Adult bone age estimation method and device, electronic equipment and storage medium

Abstract

The invention is suitable for the technical field of bone age estimation, and provides an adult bone age estimation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining image data of a pelvis of a target adult; according to the image data, determining the bone age through a pre-trained pelvic bone age estimation model; wherein the pelvic bone age estimation model is composed of a Swin Transformer model, a mesh activation function, a linear layer and a Sigmoid activation function which are connected in sequence. According to the invention, more accurate bone age estimation can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of bone age estimation, and in particular relates to a method, device, electronic equipment and storage medium for estimating the bone age of an adult. Background Art

[0002] Age estimation based on bones is an important research topic in anthropology, archaeology and forensic medicine. Traditional bone age estimation methods require experts to make subjective judgments on the classification of bone morphology. The results are greatly affected by subjectivity and the accuracy is greatly reduced.

[0003] In recent years, artificial intelligence has increasingly assisted bone age estimation, improving the accuracy and efficiency of bone age estimation. Despite the progress made in artificial intelligence, there are still some challenges in the field of bone age estimation. The prediction of adult bone age is more complicated due to the complete closure of the epiphysis, and there are differences in bone morphology between different individuals. Therefore, it is necessary to develop a more accurate algorithm model to improve the accuracy of bone age estimation. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a method, device, electronic device and storage medium for estimating bone age of an adult to achieve more accurate bone age estimation.

[0005] A first aspect of an embodiment of the present invention provides a method for estimating bone age of an adult, comprising:

[0006] Acquire image data of a target adult pelvis;

[0007] Determining bone age based on the image data using a pre-trained pelvic bone age estimation model;

[0008] The pelvic bone age estimation model is composed of a Swin Transformer model, a mish activation function, a linear layer and a Sigmoid activation function connected in sequence.

[0009] In combination with the first aspect, in a possible implementation of the first aspect, the pelvic bone age estimation model is pre-trained by the following method:

[0010] Acquire image data of different adult pelvises from a preset pelvis database to generate a training data set for the pelvic bone age estimation model;

[0011] According to the training data set, the model parameters of the initial model are iteratively optimized through a preset transfer learning algorithm to obtain the pelvic bone age estimation model.

[0012] In combination with the first aspect, in a possible implementation manner of the first aspect, the transfer learning algorithm is an AdamW optimization algorithm;

[0013] The AdamW optimization algorithm includes:

[0014] Update the first moment estimate:

[0015] m t =β1m t-1 +(1-β1)g t ;

[0016] Update the second moment estimate:

[0017]

[0018] Bias-corrected first-moment estimate:

[0019]

[0020] Bias-corrected second moment estimate:

[0021]

[0022] Update model parameters:

[0023]

[0024] Among them, α is the learning rate; β1 and β2 are the first-order moment decay rate and the second-order moment decay rate, respectively, with typical values ​​of 0.9 and 0.999; ∈ is a constant, with a typical value of 1e-8; t is the current number of iterations; g t is the gradient of the loss function; λ is the weight decay rate; θ is the model parameter.

[0025] In combination with the first aspect, in a possible implementation manner of the first aspect, the loss function set in the transfer learning algorithm is:

[0026] L = δ·MSELoss+ζ·L1Loss;

[0027]

[0028] Among them, δ and ζ are the hyperparameters that control the weights of MSELoss and L1Loss, y i is the actual value, is the predicted value, and N is the number of samples.

[0029] In combination with the first aspect, in a possible implementation manner of the first aspect, the image data includes: an MSCT image, and a 3D image reconstructed from the MSCT image;

[0030] The step of acquiring image data of different adult pelvises from a preset pelvic database to generate a training data set for the pelvic bone age estimation model includes:

[0031] The MSCT image and the 3D image are preprocessed, and a training data set for the pelvic bone age estimation model is generated based on the preprocessed image data.

[0032] In combination with the first aspect, in a possible implementation manner of the first aspect, preprocessing the MSCT image and the 3D image includes:

[0033] unifying the size of the MSCT images and the size of the 3D images;

[0034] By using image affine transformation and Gaussian noise, data enhancement is performed on the MSCT image and the 3D image respectively, so as to increase the number of samples of the MSCT image and the 3D image;

[0035] The MSCT image and the 3D image are normalized.

