Method and system for constructing bone aging degree prediction model, device and medium

Abstract

A method and system for constructing a bone aging degree prediction model, a device and a medium. The method comprises: acquiring CT images of target groups; dividing each CT image into a background region and regions of interest; on the basis of grayscale values, performing region subdivision on the regions of interest on each CT image, so as to obtain a plurality of subdivided regions in the regions of interest, and obtain the area proportion of each subdivided region in the regions of interest; and, on the basis of the CT images of the target groups and the area proportion of each subdivided region in the regions of interest on each CT image, and by means of a deep learning model framework, training to obtain a bone aging degree prediction model. The method can intelligently and more accurately predict the degrees of osteoporosis of people under test and whether they suffer from osteoporosis.

Description

A method, system, device, and medium for constructing a model to predict the degree of bone aging. Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to a method, system, device and medium for constructing a bone aging degree prediction model. Background Technology

[0002] Currently, the gold standard for diagnosing osteoporosis in clinical practice is bone mineral density (BMD) measured using dual-energy X-ray absorptiometry (DXA). This method involves passing two beams of X-rays of different energies simultaneously through human tissue and measuring the attenuation coefficient of the X-rays in the tissue based on the absorption characteristics of the tissue. However, studies show that the accuracy of inferring changes in bone strength from BMD values ​​obtained through DXA and other methods is only 60%–70%. BMD cannot be used to evaluate osteoporosis alone; it reflects the overall value of bone mineral content and does not reflect differences in bone geometry or structure. Furthermore, the heterogeneity of BMD measurements has a significant impact on the accurate assessment of osteoporosis. In other words, BMD's ability to evaluate and predict osteoporosis is limited.

[0003] Therefore, there is an urgent need to develop a model for predicting the degree of bone aging in order to assist in the early, rapid, and intelligent assessment of the degree of osteoporosis. Invention Overview

[0004] This invention provides a method, system, device, and medium for constructing a bone aging degree prediction model. Based on the labels of representative CT images of the target group and the area proportion of subdivided regions of interest on the CT images, the bone aging degree prediction model is trained to learn the relationship between the CT image features of the distal femur of the target group and the degree of osteoporosis related to aging. With one or a few representative CT images, the bone aging degree prediction model can intelligently and accurately assist in determining the degree of osteoporosis and whether the subject has osteoporosis. Technical solutions

[0005] This invention provides a method, system, device, and medium for constructing a model for predicting the degree of osteoporosis, in order to overcome the low accuracy of using dual-energy X-ray absorptiometry to measure bone mineral density to evaluate osteoporosis.

[0006] This invention provides a method for constructing a model to predict the degree of bone aging, comprising:

[0007] CT images of the target population were acquired. The target population included both osteoporotic and non-osteoporotic mammals. The CT images included multiple images of the distal femur of the target population, and each of the multiple CT images was labeled with the degree of osteoporosis and the age.

[0008] On each CT image, a background region and a region of interest are defined, where the region of interest includes the trabecular bone region and the cortical region;

[0009] Based on the grayscale value, the region of interest on each CT image is subdivided to obtain multiple subdivided regions within the region of interest, and the area ratio of each subdivided region within the region of interest is obtained.

[0010] Based on the CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, a bone aging prediction model is trained using a deep learning model framework.

[0011] In one embodiment, the mammals in the mammalian group can be animals selected from bovids, equines, felines, canines, lagos, suidae, camels, rodents, and primates, including but not limited to cattle, horses, goats, sheep, cats, rabbits, pigs, camels, alpacas, rats, mice, guinea pigs, non-human primates (such as apes, monkeys, baboons, and orangutans), and humans, preferably cattle, horses, dogs, goats, sheep, pigs, camels, rats, mice, monkeys, and humans. Preferably, the mammals are laboratory animals, including but not limited to mice, rats, rabbits, guinea pigs, hamsters, monkeys, dogs, cats, pigs, sheep, and horses.

[0012] In one implementation, acquiring CT images of the target group includes:

[0013] Using the CT image of the femoral growth plate of the target group as layer 0, CT images are acquired in the direction of the femoral shaft of the target group to obtain multiple initial CT images. The final CT image is obtained by filtering the multiple initial CT images at a preset layer interval.

[0014] For example, using the CT image of the femoral growth plate of the target group as the first layer, CT images are acquired in the direction of the leg shaft of the target group to obtain 300 initial CT images (one per layer). Then, CT images are filtered from the 300 initial CT images at a preset layer interval of 50 layers to obtain 6 final CT images.

[0015] This approach reduces data processing time while maintaining predictive accuracy, and allows for the selection of representative CT images to study bone aging, which has practical clinical significance.

[0016] In one implementation, the step of subdividing the region of interest (ROI) on each CT image based on grayscale values ​​to obtain multiple subdivided regions within the ROI, and obtaining the area proportion of each subdivided region within the ROI, includes:

[0017] The pixel density of the region of interest in each CT image is compared with multiple preset grayscale ranges, and the region of interest in each CT image is subdivided to obtain the first subdivided region, the second subdivided region, the third subdivided region, and the fourth subdivided region within the region of interest.

[0018] Divide the pixel density of each sub-region of each CT image by the total pixel density of all sub-regions of the CT image to obtain the area ratio of each sub-region in the region of interest.

