Method and system for constructing osteoporosis detection data model

Abstract

The invention relates to an osteoporosis detection data model construction method and system, and the method comprises the steps: setting an age group layer according to the equal difference of five years, setting a height section layer according to the equal difference of the height of 5cm, and setting a gender section layer according to the male and female classification, and carrying out the grouping sampling of people with healthy bones through the three dimensions, making a healthy crowd bone mineral density medical image standard data set and an unhealthy crowd bone mineral density medical image standard data set; extracting image features from the bone mineral density medical image standard data set of the healthy people and the bone mineral density medical image standard data set of the unhealthy people under the same dimension, and calculating a bone mineral density medical image feature difference value of the unhealthy people under the dimension combination; and circulating the feature extraction and calculation process until the bone mineral density medical image feature difference values of the unhealthy people under all different dimension combinations are completed. According to the invention, the position and degree of osteoporosis of unspecific people can be more accurately and efficiently judged.

Description

Technical Field

[0001] The present invention relates to the field of biomedicine, and in particular, to a method and system for constructing an osteoporosis detection data model. Background Art

[0002] In the prior art, there are various medical imaging methods and medical imaging display methods to detect the degree and location of osteoporosis, but no one has proposed to improve the accuracy and efficiency of osteoporosis detection in an unspecified population by starting with the reference standard data used for judgment. There is an urgent need for a detection data model construction method and system that uses cross-values ​​of different segments as reference standards for judging osteoporosis according to multiple dimensions such as age, height, and gender. Summary of the invention

[0003] In view of the above technical problems, in order to further improve the detection accuracy and detection efficiency of osteoporosis in unspecified populations, the present invention provides a method for constructing an osteoporosis detection data model, including: step S1, setting age segments according to 5 years of age, setting body length segments according to 5 cm of height, and setting gender segments according to male and female classification. segment), group and sample the bone healthy population with these three dimensions, and produce the bone density medical image standard data group of the healthy population corresponding to the above three dimensions; step S2, group and sample the osteoporosis population with the same three dimensions, and produce the bone density medical image standard data group of the unhealthy population corresponding to the above three dimensions; step S3, extract image features from the bone density medical image standard data group of the healthy population and the bone density medical image standard data group of the unhealthy population under the same dimensions and calculate the bone density medical image feature difference of the unhealthy population under the dimensional combination; step S4, loop the above step S3 until the bone density medical image feature difference of the unhealthy population under all different dimensional combinations is completed, thereby obtaining the bone density medical image feature difference data group of the unhealthy population under the full dimensional combination, wherein the bone density medical image feature difference data group of the unhealthy population under the full dimensional combination at least includes the bone density medical image feature difference data group related to the upper limb joints and the bone density medical image feature difference data group related to the lower limb joints.

[0004] In addition, preferably, in the osteoporosis detection data model construction method of the present invention, the upper limb joint-related bone density medical image feature difference data group also includes a wrist-related bone density medical image feature difference data group, a finger-related bone density medical image feature difference data group, an elbow-related bone density medical image feature difference data group, and an arm-related bone density medical image feature difference data group; the lower limb joint-related bone density medical image feature difference data group also includes an ankle-related bone density medical image feature difference data group, a toe-related bone density medical image feature difference data group, a knee-related bone density medical image feature difference data group, and a thigh-related bone density medical image feature difference data group.

[0005] In addition, preferably, in the osteoporosis detection data model construction method of the present invention, in step S3, image features are extracted from a standard data group of bone density medical images of healthy people and a standard data group of bone density medical images of unhealthy people under the same dimension, and it also includes: pre-processing the bone density medical images in sequence by image digitization, image normalization and image enhancement; inputting the above-processed bone density medical image data into the feature extraction model to obtain a feature map, inputting the feature map into the capsule network model to perform feature recognition on the feature map, and obtaining the standard feature value for osteoporosis judgment.