[0036] In combination with the first aspect, in a possible implementation of the first aspect, after optimizing the model parameters of the initial model by a preset transfer learning algorithm according to the training data set to obtain the pelvic bone age estimation model, the method further includes:

[0037] Testing the performance of the pelvic bone age estimation model using a validation dataset;

[0038] According to the test results, the model parameters of the pelvic bone age estimation model are adjusted.

[0039] A second aspect of an embodiment of the present invention provides an adult bone age estimation device, comprising:

[0040] An acquisition module, used for acquiring image data of a target adult pelvis;

[0041] The age estimation module is used to determine the bone age according to the image data through a pre-trained pelvic bone age estimation model; wherein the pelvic bone age estimation model is composed of a Swin Transformer model, a mish activation function, a linear layer and a Sigmoid activation function connected in sequence.

[0042] In conjunction with the second aspect, in a possible implementation manner of the second aspect, the age estimation module is further used to:

[0043] Acquire image data of different adult pelvises from a preset pelvis database to generate a training data set for the pelvic bone age estimation model;

[0044] According to the training data set, the model parameters of the initial model are iteratively optimized through a preset transfer learning algorithm to obtain the pelvic bone age estimation model.

[0045] In conjunction with the second aspect, in a possible implementation of the second aspect, the transfer learning algorithm is an AdamW optimization algorithm;

[0046] The AdamW optimization algorithm includes:

[0047] Update the first moment estimate:

[0048] m t =β1m t-1 +(1-β1)g t ;

[0049] Update the second moment estimate:

[0050]

[0051] Bias-corrected first-moment estimate:

[0052]

[0053] Bias-corrected second moment estimate:

[0054]

[0055] Update model parameters:

[0056]

[0057] Among them, α is the learning rate; β1 and β2 are the first-order moment decay rate and the second-order moment decay rate, respectively, with typical values ​​of 0.9 and 0.999; ∈ is a constant, with a typical value of 1e-8; t is the current number of iterations; g t is the gradient of the loss function; λ is the weight decay rate; θ is the model parameter.

[0058] In combination with the second aspect, in a possible implementation manner of the second aspect, the loss function set in the transfer learning algorithm is:

[0059] L = δ·MSELoss+ζ·L1Loss;

[0060]

[0061] Among them, δ and ζ are the hyperparameters that control the weights of MSELoss and L1Loss, y i is the actual value, is the predicted value, and N is the number of samples.

[0062] In combination with the second aspect, in a possible implementation manner of the second aspect, the image data includes: an MSCT image, and a 3D image reconstructed from the MSCT image;

[0063] The age estimation module is specifically used for:

[0064] The MSCT image and the 3D image are preprocessed, and a training data set for the pelvic bone age estimation model is generated based on the preprocessed image data.

[0065] In conjunction with the second aspect, in a possible implementation manner of the second aspect, the age estimation module is specifically configured to:

[0066] unifying the size of the MSCT images and the size of the 3D images;

[0067] By using image affine transformation and Gaussian noise, data enhancement is performed on the MSCT image and the 3D image respectively, so as to increase the number of samples of the MSCT image and the 3D image;

[0068] The MSCT image and the 3D image are normalized.

[0069] In conjunction with the second aspect, in a possible implementation manner of the second aspect, the age estimation module is specifically configured to:

[0070] Testing the performance of the pelvic bone age estimation model using a validation dataset;

[0071] According to the test results, the model parameters of the pelvic bone age estimation model are adjusted.

[0072] A third aspect of an embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method in the first aspect or any one of the implementations of the first aspect are implemented.

[0073] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the method in the above-mentioned first aspect or any one of the implementation methods of the first aspect.

[0074] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0075] Since the epiphysis is completely closed, the prediction of adult bone age is more complicated. The pubic symphysis is the most important bone marker for estimating age. Therefore, the embodiment of the present invention estimates bone age through image data of the adult pelvis. The embodiment of the present invention constructs a pelvic bone age estimation model, which adds a mish activation function, a linear layer and a Sigmoid activation function at the end of the Swin Transformer model, which can enhance the model to learn the complex relationship of pelvic data, and express the extracted features within a specified range, generate a probabilistic prediction of bone age, which is suitable for bone age estimation tasks and improves the accuracy of bone age estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0077] Figure 1 The following is a schematic diagram of the implementation process of the method for estimating adult bone age provided by the embodiment of the present invention. Figure 1 ;

[0078] Figure 2 is a schematic diagram of the structure of a pelvic bone age estimation model provided by an embodiment of the present invention;

