Bone age prediction model training method, and bone age prediction method and device

Abstract

The invention provides a training method of a bone age prediction model and a bone age prediction method and device. The invention relates to the technical field of computers, in particular to a bone age prediction model training method which comprises the steps that a plurality of first samples and a plurality of second samples are obtained, the first samples comprise hand bone image samples of human bodies in a first group, and the second samples comprise hand bone image samples of human bodies in a second group; the group scale of the first group is larger than that of the second group; and using the sequence knowledge of the relative ages contained in the plurality of first samples to assist the plurality of second samples to train the bone age prediction model to obtain a target bone age prediction model for the second group.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and more specifically, to a training method for a bone age prediction model, a bone age prediction method, and an apparatus. Background Technology

[0002] In the field of pediatric medicine, bone age assessment is a crucial task in pediatric imaging, as it relates to determining the skeletal maturity of children and adolescents and assisting in the diagnosis and treatment of growth and development-related diseases. In recent years, significant progress has been made in automated bone age assessment with the help of deep learning and publicly available hand X-ray datasets, enabling end-to-end networks to directly predict bone age values ​​from images.

[0003] In the process of realizing the concept of this disclosure, it was found that at least the following problems exist in the related technology: the results output by the existing methods for predicting bone age are difficult to meet the requirements of stable and reliable applications. Summary of the Invention

[0004] In view of this, this disclosure provides a training method for a bone age prediction model, a bone age prediction method, and an apparatus.

[0005] One aspect of this disclosure provides a method for training a bone age prediction model, comprising: acquiring multiple first samples and multiple second samples, wherein the first samples include hand bone image samples of humans in a first group, the second samples include hand bone image samples of humans in a second group, and the group size of the first group is larger than that of the second group; and using the relative age order knowledge contained in the multiple first samples to assist the multiple second samples in training the bone age prediction model, thereby obtaining a target bone age prediction model for the second group.

[0006] According to embodiments of this disclosure, the bone age prediction model is trained by using the relative age order knowledge contained in multiple first samples to assist multiple second samples in training the model, including: in each training round, the bone age prediction model is trained by relative bone age order modeling using at least one batch of first samples; and the bone age prediction model is trained by absolute bone age calibration using at least one batch of second samples.

[0007] According to embodiments of this disclosure, a bone age prediction model is trained using first samples from at least one training batch for relative bone age order modeling, comprising: inputting multiple first samples from the training batch into the bone age prediction model to obtain multiple first bone age prediction values; calculating intra-batch ranking loss based on the multiple first bone age prediction values ​​to obtain intra-batch ranking loss values; calculating regression loss based on the first bone age label values ​​of each of the multiple first samples in the training batch and the multiple first bone age prediction values ​​to obtain a first regression loss value; performing a weighted summation of the intra-batch ranking loss value and the first regression loss value to obtain a first loss value; and adjusting the parameters of the bone age prediction model using the first loss value.

[0008] According to embodiments of this disclosure, an intra-batch ranking loss calculation is performed based on multiple first bone age prediction values ​​to obtain an intra-batch ranking loss value, including: determining multiple sample pairs from multiple first samples in a training batch based on the relative order of their respective first bone age label values; calculating a ranking loss based on two first bone age prediction values ​​associated with each sample pair to obtain a ranking loss value; and calculating the mean of the ranking loss values ​​of the multiple samples to obtain an intra-batch ranking loss value.

[0009] According to embodiments of this disclosure, training an absolute bone age calibration model using at least one training batch of second samples includes: inputting multiple second samples from the training batch into the bone age prediction model to obtain multiple second bone age prediction values; calculating a regression loss based on the second bone age label values ​​of each of the multiple second samples from the training batch and the multiple second bone age prediction values ​​to obtain a second loss value; and adjusting the parameters of the bone age prediction model using the second loss value.

[0010] According to an embodiment of this disclosure, in each training round, the number of training batches for the first sample is the number of the first batch, and the number of training batches for the second sample is the number of the second batch; the method further includes: determining the number of the first batch and the number of the second batch in the current training round based on the round number of the current training round.

[0011] According to embodiments of this disclosure, the quantity of the first batch decreases as the number of training rounds increases, while the quantity of the second batch increases as the number of training rounds increases.