[0006] In addition, preferably, in the osteoporosis detection data model construction method of the present invention, the capsule network model includes a first capsule layer and a second capsule layer, wherein, in the first capsule layer, the feature map is flattened into a one-dimensional vector, m one-dimensional vectors are defined, and each vector node is assigned a capsule; in the second capsule layer, the image prediction capsules formed after affine transformation of all one-dimensional vector features are routed iteratively to obtain an updated vector n as the osteoporosis detection result output.

[0007] In addition, preferably, in the osteoporosis detection data model construction method of the present invention, the affine transformation is the product of a one-dimensional vector and an affine matrix, and the result is the input vector of the routing iteration; the parameters of the affine matrix are updated by the one-dimensional vector and the vector n updated by the routing iteration through the back propagation algorithm; the routing iteration specifically includes: setting unnormalized weights, where a is the iteration round, b represents the xth of all one-dimensional vectors and the normalized weights are obtained by calculating the first function, the intermediate quantity is obtained by summing the formula, and the vector of normalized length is obtained by the second function, which is the output vector of the routing iteration; judging whether the iteration round a meets the conditions, if the iteration round a meets the conditions, the final output vector n is obtained; if the iteration round a does not meet the conditions, the weight of the next round is obtained by the formula, and one round of the routing iteration is completed, and then the iteration round a value is increased by one, and the next round of routing iteration is performed.

[0008] In addition, preferably, in the osteoporosis detection data model construction method of the present invention, the image digitization processing is to convert the bone density medical image into grayscale image data, and unify the grayscale image data into a grayscale image with a channel number of 1 and a length and width of 224 through an interpolation function; the image normalization processing is to divide the numerical value in the grayscale image after the image digitization processing by the mean of the grayscale image, and normalize the grayscale image; the image enhancement processing is to enhance the brightness of the grayscale image after normalization processing to enhance the bone density characteristics.

[0009] In addition, preferably, in the osteoporosis detection data model construction method of the present invention, the feature extraction model includes three two-dimensional convolution layers, and the three two-dimensional convolution layers are set; each convolution kernel performs a convolution operation on the bone density medical image in turn to obtain multiple eigenvalues, and all the eigenvalues ​​are spliced ​​according to the positions of the product accumulation to form a feature map.

[0010] In addition, preferably, the present invention also provides a system for constructing a data model for detecting osteoporosis, including: a module for forming a standard data group of medical images of bone density of healthy people, which sets age groups according to the equal interval of 5 years, sets height groups according to the equal interval of 5 cm, and sets gender groups according to the classification of men and women, and groups and samples people with healthy bones in these three dimensions to produce a standard data group of medical images of bone density of healthy people corresponding to the above three dimensions; a module for forming a standard data group of medical images of bone density of unhealthy people, which groups and samples people with osteoporosis in the same three dimensions to produce a standard data group of medical images of bone density of unhealthy people corresponding to the above three dimensions; a module for extracting feature difference of medical images of bone density of unhealthy people The control module comprises a block for extracting image features from a standard data group of bone density medical images of healthy people and a standard data group of bone density medical images of unhealthy people under the same dimension and calculating a difference value of bone density medical images of unhealthy people under the dimensional combination; a loop calculation control module for looping the above step S3 until the difference values ​​of bone density medical images of unhealthy people under all different dimensional combinations are completed, thereby obtaining a data group of bone density medical images of unhealthy people under a full dimensional combination, wherein the data group of bone density medical images of unhealthy people under the full dimensional combination at least includes a data group of bone density medical images of upper limb joints and a data group of bone density medical images of lower limb joints.

[0011] By utilizing the present invention, it is possible to cross-sample values ​​from different segments according to multiple dimensions such as age, height, and gender as reference standards for determining osteoporosis. Starting from the reference standard data used for determination, the accuracy of osteoporosis detection and the efficiency of detection operations for an unspecified population can be greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0013] Figure 1 is a flowchart illustrating a method for constructing an osteoporosis detection data model according to an embodiment of the present invention.

[0014] Figure 2 Detailed description of the invention is a block diagram showing a system for building a data model for osteoporosis detection according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Other embodiments or variations obtained based on the embodiments in the present application by ordinary technicians in this field without creative work are all within the scope of protection of this application.