[0079] Figure 3 The following is a schematic diagram of the implementation process of the method for estimating adult bone age provided by the embodiment of the present invention. Figure 2 ;

[0080] Figure 4 is a performance diagram of the pelvic bone age estimation model provided by an embodiment of the present invention;

[0081] Figure 5 is a schematic diagram of the structure of an adult bone age estimation device provided by an embodiment of the present invention;

[0082] Figure 6 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0083] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0084] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0085] Compared with traditional age estimation methods, artificial intelligence technology provides more objective and broader observation possibilities. Despite the progress made in artificial intelligence, there are still some challenges in the field of age estimation. The prediction of adult bone age is more complicated due to the complete closure of the epiphysis. In forensic anthropology, the pubic symphysis is the most important bone marker for estimating age, but because the adult pelvis is often affected by diseases, nutritional status, living habits and other factors, its morphological diversity is strong, which reduces the accuracy of the estimation. Due to the differences in bone morphology between different individuals, a comprehensive validation study of different populations and bone conditions is needed, which is crucial to ensure the stability of the application of artificial intelligence in bone age estimation.

[0086] The application of deep learning technology has demonstrated the potential to improve the accuracy and efficiency of age estimation and break the limitations of traditional methods. In view of the problem that traditional pelvic adult age estimation is easily affected by the subjective factors of the observer, which greatly reduces the accuracy of adult age estimation results, the embodiment of the present invention adopts deep learning technology to propose an adult bone age estimation method that optimizes SwinTransformer.

[0087] See also Figure 1 As shown, the method for estimating adult bone age includes:

[0088] Step S101, acquiring image data of a target adult pelvis.

[0089] Here, the image data of the target adult pelvis includes: MSCT images (i.e., multislieces helieal CT, multi-slice spiral CT), and 3D images reconstructed from the MSCT images. In forensic anthropology, the pubic symphysis is the most important bone marker for estimating age, so this embodiment estimates age through the image data of the pelvis.

[0090] The MSCT image of the pelvis can be acquired by scanning with an MSCT device, and a 3D image can be obtained by three-dimensionally reconstructing the MSCT image.

[0091] After acquiring the image data of the target adult pelvis, the MSCT image and the 3D image in the image data may be preprocessed. For example, the number of slices of the MSCT image of each sample is different, so the MSCT image needs to be preprocessed into a uniform size (224, 224, 39). At the same time, the 3D image also needs to be processed into the same size (224, 224, 3) due to its different sizes.

[0092] Step S102, determining the bone age according to the image data through a pre-trained pelvic bone age estimation model; wherein the pelvic bone age estimation model is composed of a Swin Transformer model, a mish activation function, a linear layer and a Sigmoid activation function connected in sequence.

[0093] Swin Transformer is a deep learning model based on the Transformer architecture. It introduces the concepts of hierarchical and shifted windows to enable the model to effectively process image data.

[0094] The design of Swin Transformer allows it to be used as a general backbone network for computer vision tasks, suitable for a variety of tasks such as image classification, object detection, semantic segmentation, etc.

[0095] Since the existing Swin Transformer model cannot complete the task of enhancing the model to learn the complex relationship of pelvic data and represent the extracted features within a specified range to generate the regression task of probabilistic prediction of bone age, this embodiment constructs an improved Swin Transformer model, such as Figure 2 As shown in the figure, a mish activation function, a linear layer and a Sigmoid activation function are added at the end of the Swin Transformer model, which can enhance the model to learn the complex relationship of pelvic data, represent the extracted features within the specified range, and generate a probabilistic prediction of bone age, which is suitable for bone age estimation tasks.

[0096] For the above-mentioned improved Swin Transformer model, this embodiment optimizes the model parameters of the Swin Transformer model through transfer learning technology. The optimized Swin Transformer model is applied to the test set to estimate the adult pelvic age. Finally, the model performance index is output to evaluate the accuracy and effect of the model in predicting bone age.

[0097] The following, combined Figure 3 As shown, the training process of the model is explained:

[0098] Step S301, obtaining image data of different adult pelvises from a preset pelvis database to generate a training data set for a pelvic bone age estimation model.

[0099] Exemplarily, the pelvis database consists of 164 MSCT images and 3D images of the pelvis of adults aged 38 to 74. The MSCT images of the pelvis are acquired by scanning with an MSCT device, and the MSCT images of the samples are three-dimensionally reconstructed to obtain the 3D images of the samples.