[0019] In one implementation, the deep learning model framework includes multiple convolutional neural networks and multiple fully connected networks. The bone aging degree prediction model is trained based on the CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, using the deep learning model framework. This includes:

[0020] Based on the CT images of the target group and the area ratio of each sub-region in the region of interest of each CT image, multi-dimensional feature extraction is performed through multiple convolutional neural networks to obtain multiple features. These features are then flattened and stacked into a one-dimensional feature vector, and an activation function is used to perform non-linear activation on the one-dimensional feature vector.

[0021] By using multiple fully connected networks to perform global analysis on nonlinearly activated one-dimensional feature vectors, the mapping relationship between each CT image and the degree of osteoporosis in the target population during growth is learned, so as to train a model for predicting the degree of bone aging.

[0022] Furthermore, the multiple features obtained through multi-dimensional feature extraction using multiple convolutional neural networks can be understood as feature correlations between multiple sub-regions (such as quantitative features, qualitative features, spatial configuration features, etc.). This is a characteristic of statistics; they are mutually constraining and restrictive, not a single factor, but rather an expression of the correlation features of multiple sub-regions. For example, an individual in the target group has 6 CT images, each with four sub-regions, thus yielding 4*6=24 interdependent feature sets.

[0023] In one implementation, the step of training a bone aging prediction model based on a deep learning model framework, using CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, includes:

[0024] During model training, the predicted osteoporosis level and osteoporosis status of the target group are compared with the actual osteoporosis level and osteoporosis status of the target group using a loss function. The difference between the predicted and actual values ​​is then reduced and the model parameters of multiple convolutional neural networks are updated.

[0025] In one implementation, the deep learning model framework includes four convolutional neural networks and two fully connected networks, with the activation function being the ReLU function and the loss function being the BCEWithLogitsLoss function. The convolutional kernel sizes of the four convolutional neural networks are all different.

[0026] This invention also provides a system for constructing a model for predicting the degree of bone aging, comprising:

[0027] The image acquisition module is used to acquire CT images of the target group, which includes a mammalian group with osteoporosis and a mammalian group without osteoporosis. The CT images include multiple CT images of the distal femur of the target group, and each of the multiple CT images is labeled with an osteoporosis degree label and an age label.

[0028] The region segmentation module is used to: segment the background region and the region of interest on each CT image, wherein the region of interest includes the trabecular bone region and the cortical region;

[0029] The region subdivision module is used to: subdivide the region of interest on each CT image according to the gray value, obtain multiple subdivision regions within the region of interest, and obtain the area ratio of each subdivision region within the region of interest.

[0030] The model training module is used to train a bone aging prediction model based on a deep learning model framework, using CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image.

[0031] The present invention also provides an osteoporosis severity prediction system, comprising:

[0032] The image receiving module is used to receive CT images and age data of the subject, wherein the subject is a mammal whose osteoporosis is to be assessed, and the type of the subject is consistent with the type of the target group in any of the above-described methods for constructing a bone aging degree prediction model, and the CT image is a CT image located at the distal femur of the subject.

[0033] The prediction module is used to: obtain the degree of osteoporosis of the test subject based on the CT images and age data of the test subject, through the bone aging degree prediction model obtained by any of the above-described methods for constructing the bone aging degree prediction model;

[0034] The determination module is used to: determine whether the subject has osteoporosis based on the subject's CT images and age data, using the bone aging degree prediction model obtained through any of the above-described methods for constructing the bone aging degree prediction model.

[0035] The present invention also provides an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the computer program to implement the method for constructing the bone aging degree prediction model described above.

[0036] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing the bone aging degree prediction model described above.

[0037] The present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute any of the above-described methods for constructing a bone aging degree prediction model. Beneficial effects

[0038] This invention provides a method, system, device, and medium for constructing a bone aging degree prediction model. Based on the labels of representative CT images of the target group and the area proportion of subdivided regions of interest on the CT images, the bone aging degree prediction model is trained to learn the relationship between the CT image features of the distal femur of the target group and the degree of osteoporosis related to aging. With one or a few representative CT images, the bone aging degree prediction model can intelligently and accurately assist in determining the degree of osteoporosis and whether the subject has osteoporosis. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 is a flowchart illustrating the construction method of a bone aging degree prediction model provided by the present invention.

[0041] Figure 2 is a schematic diagram of the structure of a bone aging degree prediction model provided by the present invention.

[0042] Figure 3 is a schematic diagram of the structure of an osteoporosis degree prediction system provided by the present invention.

[0043] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention. Embodiments of the present invention

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0045] The following describes the construction method, system, equipment, and medium of the bone aging degree prediction model provided by the present invention with reference to Figures 1-4.

[0046] Figure 1 is a flowchart illustrating the method for constructing a bone aging degree prediction model provided by the present invention. Referring to Figure 1, the method for constructing a bone aging degree prediction model provided by the present invention may include:

[0047] Step S110: Obtain CT images of the target group, wherein the target group includes a mammalian group with osteoporosis and a mammalian group without osteoporosis. The CT images include multiple CT images located at the distal femur of the target group, and each of the multiple CT images is labeled with an osteoporosis degree label and an age label.