[0012] Another aspect of this disclosure provides a bone age prediction method, comprising: acquiring hand bone image data of a target object, wherein the target object belongs to a second group; and inputting the hand bone image data into a target bone age prediction model to obtain a bone age prediction result for the target object; wherein the target bone age prediction model is trained using the above-described bone age prediction model training method.

[0013] Another aspect of this disclosure provides a training apparatus for a bone age prediction model, comprising:

[0014] The first acquisition module is used to acquire multiple first samples and multiple second samples; the training module is used to use the relative age order knowledge contained in the multiple first samples to assist the multiple second samples in training the bone age prediction model, so as to obtain the target bone age prediction model for the second group.

[0015] Another aspect of this disclosure provides a bone age prediction device, comprising: a second acquisition module for acquiring hand bone image data of a target object, wherein the target object belongs to a second group; and a determination module for inputting the hand bone image data into a target bone age prediction model to obtain a bone age prediction result for the target object.

[0016] According to embodiments of this disclosure, the relative age order knowledge of the majority group is used to assist the training of the bone age prediction model with a second sample, thereby obtaining a target bone age prediction model for the second group. This training method effectively solves the problems of poor generalization ability and low prediction accuracy of bone age prediction models due to insufficient data in minority group samples. It not only reduces the dependence of the bone age prediction model on labeled data but also reduces the risk of overfitting, improving the applicability and fairness of the prediction model when predicting the bone age of minority groups. Furthermore, the method of this disclosure can significantly improve sample utilization efficiency and alleviate the problem of high costs associated with medical image acquisition and labeling. Attached Figure Description

[0017] The above and other objects, features and advantages of this disclosure will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0018] Figure 1 An exemplary system architecture for training a bone age prediction model that can be applied according to embodiments of the present disclosure is illustrated.

[0019] Figure 2 A flowchart illustrating a method for training a bone age prediction model according to an embodiment of the present disclosure is shown.

[0020] Figure 3 A flowchart illustrating a training method for relative bone age order modeling training according to an embodiment of the present disclosure is shown.

[0021] Figure 4 A data flow diagram illustrating a training method for a bone age prediction model according to an embodiment of the present disclosure is shown schematically.

[0022] Figure 5 A block diagram of a training apparatus for a bone age prediction model according to an embodiment of the present disclosure is shown schematically.

[0023] Figure 6 A block diagram of a bone age prediction device according to an embodiment of the present disclosure is illustrated schematically; and

[0024] Figure 7 A block diagram of an electronic device suitable for training a bone age prediction model according to an embodiment of the present disclosure is shown schematically. Detailed Implementation

[0025] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0028] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0029] In the embodiments disclosed herein, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard user personal information security, network security, and national security.

[0030] In the embodiments disclosed herein, user authorization or consent is obtained before acquiring or collecting user personal information.

[0031] Existing methods generally rely on large-scale homogeneous data for training, and the models perform well within the training distribution. However, they are prone to generalization degradation when faced with population differences, differences in data collection equipment, and the complexity of clinical scenarios. Especially when evaluating specific populations under conditions of small sample size, traditional transfer or fine-tuning strategies are prone to overfitting and cannot meet the requirements for stable and reliable applications.

[0032] The bone age reference systems widely used in clinical practice are established based on specific eras and populations, with limited population coverage, making it difficult to fully reflect the physiological differences among contemporary ethnic groups and regions. In reality, medical imaging samples from minority groups are inherently scarce, and constructing large-scale, high-quality datasets is costly and time-consuming, making it difficult to directly transfer methods based on big data learning. Therefore, there is an urgent need for a novel deep learning algorithm that can effectively inject comparative relationships and prior hierarchical information from within and outside the population, while balancing efficient adaptation and lightweight transfer, to improve the generalization and clinical usability of bone age assessment using small datasets.

[0033] In view of this, embodiments of the present disclosure provide a method for training a bone age prediction model, comprising: acquiring multiple first samples and multiple second samples, wherein the first samples include hand bone image samples of humans in a first group, the second samples include hand bone image samples of humans in a second group, and the group size of the first group is larger than that of the second group; and using the relative age order knowledge contained in the multiple first samples to assist the multiple second samples in training the bone age prediction model, thereby obtaining a target bone age prediction model for the second group.

[0034] Figure 1 An exemplary system architecture 100, according to embodiments of the present disclosure, is illustrated in schematic form, in which a training method for a bone age prediction model can be applied. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0035] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0036] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social media platform software, etc. (for example only).