[0100] Step S302, preprocessing the MSCT image and the 3D image, and generating a training data set for the pelvic bone age estimation model based on the preprocessed image data.

[0101] Here, a part of the samples can be selected as the training data set for model training, and the other part of the samples can be selected as the validation data set for model testing.

[0102] Among them, the MSCT image and the 3D image are preprocessed, including:

[0103] (1) Unify the size of MSCT images and 3D images.

[0104] Since the number of slices of the MSCT images of each sample is different, the MSCT images need to be preprocessed into a uniform size (224, 224, 39). At the same time, the 3D images are also processed into the same size (224, 224, 3) due to their different sizes.

[0105] (2) Through image affine transformation and Gaussian noise, the data of MSCT images and 3D images are enhanced respectively, thereby increasing the number of samples of MSCT images and 3D images.

[0106] Due to the small number of samples, each slice of the MSCT image of each sample is regarded as a sample with the same label, as is each channel of the 3D image of each sample. In addition, image affine transformation and Gaussian noise are used as data augmentation techniques to expand the number of samples.

[0107] (3) Normalize the MSCT images and 3D images.

[0108] The MSCT image and 3D image data are normalized to a range of 0 to 1, so that the model constructed later can converge more easily.

[0109] Step S303, based on the training data set, the model parameters of the initial model are iteratively optimized through a preset transfer learning algorithm to obtain a pelvic bone age estimation model.

[0110] The optimization problem of the model can be solved using AdamW. AdamW uses Adam's momentum and adaptive learning rate and adds a weight decay method. It has better stability and convergence speed when processing high-dimensional and complex neural networks, which helps to improve the accuracy and generalization ability of the model.

[0111] Adam is widely used in many deep learning frameworks and performs well on various tasks, especially in natural language processing and computer vision. This embodiment improves on the basis of the Adam optimizer. Unlike the traditional Adam optimizer, this embodiment adjusts the implementation of Adam weight decay to improve the performance and convergence of the model.

[0112] Adam adjusts the learning rate for each parameter by estimating the exponentially weighted average of the gradients (called the first-order moment estimate) and the square of the gradients (called the second-order moment estimate).

[0113] The AdamW optimization algorithm includes:

[0114] Update the first moment estimate:

[0115] m t =β1m t-1 +(1-β1)g t ;

[0116] Update the second moment estimate:

[0117]

[0118] Bias-corrected first-moment estimate:

[0119]

[0120] Bias-corrected second moment estimate:

[0121]

[0122] Update model parameters:

[0123]

[0124] Among them, α is the learning rate; β1 and β2 are the first-order moment decay rate and the second-order moment decay rate, respectively, with typical values ​​of 0.9 and 0.999; ∈ is a constant, with a typical value of 1e-8; t is the current number of iterations; g t is the gradient of the loss function; λ is the weight decay rate; θ is the model parameter.

[0125] As a possible implementation, this embodiment uses MSELoss (mean square error loss) combined with L1Loss as a loss function to improve the performance of the improved optimized Swin Transformer model in adult age estimation. Since adult age estimation is a regression problem, age estimation needs to be accurate. Using MSELoss helps ensure that the model is sufficiently resilient to large errors in age estimation. At the same time, L1Loss can help the model not overfit specific abnormal data points when facing some complex or noisy images, thereby improving the overall prediction accuracy and robustness of the model.

[0126] The loss function set in the transfer learning algorithm is:

[0127] L = δ·MSELoss+ζ·L1Loss;

[0128]

[0129] Among them, δ and ζ are the hyperparameters that control the weights of MSELoss and L1Loss, y i is the actual value, is the predicted value, and N is the number of samples.

[0130] Step S304, testing the performance of the pelvic bone age estimation model by using a validation data set; and adjusting the model parameters of the pelvic bone age estimation model according to the test results.

[0131] This embodiment uses the validation data set to test and adjust the obtained optimal model to make the model more suitable for estimating the adult age of the pelvis.

[0132] In one example, for the pelvic bone age estimation model, the following indicators were used: MSE (mean square error), RMSE (root mean square error), MAE (mean absolute error), R 2 (coefficient of determination) and Loss function to evaluate the performance of the model. The results are shown in Table 1. The predicted values ​​and actual values ​​are shown in Table 1. Figure 4 As shown (Train Loss in the figure is the loss on the training data set, and Valid Loss is the loss on the validation data set).