[0048] Step S120: Divide the background region and region of interest on each CT image, wherein the region of interest includes the trabecular bone region and the cortical region;

[0049] Step S130: Based on the grayscale value, subdivide the region of interest on each CT image to obtain multiple subdivided regions within the region of interest, and obtain the area ratio of each subdivided region within the region of interest.

[0050] Step S140: Based on the CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, a bone aging degree prediction model is trained using a deep learning model framework.

[0051] In one embodiment, the mammals in the mammalian group can be animals selected from bovids, equines, felines, canines, lagos, suidae, camels, rodents, and primates, including but not limited to cattle, horses, goats, sheep, cats, rabbits, pigs, camels, alpacas, rats, mice, guinea pigs, non-human primates (such as apes, monkeys, baboons, and orangutans), and humans, preferably cattle, horses, dogs, goats, sheep, pigs, camels, rats, mice, monkeys, and humans. Preferably, the mammals are laboratory animals, including but not limited to mice, rats, rabbits, guinea pigs, hamsters, monkeys, dogs, cats, pigs, sheep, and horses.

[0052] In one embodiment, step S110 may use the CT image of the leg growth plate of the target group as the first layer, acquire CT images in the direction of the leg shaft segment of the target group to obtain multiple initial CT images, and filter the CT images from the multiple initial CT images at a preset layer interval to obtain the final CT image.

[0053] For example, a single scan of the femur of each rat can generate 1500 CT images. Since human errors such as tilting can occur during CT image capture, image processing tools can be used to rotate the CT images to a positive orientation first. Experiments have shown that for rats of different ages, the rate of change in trabecular bone area in the diaphysis is relatively small, while the rate of change in trabecular bone area in the growth plate is relatively large. Therefore, the CT images of the growth plate can be used as the first layer, focusing on studying 300 CT images along the diaphysis direction. ).

[0054] Meanwhile, the experiment found that the CT images of adjacent layers had a high degree of similarity. Therefore, in this embodiment, one image was selected from every 50 layers, resulting in a total of 6 CT images per rat. This approach reduces the time cost of data processing while maintaining accuracy, and allows for the selection of representative CT images to assess bone aging, which has practical clinical significance.

[0055] The osteoporosis level label and age label on CT images can be pre-labeled manually. The osteoporosis level label describes the degree of deviation between the bone strength shown in the CT image and the baseline bone strength for the corresponding age, thus indicating the degree of osteoporosis of the individual in the CT image.

[0056] A CT image typically contains a region of interest (ROI) and a background region. This embodiment primarily studies the relationship between the area of ​​trabecular and cortical regions and the degree of bone aging. Therefore, step S120 can designate the region in the CT image including trabecular and cortical regions as the ROI, and the remaining regions as the background region. To eliminate the influence of the background region on the model training results, mimics software can be used to separate the ROI from the background region. The shaded area represents the ROI, and the other areas are the background region. This embodiment studies CT images of six layers (… All regions are divided into regions of interest, resulting in 6 layers of regions of interest, denoted as ( ).

[0057] In one embodiment, step S130 can compare the pixel density of the region of interest in each CT image with multiple preset grayscale ranges, subdivide the region of interest in each CT image to obtain a first subdivision region, a second subdivision region, a third subdivision region, and a fourth subdivision region within the region of interest, and then divide the pixel density of each subdivision region of each CT image by the total pixel density of all subdivision regions of the CT image to obtain the area ratio of each subdivision region in the region of interest.

[0058] This embodiment defines multiple preset grayscale ranges, including: image areas with grayscale ranges in [0, 50] are red areas, image areas with grayscale ranges in (50, 100] are yellow areas, image areas with grayscale ranges in (100, 150] are blue areas, and image areas with grayscale ranges in (150, 255] are green areas. By comparing the pixel density of the region of interest in the CT image with multiple preset grayscale ranges, the first subdivision region, the second subdivision region, the third subdivision region, and the fourth subdivision region within the region of interest can be obtained.

[0059] Specifically, will Defined as the CT image to be processed, the corresponding region of interest is The length of the image is defined as `width`, and the width of the image is defined as `height`. PixelRed, PixelYellow, PixelBlue, and PixelGreen are defined to represent the number of pixels in four sub-regions, with initial values ​​set to 0. Through computer image processing techniques, the first, second, third, and fourth sub-regions within the region of interest can be obtained, and the pixels in each sub-region can be labeled with their corresponding colors.

[0060] Because bone area varies significantly among individuals, the area occupied by different sub-regions also varies greatly. Therefore, in this embodiment, the pixel density of each sub-region in each CT image is divided by the total pixel density of all sub-regions in the CT image to normalize the sub-regions and obtain the area proportion of each sub-region in the region of interest, in the range [0, -1].

[0061] For each individual in the target group, this embodiment will obtain the area proportion of the four sub-regions after normalization corresponding to the six levels ( Among them, AreaRed, AreaYellow, AreaBlue, and AreaGreen represent the red, yellow, blue, and green regions within the region of interest (the proportion of the area of ​​the first, second, third, and fourth sub-regions to the total area of ​​the region of interest), respectively. This part of the output data will be used as the input for the model training task.