[0016] Figure 1 FIG. 1 is a flow chart showing a method for constructing an osteoporosis detection data model according to an embodiment of the present invention. Figure 1As shown, a method for constructing an osteoporosis detection data model according to an embodiment of the present invention includes: step S1, setting an age segment according to 5 years of age, setting a height segment according to 5 cm of height, and setting a gender segment according to male and female classification. segment), group and sample the bone healthy population with these three dimensions, and produce the bone density medical image standard data group of the healthy population corresponding to the above three dimensions; step S2, group and sample the osteoporosis population with the same three dimensions, and produce the bone density medical image standard data group of the unhealthy population corresponding to the above three dimensions; step S3, extract image features from the bone density medical image standard data group of the healthy population and the bone density medical image standard data group of the unhealthy population under the same dimensions and calculate the bone density medical image feature difference of the unhealthy population under the dimensional combination; step S4, loop the above step S3 until the bone density medical image feature difference of the unhealthy population under all different dimensional combinations is completed, thereby obtaining the bone density medical image feature difference data group of the unhealthy population under the full dimensional combination, wherein the bone density medical image feature difference data group of the unhealthy population under the full dimensional combination at least includes the bone density medical image feature difference data group related to the upper limb joints and the bone density medical image feature difference data group related to the lower limb joints.

[0017] In addition, as another preferred embodiment of the present invention, in the osteoporosis detection data model construction method of the present invention, the upper limb joint-related bone density medical image feature difference data group also includes a wrist-related bone density medical image feature difference data group, a finger-related bone density medical image feature difference data group, an elbow-related bone density medical image feature difference data group, and an arm-related bone density medical image feature difference data group; the lower limb joint-related bone density medical image feature difference data group also includes an ankle-related bone density medical image feature difference data group, a toe-related bone density medical image feature difference data group, a knee-related bone density medical image feature difference data group, and a thigh-related bone density medical image feature difference data group.

[0018] In addition, as another preferred embodiment of the present invention, in the osteoporosis detection data model construction method of the present invention, in step S3, image features are extracted from a standard data group of bone density medical images of healthy people and a standard data group of bone density medical images of unhealthy people under the same dimension, and it also includes: pre-processing the bone density medical images in sequence by image digitization, image normalization and image enhancement; inputting the above-processed bone density medical image data into the feature extraction model to obtain a feature map, inputting the feature map into the capsule network model to perform feature recognition on the feature map, and obtaining the standard feature value for osteoporosis judgment.

[0019] In addition, as another preferred embodiment of the present invention, in the osteoporosis detection data model construction method of the present invention, the capsule network model includes a first capsule layer and a second capsule layer, wherein, in the first capsule layer, the feature map is flattened into a one-dimensional vector, m one-dimensional vectors are defined, and each vector node is assigned a capsule; in the second capsule layer, the image prediction capsules formed after affine transformation of all one-dimensional vector features are routed iteratively to obtain an updated vector n as the osteoporosis detection result output.

[0020] In addition, as another preferred embodiment of the present invention, in the osteoporosis detection data model construction method of the present invention, the affine transformation is the product of a one-dimensional vector and an affine matrix, and the result is the input vector of the routing iteration; the parameters of the affine matrix are updated by the one-dimensional vector and the vector n updated by the routing iteration through the back propagation algorithm; the routing iteration specifically includes: setting an unnormalized weight, where a is the iteration round, b represents the xth of all one-dimensional vectors and the normalized weight is obtained by calculating the first function, the intermediate quantity is obtained by summing the formula, and the vector of normalized length is obtained by the second function, which is the output vector of the routing iteration; judging whether the iteration round a meets the conditions, if the iteration round a meets the conditions, the final output vector n is obtained; if the iteration round a does not meet the conditions, the weight of the next round is obtained by the formula, and one round of the routing iteration is completed, and then the iteration round a value is increased by one, and the next round of routing iteration is performed.