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0038] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0039] It should be noted that the training method for the bone age prediction model provided in this embodiment can generally be executed by server 105. Correspondingly, the training device for the bone age prediction model provided in this embodiment can generally be located in server 105. The training method for the bone age prediction model provided in this embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the training device for the bone age prediction model provided in this embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Alternatively, the training method for the bone age prediction model provided in this embodiment can also be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103, or by other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103. Accordingly, the training device for the bone age prediction model provided in this embodiment can also be set in the first terminal device 101, the second terminal device 102 or the third terminal device 103, or in other terminal devices different from the first terminal device 101, the second terminal device 102 or the third terminal device 103.

[0040] For example, the sample to be processed may originally be stored in any one of the first terminal device 101, the second terminal device 102, or the third terminal device 103 (e.g., the first terminal device 101, but not limited thereto), or it may be stored on an external storage device and imported into the first terminal device 101. Then, the first terminal device 101 may locally execute the training method of the bone age prediction model provided in the embodiments of this disclosure, or send the sample to be processed to other terminal devices, servers, or server clusters, and have the other terminal devices, servers, or server clusters that receive the sample to be processed execute the training method of the bone age prediction model provided in the embodiments of this disclosure.

[0041] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0042] Figure 2 A flowchart illustrating a method for training a bone age prediction model according to an embodiment of the present disclosure is shown.

[0043] like Figure 2 As shown, the method includes operations S210~S220.

[0044] In operation S210, multiple first samples and multiple second samples are acquired. The first samples include hand bone image samples of humans in a first group, and the second samples include hand bone image samples of humans in a second group. The group size of the first group is larger than the group size of the second group.

[0045] According to embodiments of this disclosure, both the first and second groups can include individuals from multiple age groups, with the first group being larger than the second group. Therefore, the first sample can be data collected from areas where samples are readily available and where there is a large ethnic group or a large population; in other words, the first sample is more representative of the hand bone images of the majority of the population. The second sample can be data collected from areas where samples are difficult to obtain and where there is a small ethnic group or a small population; that is, the second sample consists of hand bone images of a population that existing prediction models struggle to predict.

[0046] In operation S220, the relative age order knowledge contained in multiple first samples is used to assist multiple second samples in training the bone age prediction model, thereby obtaining the target bone age prediction model for the second group.

[0047] According to embodiments of this disclosure, bone age is an indicator used to assess an individual's skeletal maturity by analyzing skeletal development stages. It is commonly used in pediatric medicine to determine whether growth and development are normal and to predict adult height. In this disclosure, prediction is based on hand bone images. The aforementioned order knowledge can represent the relationship between hand bone images of people of different ages obtained by learning from a large number of samples in the first sample. Order knowledge allows the model to learn relative order relationships, such as the relationship between the hand bone images of A and B when A is older than B, rather than simply focusing on the absolute difference between the predicted and true values. When training using order knowledge and a second sample, a bone age prediction model specific to the second sample can be trained.

[0048] According to embodiments of this disclosure, even if the amount of data in the second sample is scarce, the model's predicted value distribution can be prevented from collapsing due to the use of order knowledge for auxiliary training, thus balancing order learning and numerical accuracy, enabling the model to accurately determine bone age trends even in a small number of people.

[0049] According to embodiments of this disclosure, the relative age order knowledge of the majority group is used to assist the training of the bone age prediction model with a second sample, thereby obtaining a target bone age prediction model for the second group. This training method effectively solves the problems of poor generalization ability and low prediction accuracy of bone age prediction models due to insufficient data in minority group samples. It not only reduces the dependence of the bone age prediction model on labeled data but also reduces the risk of overfitting, improving the applicability and fairness of the prediction model when predicting the bone age of minority groups. Furthermore, the method of this disclosure can significantly improve sample utilization efficiency and alleviate the problem of high costs associated with medical image acquisition and labeling.

[0050] According to embodiments of this disclosure, the bone age prediction model is trained by using the relative age order knowledge contained in multiple first samples to assist multiple second samples in training the model, including: in each training round, the bone age prediction model is trained by relative bone age order modeling using at least one batch of first samples; and the bone age prediction model is trained by absolute bone age calibration using at least one batch of second samples.