[0062] In one embodiment, the deep learning model framework includes multiple convolutional neural networks and multiple fully connected networks, and step S140 may include:

[0063] Based on the CT images of the target group and the area ratio of each sub-region in the region of interest of each CT image, multi-dimensional feature extraction is performed through multiple convolutional neural networks to obtain multiple features. These features are then flattened and stacked into a one-dimensional feature vector, and an activation function is used to perform non-linear activation on the one-dimensional feature vector.

[0064] By using multiple fully connected networks to perform global analysis on nonlinearly activated one-dimensional feature vectors, the mapping relationship between each CT image and the degree of osteoporosis in the target population during growth is learned, so as to train a model for predicting the degree of bone aging.

[0065] During model training, the predicted osteoporosis level and osteoporosis status of the target group are compared with the actual osteoporosis level and osteoporosis status of the target group using a loss function. The difference between the predicted and actual values ​​is then reduced and the model parameters of multiple convolutional neural networks are updated.

[0066] The bone aging prediction model in this embodiment mainly includes a convolutional neural network, an activation function, a fully connected network, and a loss function.

[0067] A convolutional neural network (CNN) is a typical example of a deep neural network model. This network uses a fixed-size convolutional kernel to slide regularly across the input data with a defined stride, thereby effectively extracting features from the current input.

[0068] In this embodiment, the kernel size is and the stride is 1. The calculation formula for the convolutional neural network is:

[0069] ,

[0070] This represents the input to the convolutional neural network. This represents the output of the convolutional neural network. This represents the parameters of a convolutional neural network, which are continuously updated during the model's learning process.

[0071] A fully connected layer, located at the end of a deep learning model, is primarily used for global analysis of the input features. Its main function is to transform the output of the convolutional neural network into the final classification or regression result, or to convert features into a fixed shape. The computational expression for a fully connected layer is:

[0072] ,

[0073] Assumption This represents the input of a fully connected network. This represents the output of a fully connected network. Assume... The shape is The shape is Then the shape of the parameter w of the fully connected network needs to be... The shape of parameter b needs to be The parameters w and b will be updated during the model learning process.

[0074] Specifically, referring to Figure 2, in this embodiment, to increase the richness of features, the deep learning model framework includes four convolutional neural networks (with different kernel sizes, strides, etc.) for multi-dimensional feature extraction from the input data and two fully connected networks for fusing the extracted features. The activation function is the ReLU function, and the loss function is the BCEWithLogitsLoss function.

[0075] In this embodiment, the expression for the bone aging degree prediction model is defined as follows:

[0076] in ,

[0077]

[0078] Y represents the input data; Model represents the deep learning model, which is a set of mathematical expressions; Y is the output of the model, which is the prediction of the degree of osteoporosis and whether the mouse sample has osteoporosis after the deep learning model extracts and identifies features from CT images.

[0079] This embodiment input (Indicating that X belongs to a 6×4 matrix) will pass through these four volumes respectively.

[0080] The integral neural network performs calculations to obtain output features.

[0081]

[0082] Then, the four output features are flattened and stacked into a one-dimensional feature vector:

[0083] ,

[0084] Then, the ReLU function is used to perform non-linear activation on the output features:

[0085] ,

[0086] Finally, two fully connected layers are used to map the output features to the output layer:

[0087] ,

[0088] in This indicates the output of the model.

[0089] In this embodiment, the train_test_split() function from the sklearn library is used to randomly select 60% of the samples as the training set and the remaining 40% as the test set.

[0090] During the training phase, only the training set data is used. Each training batch consists of 8 data points. The Adam optimizer is used, with an initial learning rate of 0.01. The training epochs are 500. BCEWithLogitsLoss is used as the loss function to supervise and optimize the model's learning process, and is used to calculate the model's predicted output. With real labels The distance between them is defined as follows:

[0091]

[0092] in This represents the Sigmoid activation function.

[0093] During the testing phase, only the test set data was used. Since the training and test sets were randomly selected, to reduce error, the experiment was repeated 20 times, and the accuracy of the deep learning model on the test set was recorded in these 20 experiments. The accuracy (P) was defined as follows:

[0094]

[0095] Where TP represents the number of samples correctly predicted by the model, and FP represents the number of samples incorrectly predicted by the model.

[0096] This embodiment collected 80 rat leg samples at ages of 4, 8, 12, and 24 weeks. The normal group represented naturally growing rats, while the experimental group represented ovariectomized rats (simulating osteoporosis). CT images were used for model training and prediction. The bone aging prediction model constructed in this invention can accurately learn the age-related osteoporosis patterns at 4, 8, 12, and 24 weeks, and can more effectively and accurately predict the early degree of osteoporosis and whether the subjects have osteoporosis.

[0097] The method for constructing a bone aging degree prediction model provided by this invention trains the bone aging degree prediction model to learn the relationship between the CT image features of the leg shaft segment of the target group and the degree of osteoporosis related to aging during the development of the target group, based on the labels of specific representative CT images of the target group and the area proportion of the subdivided regions of interest on the CT images. With one or a few representative CT images, the bone aging degree prediction model can intelligently and accurately assist in judging the degree of osteoporosis and whether the subject has osteoporosis.

