayscale feature extraction and calculates the average grayscale value and standard deviation of the bone region. The bone mineral density prediction and osteoporosis assessment module normalizes or standardizes the extracted features, unifying features of different dimensions into the same scale range. Based on the preprocessed feature data, it predicts the patient's bone mineral density T-score and compares the predicted bone mineral density T-score with the standard threshold according to the diagnostic criteria for lumbar spine quantitative CT bone mineral density, to determine whether the patient has osteoporosis and the severity of osteoporosis.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the osteoporosis assessment method based on pelvic CT plain scan images as described in any one of claims 1-6.

9. A computer-storable medium storing a computer program therein, characterized in that: When the program is executed, it sequentially executes any one of the claims 1-6 above, namely, the osteoporosis assessment method based on pelvic CT plain scan images.

Images