[0133] Table 1 Performance evaluation results

[0134] index MSE RMSE MAE <![CDATA[R 2 ]]> Numeric 4.2682 2.066 1.8148 0.9371

[0135] It can be clearly seen from Table 1 that the coefficient of determination (R 2 ) is 0.9371 and the root mean square error (RMSE) is 2.066, which effectively shows that the model's estimate of adult age is accurate. Figure 4It can be seen that the loss of the model on the training set and the validation set decreases rapidly within 75 epochs and then decreases slowly, which fully demonstrates the effectiveness of the loss function composed of MREloss and L1Loss.

[0136] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0137] Figure 5 : is a schematic diagram of the structure of an adult bone age estimation device 50 provided in an embodiment of the present invention, the device comprising:

[0138] The acquisition module 51 is used to acquire image data of the target adult pelvis.

[0139] The age estimation module 52 is used to determine the bone age based on the image data through a pre-trained pelvic bone age estimation model; wherein the pelvic bone age estimation model is composed of a Swin Transformer model, a mish activation function, a linear layer and a Sigmoid activation function connected in sequence.

[0140] As a possible implementation, the age estimation module 52 is further configured to:

[0141] Obtain image data of different adult pelvises from a preset pelvis database to generate a training data set for a pelvic bone age estimation model;

[0142] According to the training data set, the model parameters of the initial model are iteratively optimized through the preset transfer learning algorithm to obtain the pelvic bone age estimation model.

[0143] As a possible implementation method, the transfer learning algorithm is the AdamW optimization algorithm;

[0144] The AdamW optimization algorithm includes:

[0145] Update the first moment estimate:

[0146] m t =β1m t-1 +(1-β1)g t ;

[0147] Update the second moment estimate:

[0148]

[0149] Bias-corrected first-moment estimate:

[0150]

[0151] Bias-corrected second moment estimate:

[0152]

[0153] Update model parameters:

[0154]

[0155] Among them, α is the learning rate; β1 and β2 are the first-order moment decay rate and the second-order moment decay rate, respectively, with typical values ​​of 0.9 and 0.999; ∈ is a constant, with a typical value of 1e-8; t is the current number of iterations; g t is the gradient of the loss function; λ is the weight decay rate; θ is the model parameter.

[0156] As a possible implementation method, the loss function set in the transfer learning algorithm is:

[0157] L = δ·MSELoss+ζ·L1Loss;

[0158]

[0159] Among them, δ and ζ are the hyperparameters that control the weights of MSELoss and L1Loss, y i is the actual value, is the predicted value, and N is the number of samples.

[0160] As a possible implementation, the image data includes: an MSCT image, and a 3D image reconstructed from the MSCT image;

[0161] The age estimation module 52 is specifically used for:

[0162] The MSCT images and 3D images are preprocessed, and a training data set for the pelvic bone age estimation model is generated based on the preprocessed image data.

[0163] As a possible implementation, the age estimation module 52 is specifically used for:

[0164] Standardize the size of MSCT images and 3D images;

[0165] Through image affine transformation and Gaussian noise, data enhancement is performed on MSCT images and 3D images respectively, thereby increasing the number of samples of MSCT images and 3D images.

[0166] The MSCT images and 3D images were normalized.

[0167] As a possible implementation, the age estimation module 52 is specifically used for:

[0168] The performance of the pelvic bone age estimation model was tested using a validation dataset;

[0169] According to the test results, the model parameters of the pelvic bone age estimation model are adjusted.

[0170] Since the epiphysis is completely closed, the prediction of adult bone age is more complicated. The pubic symphysis is the most important bone marker for estimating age. Therefore, the embodiment of the present invention estimates bone age through image data of the adult pelvis. The embodiment of the present invention constructs a pelvic bone age estimation model, which adds a mish activation function, a linear layer and a Sigmoid activation function at the end of the Swin Transformer model, which can enhance the model to learn the complex relationship of pelvic data, and express the extracted features within a specified range, generate a probabilistic prediction of bone age, which is suitable for bone age estimation tasks and improves the accuracy of bone age estimation.

[0171] Figure 6 FIG. 6 is a schematic diagram of the structure of an electronic device 60 provided by an embodiment of the present invention. Figure 6 As shown, the electronic device 60 of this embodiment includes: a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61, such as an adult bone age estimation program. When the processor 61 executes the computer program 63, the steps in the above-mentioned embodiments of the adult bone age estimation method are implemented, such as Figure 1 Alternatively, when the processor 61 executes the computer program 63, the functions of the modules in the above-mentioned device embodiments are realized, for example, Figure 5 The functions of modules 51 and 52 are shown.