[0098] The following describes the construction system of the bone aging degree prediction model provided by the present invention. The construction system of the bone aging degree prediction model described below can be referred to in correspondence with the construction method of the bone aging degree prediction model described above.

[0099] The present invention provides a system for constructing a model for predicting bone aging, which may include:

[0100] The image acquisition module is used to: acquire CT images of the target group, wherein the target group includes a mammalian group with osteoporosis and a mammalian group without osteoporosis. The CT images include multiple CT images located at the distal femur of the target group, and each of the multiple CT images is labeled with an osteoporosis degree label and an age label.

[0101] The region segmentation module is used to: segment the background region and the region of interest on each CT image, wherein the region of interest includes the trabecular bone region and the cortical region;

[0102] The region subdivision module is used to: subdivide the region of interest on each CT image according to the gray value, obtain multiple subdivision regions within the region of interest, and obtain the area ratio of each subdivision region within the region of interest.

[0103] The model training module is used to train a bone aging prediction model based on a deep learning model framework, using CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image.

[0104] Referring to Figure 3, the present invention also provides an osteoporosis severity prediction system, comprising:

[0105] The image receiving module is used to receive CT images and age data of the subject, wherein the subject is a mammal whose osteoporosis is to be assessed, and the type of the subject is consistent with the type of the target group in any of the above-described methods for constructing a bone aging degree prediction model, and the CT image is a CT image located at the distal femur of the subject.

[0106] The prediction module is used to: obtain the degree of osteoporosis of the test subject based on the CT images and age data of the test subject, through the bone aging degree prediction model obtained by any of the above-described methods for constructing the bone aging degree prediction model;

[0107] The determination module is used to: determine whether the subject has osteoporosis based on the subject's CT images and age data, using the bone aging degree prediction model obtained through any of the above-described methods for constructing the bone aging degree prediction model.

[0108] In one embodiment, the mammals in the mammalian group can be animals selected from bovids, equines, felines, canines, lagos, suidae, camels, rodents, and primates, including but not limited to cattle, horses, goats, sheep, cats, rabbits, pigs, camels, alpacas, rats, mice, guinea pigs, non-human primates (such as apes, monkeys, baboons, and orangutans), and humans, preferably cattle, horses, dogs, goats, sheep, pigs, camels, rats, mice, monkeys, and humans. Preferably, the mammals are laboratory animals, including but not limited to mice, rats, rabbits, guinea pigs, hamsters, monkeys, dogs, cats, pigs, sheep, and horses.

[0109] Figure 4 illustrates a schematic diagram of the physical structure of an electronic device. As shown in Figure 3, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. The processor 810, communication interface 820, and memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a method for constructing a bone aging degree prediction model. This method includes:

[0110] CT images of the target population were acquired. The target population included both osteoporotic and non-osteoporotic mammals. The CT images included multiple images of the distal femur of the target population, and each of the multiple CT images was labeled with the degree of osteoporosis and the age.

[0111] On each CT image, a background region and a region of interest are defined, where the region of interest includes the trabecular bone region and the cortical region;

[0112] Based on the grayscale value, the region of interest on each CT image is subdivided to obtain multiple subdivided regions within the region of interest, and the area ratio of each subdivided region within the region of interest is obtained.

[0113] Based on the CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, a bone aging prediction model is trained using a deep learning model framework.

[0114] And / or, when the processor 810 invokes a logical instruction in memory 830, it also performs the following steps:

[0115] The test subject receives CT images and age data, wherein the test subject is a mammal whose osteoporosis is to be assessed, and the type of the test subject is consistent with the type of the target population in any of the above-described methods for constructing a bone aging degree prediction model. The CT images are CT images located at the distal femur of the test subject.

[0116] Based on the CT images and age data of the subject, the degree of osteoporosis of the subject is obtained by constructing a bone aging degree prediction model using any of the above-described methods.

[0117] The test determines whether the subject has osteoporosis based on the degree of osteoporosis.

[0118] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0119] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is capable of executing the method for constructing the bone aging degree prediction model provided by the above methods, the method comprising:

[0120] CT images of the target population were acquired. The target population included both osteoporotic and non-osteoporotic mammals. The CT images included multiple images of the distal femur of the target population, and each of the multiple CT images was labeled with the degree of osteoporosis and the age.

[0121] On each CT image, a background region and a region of interest are defined, where the region of interest includes the trabecular bone region and the cortical region;

[0122] Based on the grayscale value, the region of interest on each CT image is subdivided to obtain multiple subdivided regions within the region of interest, and the area ratio of each subdivided region within the region of interest is obtained.

[0123] Based on the CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, a bone aging prediction model is trained using a deep learning model framework.

[0124] And / or, when a computer program is executed by a processor, the computer is also able to perform the following steps:

[0125] The test subject receives CT images and age data, where the test subject is an experimental mouse, and the CT images are CT images located at the distal femur of the test subject.

[0126] Based on the CT images and age data of the subject, the degree of osteoporosis of the subject is obtained by constructing a bone aging degree prediction model using any of the above-described methods.

[0127] The test determines whether the subject has osteoporosis based on the degree of osteoporosis.

