Bone age measurement method, system, electronic device and computer readable storage medium

Through deep learning models, ultrasound images are screened in quality and extracted ossification center areas, and multi-objective features are fusion combined with clinical information, which solves the problem of automated bone age measurement in ultrasound technology, realizes radiation-free and automated bone age measurement, and improves detection Accuracy and stability of results.

CN114429468BActive Publication Date: 2024-11-19TONGJI HOSPITAL ATTACHED TO TONGJI MEDICAL COLLEGE HUAZHONG SCI TECH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210097866.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-27
Publication Date
2024-11-19
Estimated Expiration
2042-01-27

AI Technical Summary

Technical Problem

The existing ultrasound technology has manual operation errors and no automated measurements are achieved in bone age measurement, which makes bone age assessment inconvenient for ultrasound physicians to use and promote.

Method used

The quality control model and multi-objective segmentation model were used to screen ultrasound images in quality and extract the ossification center area, combine clinical information to perform multi-objective feature fusion, and bone age measurement was achieved through XGBoost regression analysis.

Benefits of technology

Radiation-free and automated bone age measurement is achieved, the accuracy and stability of the test results are improved, the operation process is simplified, and it is suitable for ultrasonic doctors of all levels.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114429468B_ABST
    Figure CN114429468B_ABST
Patent Text Reader

Abstract

The present invention discloses a bone age measurement method, system, electronic device and computer-readable storage medium, the method comprising the following steps: collecting ultrasonic images and preprocessing them; using a trained quality control model based on deep learning to perform quality screening on the ultrasonic images; using a trained multi-target segmentation model based on deep learning to extract the ossification center area and cartilage area of ​​the ultrasonic images that have passed the quality screening; performing multi-target feature fusion on the ossification center area, cartilage area and clinical information, performing XGBoost regression analysis on the fused multi-target features, and obtaining bone age measurement values. The bone age measurement method and system of the present invention are non-invasive and radiation-free for the human body, have high safety, are convenient for detection, and have accurate detection results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of bone age measurement, and in particular to a bone age measurement method, system, electronic device and computer-readable storage medium. Background Art

[0002] Bone age is an indicator of the maturity of children's bones. The assessment of bone age and the difference between it and actual age play an important role in pediatric clinical work. Currently, the most commonly used method for bone age assessment is X-rays of the hands and wrists, but X-ray examinations have ionizing radiation, which is not conducive to short-term multiple examinations of children. Ultrasound is a non-invasive, non-ionizing, convenient, and economical imaging method, which has been widely used in clinical work.

[0003] Recent studies have shown that the use of conventional ultrasound to measure the ossification center of bones and the height of epiphyseal cartilage to obtain the ossification rate can be used to assess bone age. However, it has problems such as manual operation errors and the lack of automated measurement, which makes it inconvenient for ultrasound physicians to use and promote. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the above-mentioned background technology and provide a bone age measurement method, system, electronic device and computer-readable storage medium. The method is non-invasive and radiation-free to the human body, has high safety, is easy to detect, and has accurate test results.

[0005] To achieve the above object, the present invention provides a bone age measurement method, comprising the following steps:

[0006] Step S1: Acquire ultrasonic images and pre-process them;

[0007] Step S2: using the trained deep learning-based quality control model to perform quality screening on the ultrasound image;

[0008] Step S3: using the trained multi-objective segmentation model based on deep learning to extract the ossification center area and cartilage area from the ultrasound images that have passed the quality screening;

[0009] Step S4: performing multi-target feature fusion on the ossification center area, cartilage area and clinical information, performing XGBoost regression analysis on the fused multi-target features, and obtaining bone age measurement values.

[0010] Furthermore, in step S1, the ultrasound image is preprocessed, including:

[0011] 1) The collected ultrasound images are denoised using the spatial mean filtering method;

[0012] 2) Flip, scale, rotate and / or adjust the brightness of the denoised ultrasound image, and retain the ultrasound image after each type of operation to achieve the expansion of the number of ultrasound images.