[0051] According to embodiments of this disclosure, the bone age prediction model can be trained by using a first sample from at least one training batch to model the relative bone age order, thereby obtaining order knowledge; the bone age prediction model can be trained by using a second sample from at least one training batch to calibrate the absolute bone age, thereby using the order knowledge to assist the model in calibrating the absolute bone age for the second sample.

[0052] The majority population corresponding to the first sample can be divided into multiple training batches, and the minority population corresponding to the second sample can be divided into multiple training batches. Then, different batches can be used for training to obtain order knowledge and perform absolute bone age calibration on the second sample. The order knowledge enhances the model's generalization ability, while the absolute bone age calibration enables the model to accurately predict the bone age of the population in the second sample.

[0053] Figure 3 A flowchart illustrating a training method for relative bone age order modeling training according to an embodiment of the present disclosure is shown.

[0054] like Figure 3 As shown, the training method for relative bone age order modeling training includes operations S310 to S350.

[0055] According to embodiments of this disclosure, a relative bone age order modeling training is performed on a bone age prediction model using first samples from at least one training batch, including:

[0056] In operation S310, multiple first samples from the training batch are input into the bone age prediction model to obtain multiple first bone age prediction values.

[0057] In operation S320, based on multiple first bone age prediction values, the intra-batch sorting loss is calculated to obtain the intra-batch sorting loss value.

[0058] In operation S330, based on the first bone age label value and multiple first bone age prediction values ​​of each of the first samples in the training batch, regression loss is calculated to obtain the first regression loss value.

[0059] In operation S340, the weighted sum of the intra-batch ranking loss value and the first regression loss value is performed to obtain the first loss value.

[0060] When operating S350, the parameters of the bone age prediction model are adjusted using the first loss value.

[0061] According to embodiments of this disclosure, the aforementioned first sample can come from a majority population sample. Firstly, multiple first samples from the current training batch of the majority population can be input into the bone age prediction model. The bone age prediction model then predicts and outputs a first bone age prediction value corresponding to the first sample. Additionally, the first sample also includes a first bone age label value. Next, while satisfying the order condition of the first bone age label values, the intra-batch ranking loss can be calculated by combining the first bone age prediction values ​​to obtain the intra-batch ranking loss value for the current training batch. During training, the bone age prediction model can associate the magnitude relationship of multiple first bone age label values ​​in the majority population with the hand bone image corresponding to the first sample by reducing the intra-batch ranking loss value, allowing the bone age prediction model to obtain the aforementioned order knowledge.

[0062] Furthermore, the first bone age prediction value and the first bone age label value obtained above can be used to calculate the regression loss to obtain the first regression loss value of the current training batch. During training, the bone age prediction model reduces the first regression loss value to make the first bone age prediction value closer to the first bone age label value, that is, to make the bone age prediction model predict more and more accurately.

[0063] Next, the intra-batch ranking loss value and the first regression loss value obtained above are weighted and summed to obtain the first loss value for the current training batch. Then, the parameters of the bone age prediction model are adjusted based on the first loss value to achieve the purpose of model training. Integrating the two loss values ​​for training can not only make the model prediction more and more accurate, but also enable the model to learn the correlation between the first samples, further enabling the bone age prediction model to better perform subsequent instruction transfer operations.

[0064] According to embodiments of this disclosure, an intra-batch ranking loss calculation is performed based on multiple first bone age prediction values ​​to obtain an intra-batch ranking loss value, including: determining multiple sample pairs from multiple first samples in a training batch based on the relative order of their respective first bone age label values; calculating a ranking loss based on two first bone age prediction values ​​associated with each sample pair to obtain a ranking loss value; and calculating the mean of the ranking loss values ​​of the multiple samples to obtain an intra-batch ranking loss value.

[0065] According to embodiments of this disclosure, the method for obtaining the order loss value firstly determines multiple sample pairs in the current training batch. The sample pairs must have a size relationship, that is, satisfy the above-mentioned order condition. Then, the order loss value of each sample pair that satisfies the order condition is calculated. Then, the order loss of all sample pairs in the current training batch is averaged to obtain the intra-batch order loss value.

[0066] According to embodiments of this disclosure, training batches for the majority population can be marked as ,in, This represents the majority group. For a given sample pair in the current training batch within the majority group, i.e., the first sample... and the first sample When the first sample tag value Larger than the first sample tag value At that time, the first sample For the first sample loss It can be calculated using the following formula (1).