[0021] In addition, as another preferred embodiment of the present invention, in the osteoporosis detection data model construction method of the present invention, the image digitization processing is to convert the bone density medical image into grayscale image data, and unify the grayscale image data into a grayscale image with a channel number of 1 and a length and width of 224 through an interpolation function; the image normalization processing is to divide the numerical value in the grayscale image after the image digitization processing by the mean of the grayscale image, and normalize the grayscale image; the image enhancement processing is to enhance the brightness of the grayscale image after normalization processing to enhance the bone density characteristics.

[0022] In addition, as another preferred embodiment of the present invention, in the osteoporosis detection data model construction method of the present invention, the feature extraction model includes three two-dimensional convolution layers, and the three two-dimensional convolution layers are set; each convolution kernel performs a convolution operation on the bone density medical image in turn to obtain multiple eigenvalues, and all the eigenvalues ​​are spliced ​​according to the positions of the product accumulation to form a feature map.

[0023] Figure 2 Detailed description of the invention is a block diagram showing a system for building a data model for osteoporosis detection according to an embodiment of the present invention.

[0024] In addition, as another preferred embodiment of the present invention, the present invention also provides an osteoporosis detection data model construction system, such as Figure 2 As shown, the osteoporosis detection data model construction system includes: a module for forming a standard data set of bone density medical images of healthy people, which sets age groups according to 5 years of age, sets height groups according to 5 cm of height, and sets gender groups according to male and female classification, and groups and samples people with healthy bones in these three dimensions to produce a standard data set of bone density medical images of healthy people corresponding to the above three dimensions; a module for forming a standard data set of bone density medical images of unhealthy people, which groups and samples people with osteoporosis in the same three dimensions to produce a standard data set of bone density medical images of unhealthy people corresponding to the above three dimensions; a module for extracting feature differences of bone density medical images of unhealthy people, which extracts feature differences of bone density medical images of unhealthy people from the same dimensions. The image features are extracted from the standard data group of bone density medical images of healthy people and the standard data group of bone density medical images of unhealthy people under the dimensional combination, and the bone density medical image feature difference of the unhealthy people under the dimensional combination is calculated; the loop calculation control module loops the above step S3 until the bone density medical image feature difference of the unhealthy people under all different dimensional combinations is completed, thereby obtaining the bone density medical image feature difference data group of the unhealthy people under the full dimensional combination, wherein the bone density medical image feature difference data group of the unhealthy people under the full dimensional combination at least includes the bone density medical image feature difference data group related to the upper limb joints and the bone density medical image feature difference data group related to the lower limb joints.

[0025] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0026] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0027] The above is only a specific embodiment of the present invention, but the protection scope of the present invention 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 by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

[0028] The above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for constructing an osteoporosis detection data model, characterized in that: include: Step S1, setting age groups by 5 years, setting height groups by 5 cm, and setting gender groups by gender, and grouping and sampling people with healthy bones based on these three dimensions to create a standard data set of bone density medical images of healthy people corresponding to the above three dimensions; Step S2, grouping and sampling the osteoporosis population using the same three dimensions to produce a standard data set of bone density medical images of unhealthy people corresponding to the above three dimensions; Step S3, extracting image features from a standard data set of bone density medical images of healthy people and a standard data set of bone density medical images of unhealthy people under the same dimension and calculating the difference of bone density medical image features of unhealthy people under the dimensional combination; Step S4, looping the above step S3 until the unhealthy population bone density medical image feature difference under all different dimensional combinations is completed, thereby obtaining the unhealthy population bone density medical image feature difference data set under all dimensional combinations, wherein: The bone density medical image feature difference data set of unhealthy people under the full-dimensional combination at least includes a bone density medical image feature difference data set related to upper limb joints and a bone density medical image feature difference data set related to lower limb joints.

2. The osteoporosis detection data model construction method according to claim 1, characterized in that: The upper limb joint-related bone density medical image feature difference data set also includes a wrist-related bone density medical image feature difference data set, a finger-related bone density medical image feature difference data set, an elbow-related bone density medical image feature difference data set, and an arm-related bone density medical image feature difference data set; The lower limb joint-related bone density medical image feature difference data group also includes an ankle-related bone density medical image feature difference data group, a toe-related bone density medical image feature difference data group, a knee-related bone density medical image feature difference data group, and a thigh-related bone density medical image feature difference data group.