[0013] Furthermore, in step S2, the training process of the quality control model based on deep learning includes:

[0014] Based on experience, a corresponding quality level reference value is marked on each frame of ultrasound image;

[0015] A quality control model based on a convolutional neural network structure is constructed, and the loss function is set to a classification cross loss function; the ultrasound image and its quality grade reference value are used as input to train the quality control model to obtain a quality control model training weight file;

[0016] Through the back propagation of the training process, the weight of each layer of the quality control model is continuously updated in a way that minimizes the loss function, so that the output of the quality control model iteratively approaches the true value, until the classification cross loss value between the predicted value of the forward propagation of the quality control model and the labeled quality grade reference value tends to converge, and an optimized quality control model weight file is obtained. The optimized quality control model weight file is used as the input of the quality control model.

[0017] Furthermore, in step S3, the training process of the multi-target segmentation model based on deep learning includes:

[0018] Based on experience, each frame of ultrasound image that has passed the quality screening is labeled with the ossification center reference area and the cartilage reference area;

[0019] A multi-objective segmentation model based on a convolutional neural network structure was built, and the loss function was set to the Dice coefficient. The ultrasound images that passed the quality screening and their ossification center reference area and cartilage reference area were used as input to train the multi-objective segmentation model, and the training weight file of the multi-objective segmentation model was obtained.

[0020] Through the back propagation of the training process, the weight of each layer of the multi-target segmentation model is continuously updated in a way that minimizes the loss function, so that the output of the multi-target segmentation model iteratively approaches the true value, until the Dice coefficient between the predicted segmentation result of the forward propagation of the multi-target segmentation model and the annotated ossification center reference area and cartilage reference area tends to converge, and an optimized multi-target segmentation model weight file is obtained; the optimized multi-target segmentation model weight file is used as the input of the multi-target segmentation model.

[0021] Furthermore, in step S4, multi-target feature fusion is performed on the ossification center region, cartilage region and clinical information, including:

[0022] respectively extracting high-throughput imaging features of the ossification center region and the cartilage region;

[0023] The image features are subjected to variance analysis to eliminate features with constant values, so as to eliminate irrelevant features; and then the redundant features are deleted by single factor significance analysis to obtain relevant image features after initial screening;

[0024] The clinical information of the examined object is fused with the relevant imaging features after the initial screening to obtain the fused multi-target features;

[0025] The clinical information includes at least gender, actual age and family history of genetic diseases.

[0026] Furthermore, the image features include at least one or more of intensity features, shape features, texture features and wavelet features.

[0027] Furthermore, in step S4, performing XGBoost regression analysis on the fused multi-target features to obtain bone age measurement values ​​includes:

[0028] The bone age reference value of each frame of ultrasound image is annotated based on experience, the bone age reference value and the fused multi-target features are input into the trained XGBoost prediction model, the fused multi-target features are fitted with the bone age reference value using regression analysis, and the bone age measurement value is obtained based on the fitting result.

[0029] The present invention also provides a system for implementing the above-mentioned bone age measurement method, comprising an image acquisition and preprocessing module, a quality control module, a multi-target segmentation module and a prediction module, wherein:

[0030] An image acquisition and preprocessing module, used for acquiring ultrasound images and preprocessing them;

[0031] A quality control module, used for performing quality screening on the ultrasound images using a trained quality control model based on deep learning;

[0032] The multi-objective segmentation module is used to extract the ossification center area and cartilage area from the ultrasound images that have passed the quality screening by using the trained multi-objective segmentation model based on deep learning;

[0033] The prediction module is used to perform multi-target feature fusion on the ossification center area, cartilage area and clinical information, and perform XGBoost regression analysis on the fused multi-target features to obtain bone age measurement values.

[0034] The present invention also provides an electronic device, comprising:

[0035] Memory for storing computer software programs;

[0036] The processor is used to read and execute the computer software program to implement the above-mentioned bone age measurement method.

[0037] The present invention also provides a computer-readable storage medium, in which a computer software program for implementing the above-mentioned bone age measurement method is stored.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] Firstly, the bone age measurement method and system provided by the present invention collects ultrasonic images of the radius of the wrist of the subject as the basis for detection, thus avoiding the radiation problem to the human body caused by traditional X-ray detection.