[0067] (1)

[0068] in, This can indicate taking a larger value. It can represent preset parameters , It can represent the first sample The predicted value, It can represent the first sample The predicted value.

[0069] Then, for the intra-batch sorting loss value of the current batch... It can be calculated using equation (2).

[0070] (2)

[0071] in, This can represent the set of predicted values ​​obtained by the bone age prediction model through the current training batch. It can represent the first bone age label value corresponding to the current training batch; ; It can represent an indicator function, used to determine... The output should check if the condition is met. If the condition is met, output 1; otherwise, output 0. This can represent the number of elements in the first batch of the current training batch. This can represent the ranking loss.

[0072] According to embodiments of this disclosure, by Regularization calculates the first regression loss value of the current training batch. It can be shown in equation (3).

[0073] (3)

[0074] After obtaining the intra-batch ranking loss and the first regression loss for the current training batch, they can be weighted and summed to obtain the first loss value. As shown in equation (4).

[0075] (4)

[0076] in, It can represent the loss weight.

[0077] According to embodiments of this disclosure, when training a bone age prediction model using majority population samples, two loss values ​​are used to adjust the model parameters. First, to represent the correlation between multiple first samples within the same training batch, sample pairs that meet the order condition are selected, and their intra-batch loss ranking is calculated. The intra-batch ranking loss value is used to allow the bone age prediction model to learn the correlation between hand bone images of the first samples when the first bone age label values ​​have a certain correlation; second, using... The first regression loss value is calculated using regularization to gradually bring the predicted value closer to the labeled value. This approach then combines the two loss values ​​mentioned above, avoiding scale collapse caused by learning only relative order and effectively suppressing overfitting, thus improving model reliability. Loss weights are used to adjust the relationship between the two loss values, and finally, the bone age prediction model is obtained by adjusting the parameters. This improves the robustness of bone age prediction, reduces dependence on large-scale labeled data, and better lays the foundation for subsequent knowledge transfer.

[0078] According to embodiments of this disclosure, training an absolute bone age calibration model using at least one training batch of second samples includes: inputting multiple second samples from the training batch into the bone age prediction model to obtain multiple second bone age prediction values; calculating a regression loss based on the second bone age label values ​​of each of the multiple second samples from the training batch and the multiple second bone age prediction values ​​to obtain a second loss value; and adjusting the parameters of the bone age prediction model using the second loss value.

[0079] According to embodiments of this disclosure, for a bone age prediction model trained using the first sample, the model is then trained using a minority group corresponding to the second sample. Specifically: first, multiple second samples from the current training batch of the minority group are input into the bone age prediction model to obtain multiple second bone age prediction values. Then, the second bone age label values ​​included in the two samples and the second bone age prediction values ​​are used to calculate the regression loss to obtain a second loss value. Finally, the parameters of the bone age prediction model are adjusted based on the second loss value.

[0080] According to embodiments of this disclosure, the current training batch of a minority group can be marked as ,in, To indicate a minority group, use Regularization is performed by calculating the second loss value using equation (5). .

[0081] (5)

[0082] in, This can represent the number of the second sample in the current training batch. It can represent the number of the second sample. It can represent the second bone age prediction value. It can represent the second bone age label value. This can represent performing training results. The result obtained from the calculation of the regularized loss value.

[0083] According to embodiments of this disclosure, a second sample is trained using L1 regularization based on a bone age prediction model trained on a first sample. This not only allows the knowledge learned from the first sample to be transferred to the second sample but also enables absolute calibration for the minority groups corresponding to the second sample. This approach allows the model to automatically select the features most influential on bone age prediction for minority groups during training, eliminating redundant and irrelevant features, thereby achieving sparse feature selection. This not only enhances the model's ability to focus on key hand bone features in the second sample and improves prediction accuracy but also effectively suppresses overfitting in minority group training scenarios, improving the model's stability and applicability.

[0084] According to an embodiment of this disclosure, in each training round, the number of training batches for the first sample is the number of the first batch, and the number of training batches for the second sample is the number of the second batch; the method further includes: determining the number of the first batch and the number of the second batch in the current training round based on the round number of the current training round.