[0128] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for constructing a bone aging degree prediction model provided by the methods described above, the method comprising:

[0129] CT images of the target population were acquired. The target population included both osteoporotic and non-osteoporotic mammals. The CT images included multiple images of the distal femur of the target population, and each of the multiple CT images was labeled with the degree of osteoporosis and the age.

[0130] On each CT image, a background region and a region of interest are defined, where the region of interest includes the trabecular bone region and the cortical region;

[0131] Based on the grayscale value, the region of interest on each CT image is subdivided to obtain multiple subdivided regions within the region of interest, and the area ratio of each subdivided region within the region of interest is obtained.

[0132] Based on the CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, a bone aging prediction model is trained using a deep learning model framework.

[0133] And / or, when a computer program is executed by a processor, it also performs the following steps:

[0134] The test subject receives CT images and age data, wherein the test subject is a mammal whose osteoporosis is to be assessed, and the type of the test subject is consistent with the type of the target population in any of the above-described methods for constructing a bone aging degree prediction model. The CT images are CT images located at the distal femur of the test subject.

[0135] Based on the CT images and age data of the subject, the degree of osteoporosis of the subject is obtained by constructing a bone aging degree prediction model using any of the above-described methods.

[0136] The test determines whether the subject has osteoporosis based on the degree of osteoporosis.

[0137] The present invention also includes the following additional embodiments:

[0138] 1. A method for constructing a computer-implemented deep learning model for predicting bone aging, comprising:

[0139] CT images of the target population were acquired. The target population included both osteoporotic and non-osteoporotic mammals. The CT images included multiple images of the distal femur of the target population, and each of the multiple CT images was labeled with the degree of osteoporosis and the age.

[0140] On each CT image, a background region and a region of interest are defined, where the region of interest includes the trabecular bone region and the cortical region;

[0141] Based on the grayscale value, the region of interest on each CT image is subdivided to obtain multiple subdivided regions within the region of interest, and the area ratio of each subdivided region within the region of interest is obtained.

[0142] Based on the CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, combined with the osteoporosis level, age and age label of the target group, a deep learning model for predicting bone aging is trained based on a deep learning model framework.

[0143] 2. According to the construction method described in Implementation Scheme 1, acquiring CT images of the target group includes:

[0144] Using the CT image of the femoral growth plate of the target group as the first layer, CT images are acquired in the direction of the femoral shaft of the target group to obtain multiple initial CT images. Then, CT images are selected from the multiple initial CT images at a preset layer interval to obtain the training CT images.

[0145] 3. According to the construction method described in Implementation Scheme 2, the step of subdividing the region of interest (ROI) on each CT image based on grayscale values ​​to obtain multiple subdivided regions within the ROI, and obtaining the area proportion of each subdivided region within the ROI, includes:

[0146] The pixel density of the region of interest in each training CT image is compared with multiple preset grayscale ranges. The region of interest in each training CT image is subdivided to obtain the first subdivided region, the second subdivided region, the third subdivided region, and the fourth subdivided region within the region of interest.

[0147] Divide the pixel density of each sub-region of each training CT image by the total pixel density of all sub-regions of the training CT image to obtain the area ratio of each sub-region in the region of interest.

[0148] 4. According to the construction method described in Implementation Scheme 3, the deep learning model framework includes multiple convolutional neural networks and multiple fully connected networks. Based on the CT images of the target group and the area ratio of each sub-region in the region of interest on each CT image, combined with the osteoporosis level, age, and age label of the target group, a deep learning model for predicting bone aging is trained using the deep learning model framework, including:

[0149] Based on the training CT images of the target group and the area ratio of each sub-region in the region of interest of each training CT image, multi-dimensional feature extraction is performed through multiple convolutional neural networks to obtain multiple features. The multiple features are flattened and stacked into a one-dimensional feature vector, and the one-dimensional feature vector is non-linearly activated using an activation function.

[0150] By using multiple fully connected networks to perform global analysis on nonlinearly activated one-dimensional feature vectors, the mapping relationship between each training CT image and the degree of osteoporosis in the target population during growth is learned, so as to train a deep learning model for predicting bone aging.

[0151] 5. According to the construction method described in any one of Implementation Schemes 1-4, the step of training a deep learning model for predicting bone aging based on the CT images of the target group and the area ratio of each sub-region in the region of interest of each CT image, combined with the osteoporosis degree, age, and age label of the target group, and based on a deep learning model framework, includes:

[0152] During model training, the predicted osteoporosis level and osteoporosis status of individuals in a target group at a specific age stage are compared with the actual osteoporosis level and osteoporosis status of individuals in the target group at the specific age stage using a loss function. This process reduces the difference between the predicted and actual values ​​and updates the model parameters of multiple convolutional neural networks.

[0153] 6. According to the construction method described in any one of the implementation schemes 1-4, the deep learning model framework for predicting bone aging includes 4 convolutional neural networks and 2 fully connected networks, with the activation function being the ReLU function and the loss function being the BCEWithLogitsLoss function, wherein the convolution kernel sizes of the 4 convolutional neural networks are all different.