[0040] Secondly, the present invention performs filtering and image expansion through image preprocessing to improve the robustness of subsequent deep learning models; the quality of the collected ultrasound image samples is screened through the quality control module to improve the accuracy of subsequent detection and achieve rapid evaluation of the image; the information of the target area to be measured is extracted through the multi-target segmentation module, the image features of the target area to be measured are extracted and multi-target feature fusion is performed, and the fused multi-target features are regressed and analyzed through the prediction model to obtain the bone age measurement value.

[0041] Third, the present invention can realize the prediction of bone age based on ultrasound, which has the advantages of no radiation and automation. Through the design of multiple AI models, the stability and availability of the prediction results can be guaranteed, helping ultrasound doctors at all levels to measure bone age.

[0042] Fourthly, compared with traditional detection methods, the bone age measurement method of the present invention has the advantages of high safety, simple operation, and accurate measurement results. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flow chart of a bone age measurement method of the present invention;

[0044] Figure 2 This is a flow chart of ultrasound image preprocessing of the present invention;

[0045] Figure 3 is a flow chart of the quality control model training process of the present invention;

[0046] Figure 4 A schematic diagram of the classification of ultrasound image quality levels of the present invention;

[0047] Figure 5 It is a flow chart of the multi-objective segmentation model training process of the present invention;

[0048] Figure 6 It is a schematic diagram of the multi-target segmentation area of ​​an ultrasound image of the present invention;

[0049] Figure 7 is a flow chart of the prediction process of the prediction model of the present invention;

[0050] Figure 8 This is a block diagram of a bone age measurement system based on ultrasound imaging according to the present invention;

[0051] Fig. 9 A block diagram of an electronic device of the present invention;

[0052] Fig.10 The present invention is a block diagram of a computer-readable storage medium. DETAILED DESCRIPTION

[0053] The following is a detailed description of the implementation of the present invention in conjunction with the implementation cases, but they do not constitute a limitation of the present invention and are only given as examples. At the same time, the advantages of the present invention will become clearer and easier to understand through the description.

[0054] Figure 1 A flow chart of a bone age measurement method provided in this embodiment, such as Figure 1 As shown, the method comprises the following steps:

[0055] Acquire and pre-process ultrasound images;

[0056] Using a trained deep learning-based quality control model to perform quality screening on the ultrasound images;

[0057] The trained multi-objective segmentation model based on deep learning is used to extract the ossification center area and cartilage area from the ultrasound images that have passed the quality screening.

[0058] Multi-target feature fusion is performed on the ossification center area, cartilage area and clinical information, and XGBoost regression analysis is performed on the fused multi-target features and clinical information to obtain bone age measurement values.

[0059] It is understandable that, based on the defects in the background technology, an embodiment of the present invention proposes a method for measuring bone age. The purpose of the present invention is to use an artificial intelligence algorithm to analyze the ultrasonic image of the radius of the human wrist, and to extract the ossification information features of the radius in the ultrasonic image (such as the ossification center area and the cartilage area) by combining the convolutional neural network algorithm with the computer vision image processing algorithm, and to analyze the features by the XGBoost algorithm, so as to realize the automatic measurement of bone age. The present invention collects the ultrasonic image of the radius of the wrist of the subject as the basis for detection, avoiding the radiation problem to the human body caused by traditional X-ray detection; through the cascade design of multiple AI models, the stability and availability of the prediction results can be guaranteed, helping ultrasound doctors at all levels to measure bone age. Compared with traditional detection methods, the bone age measurement method of the present invention has the advantages of high safety, simple operation, and accurate measurement results.

[0060] In a possible embodiment, the ultrasound image is preprocessed, such as Figure 2 As shown, including:

[0061] The collected ultrasound images were denoised using the spatial mean filter method;

[0062] The denoised ultrasonic image is flipped, scaled, rotated and / or brightness adjusted, and the ultrasonic image after each type of operation is retained to achieve the expansion of the number of ultrasonic images.

[0063] It is understandable that ultrasound images are different from other images and have more shot noise. The spatial mean filter algorithm is used to process them. The filtering formula of the image under the (2n+1)*(2n+1) spatial window at the coordinate (x, y) is as follows:

[0064]

[0065] After filtering the ultrasound images, in order to improve the robustness of the subsequent model and simulate the actual inspection situation, the image quantity is amplified, for example, by the following method:

[0066] (1) Image flipping: In view of the differences in the direction of the patient's radius during examination, the image is flipped left and right. The number of images after flipping is doubled.