[0085] According to embodiments of this disclosure, based on the above-described model training method, the number of batches used during training can vary depending on the training rounds. For example, in the early stages of training, the first sample can be used for training, so a larger number of batches are used to train the model at the beginning of training, and correspondingly, the number of the second batch can be smaller. As the number of training rounds increases, the number of the first batch can be gradually reduced and the number of the second batch can be increased, so that the training focus of the model gradually shifts towards the features of the second sample, thus training a bone age prediction model for the second sample.

[0086] According to embodiments of this disclosure, the quantity of the first batch decreases as the number of training rounds increases, while the quantity of the second batch increases as the number of training rounds increases.

[0087] According to embodiments of this disclosure, the bone age prediction model can be trained in parallel or alternately. Specifically, within each training round, training is performed several times with the first sample, followed by several times with the second sample. The number of training rounds can be gradually adjusted. The number of training rounds can be based on the number of training rounds. Dynamic adjustments are made to reduce overall target losses. It can be constructed as shown in equation (6).

[0088] (6)

[0089] in, For the first quantity, It can be larger in the early stages and decrease gradually in the later stages; This is the quantity for the second batch. This approach allows for smaller initial values ​​and progressively larger values ​​later, implementing a strategy of "learning a universal order first, then performing precise intra-domain calibration." When generating predictive outputs, a single second-sample hand bone image can be used. Output the predicted value, i.e. .in, These can represent the core parameters of a bone age prediction model. It can represent the prediction result.

[0090] According to embodiments of this disclosure, the method of dynamically adjusting the number of training batches with each training round allows the model to learn the general characteristics and patterns of most population groups at the beginning of training, thereby establishing a more universal foundation for bone age prediction. As training progresses, the number of batches for the first sample is gradually reduced while the number of batches for the second sample is increased, enabling the model to focus more on the specific characteristics and differences of minority populations, achieving accurate adaptation to the target population. This dynamic process improves the model's generalization ability and accuracy in predicting bone age for different populations, effectively avoids overfitting, and enhances the model's stability and robustness. Furthermore, this dynamic adjustment method optimizes training efficiency by rationally allocating training resources, allowing the model to achieve better performance within a limited training time, providing more reliable and accurate prediction results for bone age assessment.

[0091] Figure 4 A data flow diagram illustrating a training method for a bone age prediction model according to an embodiment of the present disclosure is shown schematically.

[0092] like Figure 4 As shown, the first sample can include hand bone image samples of humans in the first group and the corresponding first bone age label value. The hand bone image samples of humans in the first group can be input into the bone age prediction model to be trained to obtain the first bone age prediction value. Then, based on the first bone age prediction value and the first bone age label value, a loss value can be calculated. This scheme calculates two loss values ​​for the first sample: the intra-batch ranking loss value and the first regression loss value. The weighted sum of the intra-batch ranking loss value and the first regression loss value yields the first loss value. The first loss value is used to train the bone age prediction model to be trained, that is, to adjust the parameters of the bone age prediction model, resulting in the adjusted bone age prediction model.

[0093] The second sample can include hand bone image samples from the second group of people and corresponding second bone age label values. These hand bone image samples can be input into the adjusted bone age prediction model to obtain the second bone age prediction value. A second loss value is then calculated based on the second bone age prediction value and the second bone age label value. Using this second loss value to adjust the adjusted bone age prediction model yields the target bone age prediction model. The above model training is illustrated using a single epoch. In actual training, multiple epochs can be performed, and the number of training batches for multiple first and second samples can be adjusted within each epoch to obtain the target bone age prediction model.

[0094] According to an embodiment of this disclosure, a bone age prediction method includes: acquiring hand bone image data of a target object, wherein the target object belongs to a second group; and inputting the hand bone image data into a target bone age prediction model to obtain a bone age prediction result for the target object.

[0095] According to embodiments of this disclosure, after obtaining the target bone age prediction model, bone age can be predicted based on hand bone images of people belonging to the second group. Since the target bone age prediction model is adaptively trained for the second group, it can output more accurate bone age prediction values.

[0096] Figure 5 A block diagram of a training apparatus for a bone age prediction model according to an embodiment of the present disclosure is shown schematically.

[0097] like Figure 5 As shown, the training device 500 for the bone age prediction model includes a first acquisition module 510 and a training module 520.

[0098] The first acquisition module 510 is used to acquire multiple first samples and multiple second samples.