[0154] 7. A method for predicting bone aging, comprising:

[0155] Receive CT images and age data of the subject, wherein the type of the subject is consistent with the type of the target group in the construction method according to any one of embodiments 1-6, and the CT images include CT images located at the distal femur of the subject;

[0156] Based on the CT images and age data of the subject, the deep learning model for predicting bone aging, obtained by the construction method according to any one of the embodiments 1-6, is input to obtain the prediction results of whether the subject has osteoporosis and the degree of osteoporosis, wherein the degree of osteoporosis reflects whether there is a deviation between the bone strength shown in the CT image and the baseline bone strength of the corresponding age stage.

[0157] Based on the predicted osteoporosis level of the test subject, it is determined whether the test subject has bone aging.

[0158] 8. A bone aging prediction system, comprising:

[0159] One or more processors;

[0160] One or more non-transitory computer-readable media storing: a deep learning model for predicting bone aging, obtained according to the construction method of any one of embodiments 1-6;

[0161] Instructions that, when executed by the one or more processors, cause the computing system to perform operations, the operations including:

[0162] Receive CT images and age data of the subject, wherein the type of the subject is consistent with the type of the target group in the construction method according to any one of embodiments 1-6, and the CT images include CT images located at the distal femur of the subject;

[0163] The CT images and age data of the subject are input into the deep learning model for predicting bone aging to obtain the prediction results of whether the subject has osteoporosis and the degree of osteoporosis, wherein the degree of osteoporosis reflects whether the bone strength shown in the CT image deviates from the baseline bone strength of the corresponding age group.

[0164] Based on the predicted osteoporosis level of the test subject, it is determined whether the test subject has bone aging.

[0165] 9. A non-transitory computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0166] Receive CT images and age data of the subject, wherein the type of the subject is consistent with the type of the target group in the construction method according to any one of embodiments 1-6, and the CT images include CT images located at the distal femur of the subject;

[0167] The CT images and age data of the subject are input into the deep learning model for predicting bone aging to obtain the prediction results of whether the subject has osteoporosis and the degree of osteoporosis, wherein the degree of osteoporosis reflects whether the bone strength shown in the CT image deviates from the baseline bone strength of the corresponding age group.

[0168] Based on the predicted osteoporosis level of the test subject, it is determined whether the test subject has bone aging.

[0169] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0170] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Cross-references to related applications

[0172] This application claims priority to Chinese patent application No. 202410785603.0, filed on June 18, 2024, the entire contents of which are incorporated herein by reference. Industrial applicability

[0173] This invention provides a method, system, device, and medium for constructing a bone aging degree prediction model. Based on the labels of representative CT images of the target group and the area proportion of subdivided regions of interest on the CT images, the bone aging degree prediction model is trained to learn the relationship between the CT image features of the distal femur of the target group and the degree of osteoporosis related to aging. With one or a few representative CT images, the bone aging degree prediction model can intelligently and accurately assist in determining the degree of osteoporosis and whether the subject has osteoporosis.

Claims

1. A method of constructing a bone aging degree prediction model, characterized by, The method comprises the following steps: obtaining CT images of a target group, wherein the target group comprises a group of mammals with osteoporosis and a group of mammals without osteoporosis, the CT images comprise a plurality of CT images located at the distal end of the femur of the target group, and each of the plurality of CT images is labeled with an osteoporosis degree label and an age label; dividing a background region and a region of interest on each CT image, wherein the region of interest comprises a trabecular region and a cortical region; regionally subdividing the region of interest on each CT image according to a gray value, obtaining a plurality of subdivided regions within the region of interest, and obtaining an area proportion of each subdivided region in the region of interest; training a bone aging degree prediction model based on a deep learning model framework according to the CT images of the target group and the area proportion of each subdivided region in the region of interest on each CT image, and combining the osteoporosis degree age and the age label of the target group.

2. The method of claim 1, wherein the method is characterized by: The method comprises the following steps: obtaining CT images of a target group, wherein the target group comprises a group of mammals with osteoporosis and a group of mammals without osteoporosis, the CT images comprise a plurality of CT images located at the distal end of the femur of the target group, and each of the plurality of CT images is labeled with an osteoporosis degree label and an age label; 3. The method of claim 2, wherein the method further comprises: determining the degree of bone aging of the subject based on the bone aging degree prediction model. dividing a background region and a region of interest on each CT image, wherein the region of interest comprises a trabecular region and a cortical region; regionally subdividing the region of interest on each CT image according to a gray value, obtaining a plurality of subdivided regions within the region of interest, and obtaining an area proportion of each subdivided region in the region of interest; training a bone aging degree prediction model based on a deep learning model framework according to the CT images of the target group and the area proportion of each subdivided region in the region of interest on each CT image, and combining the osteoporosis degree age and the age label of the target group.