3. The osteoporosis detection data model construction method according to claim 1, characterized in that: In step S3, image features are extracted from a standard medical image data set of bone density of healthy people and a standard medical image data set of bone density of unhealthy people in the same dimension, and the following steps are also included: The bone density medical image is processed in sequence in advance for image digitization, image normalization and image enhancement; The processed bone density medical image data is input into the feature extraction model to obtain a feature map, and the feature map is input into the capsule network model to perform feature recognition on the feature map to obtain a standard feature value for determining osteoporosis.

4. The osteoporosis detection data model construction method according to claim 1, characterized in that: The capsule network model includes a first capsule layer and a second capsule layer, wherein: In the first capsule layer, the feature map is flattened into a one-dimensional vector, m one-dimensional vectors are defined, and each vector node is assigned a capsule; in the second capsule layer, the image prediction capsules formed after affine transformation of all one-dimensional vector features are routed iteratively to obtain an updated vector n as the osteoporosis detection result output.

5. The osteoporosis detection data model construction method according to claim 4, characterized in that: The affine transformation is the product of a one-dimensional vector and an affine matrix, and the result is the input vector of the routing iteration; the parameters of the affine matrix are updated by the one-dimensional vector and the vector n updated by the routing iteration through a back propagation algorithm; The routing iteration specifically includes: setting an unnormalized weight, where a is an iteration round, b represents the xth of all one-dimensional vectors and obtaining a normalized weight by calculating a first function, using a formula to sum up to obtain an intermediate quantity, and using a second function to obtain a vector of normalized length, which is an output vector of the routing iteration; judging whether the iteration round a meets a condition, if the iteration round a meets the condition, then obtaining the final output vector n; if the iteration round a does not meet the condition, then obtaining the weight of the next round by using a formula, completing one round of the routing iteration, and then adding one to the iteration round a value to perform the next round of routing iteration.

6. The osteoporosis detection data model construction method according to claim 3, characterized in that: The image digitization processing is to convert the bone density medical image into grayscale image data, and unify the grayscale image data into a grayscale image with a channel number of 1 and a length and width of 224 through an interpolation function; The image normalization process is to divide the value in the grayscale image after the image is digitized by the mean value of the grayscale image to normalize the grayscale image; The image enhancement process is to perform brightness enhancement on the grayscale image after normalization process to strengthen the bone density feature.

7. The osteoporosis detection data model construction method according to claim 6, characterized in that: The feature extraction model includes three two-dimensional convolutional layers, Setting the three two-dimensional convolutional layers; Each convolution kernel sequentially performs a convolution operation on the bone density medical image to obtain a plurality of eigenvalues, and concatenates all eigenvalues ​​according to the positions of product accumulation to form a feature map.

8. A system for constructing an osteoporosis detection data model, characterized in that: include: The module for forming the standard data set of medical images of bone density of healthy people is to set the age layer according to the 5-year interval, the height layer according to the 5-cm interval, and the gender layer according to the male and female classification. The bone healthy people are grouped and sampled according to these three dimensions to produce the standard data set of medical images of bone density of healthy people corresponding to the above three dimensions; The module for forming the standard data set of bone density medical images of unhealthy people is used to group and sample the osteoporosis population using the same three dimensions to produce the standard data set of bone density medical images of unhealthy people corresponding to the above three dimensions; The module for extracting the difference value of bone density medical images of unhealthy people extracts the image features from the standard data set of bone density medical images of healthy people and the standard data set of bone density medical images of unhealthy people under the same dimension and calculates the difference value of bone density medical images of unhealthy people under the combination of the dimensions; The loop calculation control module loops through step S3 until the bone density medical image feature differences of unhealthy people under all different dimensional combinations are completed, thereby obtaining a data set of bone density medical image feature differences of unhealthy people under all dimensional combinations, wherein: The bone density medical image feature difference data set of unhealthy people under the full-dimensional combination at least includes a bone density medical image feature difference data set related to upper limb joints and a bone density medical image feature difference data set related to lower limb joints.

Images