[0067] (2) Image scaling: The differences in magnification adjustment by different physicians during radial examination were evaluated to be approximately 0.1 times, so the images were scaled at ratios of 0.90, 0.95, 1.05, and 1.10, and the number of images was increased by 4 times.

[0068] (3) Image rotation: There are certain differences in the angles of the probes operated by physicians during radial examination. After evaluation, the differences are within 20 degrees. Therefore, the images are rotated at -20 degrees, -15 degrees, -10 degrees, -5 degrees, 5 degrees, 10 degrees, 15 degrees, and 20 degrees. After rotation, the number of images is expanded by 8 times.

[0069] (4) Image brightness adjustment: Based on the differences in machine gain adjustment by different physicians during radius examination, after evaluation, the brightness change was roughly around 0.1 times. Therefore, the image brightness was adjusted at a ratio of 0.90, 0.95, 1.05, and 1.10. After the brightness adjustment, the number of images was expanded by 4 times.

[0070] After the above-mentioned image expansion operation, the number of ultrasound image samples is greatly increased. The expanded sample set can be used as a training set for AI model training in subsequent steps, and can also be used as a test set to test the trained AI model.

[0071] In a possible embodiment, Figure 3The training process of the quality control model based on deep learning as shown includes:

[0072] (1) Based on experience, each frame of ultrasound image in the training set is labeled with a corresponding quality level reference value. Ultrasound images of different quality levels are Figure 4 shown.

[0073] It is understandable that before training the quality control model, it is necessary to set a quality grade reference value as a reference standard, so that the prediction results can be compared with it during the training process to determine the accuracy of the prediction results. In this embodiment, the ultrasound image acquisition equipment is first used to collect the radial ultrasound images, X-rays and clinical information of the normal population, and the clinical information includes at least information such as gender, actual age and family genetic history. The clinical information is reserved for subsequent steps. Then two senior and experienced physicians use TW3 and GP atlas methods to measure bone age according to the information displayed on the X-ray. When there is a difference in the measurement results, another authoritative expert measures the bone age to obtain the bone age reference value of the subject corresponding to the X-ray; the bone age reference value of the subject is annotated to the information of its ultrasound image, and this bone age reference value is reserved for the AI ​​model in the subsequent steps to predict bone age. Two senior ultrasound physicians then mark the ossification center reference area and cartilage reference area of ​​each frame of ultrasound image, and the marked ossification center reference area and cartilage reference area are reserved for subsequent multi-target segmentation model use. Two senior ultrasound physicians also evaluated the quality of each frame of ultrasound images. The evaluation results included three categories: high, medium, and low. The manual evaluation results of the quality of each frame of ultrasound images were used as the reference value of the quality level of ultrasound images, and each frame of ultrasound images was annotated. The reference value of the quality level of ultrasound images was used as a reference for training the quality control model.

[0074] (2) Build a quality control model based on a convolutional neural network structure, and set the loss function to a classification cross loss function; use the ultrasound image and its quality grade reference value as input to train the quality control model, and obtain a quality control model training weight file.

[0075] (3) Through the back propagation of the training process, the weight of each layer of the quality control model is continuously updated in a way that minimizes the loss function so that the output of the quality control model iteratively approaches the true value, until the classification cross-loss value between the predicted value of the forward propagation of the quality control model and the labeled quality grade reference value tends to converge, and an optimized quality control model weight file is obtained, and the optimized quality control model weight file is used as the input of the quality control model.

[0076] It is understandable that regarding the design of the quality control AI model, in order to prevent the model from overfitting, a lightweight convolutional neural network structure is designed. The steps are as follows: first repeat two sets of convolution modules - two-dimensional convolutional layer and Relu activation function, then build a two-dimensional pooling layer, then vectorize the features through the fully connected layer, and add a one-dimensional convolutional layer with L2 regularization constraint, and then add a one-dimensional pooling layer, and finally use the Softmax classifier to set the output results into three categories - high, medium and low image quality, so as to achieve the mapping of the image to high and low image quality through the convolutional neural network.