[0099] Training module 520 is used to utilize the relative age order knowledge contained in multiple first samples to assist multiple second samples in training the bone age prediction model, thereby obtaining a target bone age prediction model for the second group.

[0100] According to embodiments of this disclosure, training module 520 may include a first training submodule and a second training submodule.

[0101] The first training submodule is used to train the bone age prediction model for relative bone age order modeling using the first samples of at least one training batch in each training round.

[0102] The second training submodule is used to train the bone age prediction model for absolute bone age calibration using the second samples from at least one training batch.

[0103] According to embodiments of this disclosure, the first training submodule may include a first training unit, a second training unit, a third training unit, a fourth training unit, and a fifth training unit.

[0104] The first training unit is used to input multiple first samples from the training batch into the bone age prediction model to obtain multiple first bone age prediction values.

[0105] The second training unit is used to calculate the intra-batch ranking loss based on multiple first bone age prediction values, and obtain the intra-batch ranking loss value.

[0106] The third training unit is used to calculate the regression loss based on the first bone age label value and multiple first bone age prediction values ​​of each of the first samples in the training batch, and to obtain the first regression loss value.

[0107] The fourth training unit is used to perform a weighted summation of the intra-batch ranking loss value and the first regression loss value to obtain the first loss value.

[0108] The fifth training unit is used to adjust the parameters of the bone age prediction model using the first loss value.

[0109] According to embodiments of this disclosure, the second training unit may include a first training subunit, a second training subunit, and a third training subunit.

[0110] The first training subunit is used to determine multiple sample pairs from the multiple first samples in the training batch based on the relative order of the first bone age label values ​​of the multiple first samples in the training batch.

[0111] The second training subunit is used to calculate the ranking loss based on the two first bone age prediction values ​​associated with each sample pair, and obtain the ranking loss value.

[0112] The third training subunit is used to calculate the mean of the ranking loss values ​​of multiple samples to obtain the intra-batch ranking loss value.

[0113] According to embodiments of this disclosure, the second training submodule may include a sixth training unit, a seventh training unit, and an eighth training unit.

[0114] The sixth training unit is used to input multiple second samples from the training batch into the bone age prediction model to obtain multiple second bone age prediction values.

[0115] The seventh training unit is used to calculate the regression loss based on the second bone age label values ​​and multiple second bone age prediction values ​​of the second samples in the training batch, and obtain the second loss value.

[0116] The eighth training unit is used to adjust the parameters of the bone age prediction model using the second loss value.

[0117] According to embodiments of this disclosure, training module 520 may further include a third training submodule.

[0118] The third training submodule is used to determine the number of the first batch and the number of the second batch in the current training round based on the current training round.

[0119] Figure 6 A block diagram of a bone age prediction device according to an embodiment of the present disclosure is shown schematically.

[0120] like Figure 6 As shown, the bone age prediction device 600 includes a second acquisition module 610 and a determination module 620.

[0121] The second acquisition module 610 is used to acquire hand bone image data of the target object, wherein the target object belongs to the second group;

[0122] The determination module 620 is used to input hand bone image data into the target bone age prediction model to obtain the bone age prediction result of the target object.

[0123] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as hardware circuitry, such as a Field-Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a System-on-Chip, a System-on-a-Substrate, a System-on-Package, an Application-Specific Integrated Circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0124] For example, any plurality of the first acquisition module 510, training module 520, second acquisition module 610, and determination module 620 can be combined into one module / unit / subunit, or any one of these modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least part of the functionality of one or more of these modules / units / subunits can be combined with at least part of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of this disclosure, at least one of the first acquisition module 510, training module 520, second acquisition module 610, and determination module 620 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the first acquisition module 510, training module 520, second acquisition module 610, and determination module 620 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0125] It should be noted that the data processing system part in the embodiments of this disclosure corresponds to the data processing method part in the embodiments of this disclosure. The specific description of the data processing system part is referred to in the data processing method part, and will not be repeated here.

[0126] Figure 7 A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0127] like Figure 7 As shown, an electronic device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0128] RAM 703 stores various programs and data required for the operation of electronic device 700. Processor 701, ROM 702, and RAM 703 are interconnected via bus 704. Processor 701 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 702 and / or RAM 703. It should be noted that the programs may also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0129] According to embodiments of this disclosure, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the input / output (I / O) interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0130] According to embodiments of this disclosure, the method flow according to embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by processor 701, it performs the functions defined in the system of embodiments of this disclosure. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0131] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0132] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0133] For example, according to embodiments of this disclosure, a computer-readable storage medium may include the ROM 702 and / or RAM 703 described above and / or one or more memories other than ROM 702 and RAM 703.