[0172] Exemplarily, the computer program 63 may be divided into one or more modules / units, which are stored in the memory 62 and executed by the processor 61 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program 63 in the electronic device 60.

[0173] The electronic device 60 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The electronic device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will appreciate that Figure 6 It is only an example of the electronic device 60 and does not constitute a limitation of the electronic device 60. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device 60 may also include input and output devices, network access devices, buses, etc.

[0174] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0175] The memory 62 may be an internal storage unit of the electronic device 60, such as a hard disk or memory of the electronic device 60. The memory 62 may also be an external storage device of the electronic device 60, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 60. Further, the memory 62 may also include both an internal storage unit of the electronic device 60 and an external storage device. The memory 62 is used to store the computer program and other programs and data required by the electronic device 60. The memory 62 may also be used to temporarily store data that has been output or is to be output.

[0176] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0177] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0178] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0179] In the embodiments provided by the present invention, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0180] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0181] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0182] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0183] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for estimating bone age of an adult, characterized in that: include: Acquire image data of a target adult pelvis; Determining bone age based on the image data using a pre-trained pelvic bone age estimation model; The pelvic bone age estimation model is composed of a Swin Transformer model, a mish activation function, a linear layer and a Sigmoid activation function connected in sequence.

2. The method for estimating bone age of an adult according to claim 1, characterized in that: The pelvic bone age estimation model is pre-trained by the following method: Acquire image data of different adult pelvises from a preset pelvis database to generate a training data set for the pelvic bone age estimation model; According to the training data set, the model parameters of the initial model are iteratively optimized through a preset transfer learning algorithm to obtain the pelvic bone age estimation model.

3. The method for estimating bone age of an adult according to claim 2, wherein: The transfer learning algorithm is the AdamW optimization algorithm; The AdamW optimization algorithm includes: Update the first moment estimate: m t =β1m t-1 +(1-β1)g t ; Update the second moment estimate: Bias-corrected first-moment estimate: Bias-corrected second moment estimate: Update model parameters: Among them, α is the learning rate; β1 and β2 are the first-order moment decay rate and the second-order moment decay rate, respectively, with typical values ​​of 0.9 and 0.999; ∈ is a constant, with a typical value of 1e-8; t is the current number of iterations; g t is the gradient of the loss function; λ is the weight decay rate; θ is the model parameter.

4. The method for estimating bone age of an adult according to claim 2, wherein: The loss function set in the transfer learning algorithm is: L = δ·MSELoss+ζ·L1Loss; Among them, δ and ζ are the hyperparameters that control the weights of MSELoss and L1Loss, y i is the actual value, is the predicted value, and N is the number of samples.

5. The method for estimating bone age of an adult according to claim 2, wherein: The image data includes: an MSCT image, and a 3D image reconstructed from the MSCT image; The step of acquiring image data of different adult pelvises from a preset pelvic database to generate a training data set for the pelvic bone age estimation model includes: The MSCT image and the 3D image are preprocessed, and a training data set for the pelvic bone age estimation model is generated based on the preprocessed image data.

6. The method for estimating bone age of an adult according to claim 5, characterized in that: The preprocessing of the MSCT image and the 3D image comprises: unifying the size of the MSCT images and the size of the 3D images; By using image affine transformation and Gaussian noise, data enhancement is performed on the MSCT image and the 3D image respectively, so as to increase the number of samples of the MSCT image and the 3D image; The MSCT image and the 3D image are normalized.

7. The method for estimating bone age of an adult according to claim 2, wherein: After optimizing the model parameters of the initial model by a preset transfer learning algorithm according to the training data set to obtain the pelvic bone age estimation model, the method further includes: Testing the performance of the pelvic bone age estimation model using a validation dataset; According to the test results, the model parameters of the pelvic bone age estimation model are adjusted.

8. A device for estimating bone age of an adult, characterized in that: include: An acquisition module, used for acquiring image data of a target adult pelvis; The age estimation module is used to determine the bone age according to the image data through a pre-trained pelvic bone age estimation model; wherein the pelvic bone age estimation model is composed of a Swin Transformer model, a mish activation function, a linear layer and a Sigmoid activation function connected in sequence.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images