4. The method of claim 3, wherein the method is characterized by: The method comprises the following steps: obtaining CT images of a target group, wherein the target group comprises a group of mammals with osteoporosis and a group of mammals without osteoporosis, the CT images comprise a plurality of CT images located at the distal end of the femur of the target group, and each of the plurality of CT images is labeled with an osteoporosis degree label and an age label; dividing a background region and a region of interest on each CT image, wherein the region of interest comprises a trabecular region and a cortical region; regionally subdividing the region of interest on each CT image according to a gray value, obtaining a plurality of subdivided regions within the region of interest, and obtaining an area proportion of each subdivided region in the region of interest; training a bone aging degree prediction model based on a deep learning model framework according to the CT images of the target group and the area proportion of each subdivided region in the region of interest on each CT image, and combining the osteoporosis degree age and the age label of the target group. The deep learning model framework comprises a plurality of convolutional neural networks and a plurality of fully connected networks, and the training of the bone aging degree prediction model based on the deep learning model framework according to the CT images of the target group and the area proportion of each subdivided region in the region of interest on each CT image, and combining the osteoporosis degree age and the age label of the target group comprises the following steps: performing multi-dimensional feature extraction on the training CT images of the target group and the area proportion of each subdivided region in the region of interest on each training CT image through the plurality of convolutional neural networks to obtain a plurality of features, flattening and stacking the plurality of features into a one-dimensional feature vector, and performing nonlinear activation on the one-dimensional feature vector by using an activation function; performing global analysis on the one-dimensional feature vector subjected to nonlinear activation by using the plurality of fully connected networks, learning the mapping relationship between each training CT image and the osteoporosis degree during the growth of the target group, and training the bone aging degree prediction model.

5. The method of constructing a bone aging degree prediction model according to any one of claims 1 to 4, characterized in that, The bone aging degree prediction model is trained based on a deep learning model framework according to CT images of the target group and area proportions of each sub-region on each CT image in the area of interest, in combination with the osteoporosis degree age and age label of the target group. In the model training process, the prediction results of the osteoporosis degree and whether the target group individual at a specific age stage is osteoporotic are predicted according to the bone aging degree prediction model, and the prediction results are compared with the actual results of the osteoporosis degree and whether the target group individual at the specific age stage is osteoporotic, and the difference between the prediction value and the actual value is reduced by a loss function and the model parameters of the plurality of convolutional neural networks are updated.

6. The method of constructing a bone aging degree prediction model according to any one of claims 1 to 4, characterized in that, The deep learning model framework of the bone aging degree prediction model comprises four convolutional neural networks and two fully connected networks, the activation function is a ReLU function, and the loss function is a BCEWithLogitsLoss function, wherein the convolution kernel sizes of the four convolutional neural networks are all different.

7. A system for constructing a model to predict the degree of bone aging, characterized in that, The method comprises the following steps: An image acquisition module is configured to acquire CT images of a target group, wherein the target group comprises a mammalian group with osteoporosis and a mammalian group without osteoporosis, the CT images comprise a plurality of CT images of distal femoral segments of the target group, and each of the plurality of CT images is labeled with an osteoporosis degree label and an age label; A region division module is configured to divide a background region and an area of interest on each CT image, wherein the area of interest comprises a trabecular region and a cortical region; A region subdivision module is configured to perform region subdivision on the area of interest on each CT image according to a gray value, to obtain a plurality of sub-regions in the area of interest and area proportions of each of the plurality of sub-regions in the area of interest; A model training module is configured to train a bone aging degree prediction model based on a deep learning model framework according to the CT images of the target group and the area proportions of each sub-region on each CT image in the area of interest, in combination with the osteoporosis degree age and age label of the target group.

8. An osteoporosis degree prediction system characterized by comprising: The method comprises the following steps: An image receiving module is configured to receive CT images and age data of a to-be-tested individual, wherein the type of the to-be-tested individual is consistent with the type of the target group in the method for constructing the bone aging degree prediction model according to any one of claims 1-6, and the CT images are CT images of distal femoral segments of the to-be-tested individual; A prediction module is configured to obtain prediction results of whether the to-be-tested individual has osteoporosis and the osteoporosis degree according to the CT images and the age data of the to-be-tested individual by the bone aging degree prediction model obtained by the method for constructing the bone aging degree prediction model according to any one of claims 1-6, wherein the osteoporosis degree reflects whether the bone strength shown in the CT images deviates from the reference bone strength at the corresponding age stage; A determination module is configured to determine whether the to-be-tested individual has bone aging according to the prediction results of the osteoporosis degree of the to-be-tested individual.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the following steps when executing the program: receiving CT image and age data of a subject, wherein the subject is of the same type as the target group in the method for constructing the bone aging degree prediction model according to any one of claims 1-6, and the CT image is of the distal femur of the subject; obtaining a prediction result of whether the subject has osteoporosis and the degree of osteoporosis according to the CT image and the age data of the subject and the bone aging degree prediction model obtained by the method for constructing the bone aging degree prediction model according to any one of claims 1-6, wherein the degree of osteoporosis reflects whether the bone strength shown in the CT image deviates from the reference bone strength of the corresponding age stage; determining whether the subject has bone aging according to the prediction result of the degree of osteoporosis of the subject. 10.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the following steps: receiving CT image and age data of a subject, wherein the subject is of the same type as the target group in the method for constructing the bone aging degree prediction model according to any one of claims 1-6, and the CT image is of the distal femur of the subject; obtaining a prediction result of whether the subject has osteoporosis and the degree of osteoporosis according to the CT image and the age data of the subject and the bone aging degree prediction model obtained by the method for constructing the bone aging degree prediction model according to any one of claims 1-6, wherein the degree of osteoporosis reflects whether the bone strength shown in the CT image deviates from the reference bone strength of the corresponding age stage; determining whether the subject has bone aging according to the prediction result of the degree of osteoporosis of the subject.

Images