[0134] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this disclosure. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the training method and bone age prediction method of the bone age prediction model provided in the embodiments of this disclosure.

[0135] When the computer program is executed by the processor 701, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0136] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 709, and / or installed from a removable medium 711. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0137] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on a user's computing device, partially on a user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0138] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways, even if such combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of this disclosure may be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0139] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A method for training a bone age prediction model, comprising: obtaining a plurality of first samples and a plurality of second samples, wherein the first samples comprise hand bone image samples of human bodies in a first population, the second samples comprise hand bone image samples of human bodies in a second population, and a population size of the first population is greater than a population size of the second population; and using relative age order knowledge contained in the plurality of first samples to assist the plurality of second samples in training a bone age prediction model, to obtain a target bone age prediction model for the second population.

2. The method of claim 1, wherein, The using relative age order knowledge contained in the plurality of first samples to assist the plurality of second samples in training a bone age prediction model comprises: in each training round, using first samples of at least one training batch to perform relative bone age order modeling training on the bone age prediction model; and using second samples of at least one training batch to perform absolute bone age calibration training on the bone age prediction model.

3. The method of claim 2, wherein, The using first samples of at least one training batch to perform relative bone age order modeling training on the bone age prediction model comprises: inputting the plurality of first samples of the training batch into the bone age prediction model respectively to obtain a plurality of first bone age prediction values; based on the plurality of first bone age prediction values, performing intra-batch ordering loss calculation to obtain an intra-batch ordering loss value; based on first bone age label values of the plurality of first samples of the training batch and the plurality of first bone age prediction values, performing regression loss calculation to obtain a first regression loss value; performing weighted summation on the intra-batch ordering loss value and the first regression loss value to obtain a first loss value; and using the first loss value to perform parameter adjustment on the bone age prediction model.

4. The method of claim 3, wherein, The based on the plurality of first bone age prediction values, performing intra-batch ordering loss calculation to obtain an intra-batch ordering loss value comprises: based on relative orders of first bone age label values of the plurality of first samples of the training batch, determining a plurality of sample pairs from the plurality of first samples of the training batch; based on two first bone age prediction values associated with each sample pair, performing ordering loss calculation to obtain an ordering loss value; and based on mean calculation of the ordering loss values of the plurality of sample pairs, obtaining the intra-batch ordering loss value.

5. The method of claim 2, wherein, The using second samples of at least one training batch to perform absolute bone age calibration training on the bone age prediction model comprises: inputting the plurality of second samples of the training batch into the bone age prediction model respectively to obtain a plurality of second bone age prediction values; based on second bone age label values of the plurality of second samples of the training batch and the plurality of second bone age prediction values, performing regression loss calculation to obtain a second loss value; and using the second loss value to perform parameter adjustment on the bone age prediction model.

6. The method of claim 2, wherein, In each training round, a number of training batches of the first samples is a first batch number, and a number of training batches of the second samples is a second batch number. The method further comprises: based on a round of a current training round, determining the first batch number and the second batch number of the current training round.

7. The method of claim 6, wherein, The first batch quantity decreases with the increase of training rounds, and the second batch quantity increases with the increase of training rounds.

8. A bone age prediction method, comprising: obtaining hand bone image data of a target object, wherein the target object belongs to a second group; and inputting the hand bone image data into a target bone age prediction model to obtain a bone age prediction result of the target object; wherein the target bone age prediction model is obtained by training the method according to any one of claims 1-7.

9. A training device for a bone age prediction model, comprising: a first obtaining module configured to obtain a plurality of first samples and a plurality of second samples; a training module configured to use the relative age order knowledge contained in the plurality of first samples to assist the plurality of second samples in training a bone age prediction model to obtain a target bone age prediction model for the second group.

10. A bone age prediction device, comprising: a second obtaining module configured to obtain hand bone image data of a target object, wherein the target object belongs to a second group; a determining module configured to input the hand bone image data into a target bone age prediction model to obtain a bone age prediction result of the target object.

Images