[0077] After building a quality control model based on a convolutional neural network structure, the quality control model needs to be trained. Specifically, in this embodiment, the initialization is set using a truncated normal distribution, the optimization function is set to the Adam optimizer, the loss function is set to the classification cross loss function, the accuracy is used as the evaluation function, the number of learning times is set to 1000, the training batch size is set to 16, the initial learning rate is set to 0.001, and the learning rate decay is set to 0.334. The input for quality control model training is a radial ultrasound image with a quality grade reference value annotated by a physician. The back propagation algorithm and the stochastic gradient descent method are used. According to the size of the classification cross loss value between the predicted value of the forward propagation and the annotated image quality grade reference value, the back propagation is iterated to update the weight of each layer until the classification cross loss value between the predicted value of the quality control model and the annotated image quality grade reference value tends to converge, and the training model is stopped to obtain the trained model.

[0078] After the quality control model training is completed, the trained model is saved, and a large number of images can be inferred end-to-end through the server, so as to achieve rapid evaluation of image quality, and evaluate the application efficiency of the quality control AI model through ROC curve analysis, accuracy, sensitivity, specificity and precision. The quality control model is applied to the image quality screening stage of ultrasound images for bone age detection to eliminate ultrasound images with low quality levels and screen out ultrasound images with medium and high quality levels to flow to the next detection node - the multi-target segmentation model.

[0079] In a possible embodiment, Figure 5 The training process of the multi-target segmentation model based on deep learning as shown includes:

[0080] (1) Based on experience, each frame of ultrasound image that has passed the quality screening is labeled with the ossification center reference area and the cartilage reference area. The labeled ossification center reference area and the cartilage reference area are as follows: Figure 6 shown.

[0081] It is understandable that the ossification center reference area and the cartilage reference area annotated by senior and experienced physicians can be used as standard values ​​in the multi-objective segmentation model, for example, as a reference basis for training the multi-objective segmentation model.

[0082] (2) Build a multi-target segmentation model based on a convolutional neural network structure and set the loss function as the Dice coefficient. Use the ultrasound images that have passed the quality screening and their ossification center reference area and cartilage reference area as input to train the multi-target segmentation model and obtain the multi-target segmentation model training weight file.

[0083] (3) Through the back propagation of the training process, the weight of each layer of the multi-objective segmentation model is continuously updated in a way that minimizes the loss function so that the output of the multi-objective segmentation model iteratively approaches the true value, until the Dice coefficient between the predicted segmentation result of the forward propagation of the multi-objective segmentation model and the labeled ossification center reference area and cartilage reference area tends to converge, thereby obtaining an optimized multi-objective segmentation model weight file; the optimized multi-objective segmentation model weight file is used as the input of the multi-objective segmentation model.

[0084] In this embodiment, the design of the multi-target segmentation model is to facilitate the fast and good segmentation of multiple targets of the radius in the ultrasound image (i.e., the ossification center area and the cartilage area), which is achieved by building a simplified U-Net convolutional neural network structure. Specifically, the second group of convolution modules is reduced in the encoder stage of U-Net, so that the downsampled semantic features can be obtained more quickly. Correspondingly, the third group of convolution modules is reduced in the decoder stage, so that the semantic features can be transformed into a segmentation result with the same size as the original image more quickly, and a two-dimensional depth-separable convolution layer is used instead of a two-dimensional convolution layer to reduce the convolution operations in the training and reasoning process, thereby improving the speed of the model.

[0085] After the multi-target segmentation model is built, it needs to be trained. Specifically, for the training of the multi-target segmentation model, the initialization is set using the truncated normal distribution method, the optimization function is set to the Adam optimizer, the loss function is set to the Dice coefficient, the average intersection-over-union ratio is used as the evaluation function, the number of learning times is set to 100, the training batch size is set to 4, and the initial learning rate is set to 0.0001. The model input is an image with medium and high image quality evaluated by the quality control model, and the two segmentation and annotation results of the ossification center reference area and the cartilage reference area of ​​the radial part marked by the physician are used as the target for training; the back propagation algorithm and the stochastic gradient descent method are used to iterate the back propagation and update the weights of each layer according to the Dice coefficient between the predicted segmentation result and the annotation result of the forward propagation, until the Dice coefficient between the predicted segmentation result and the annotation result of the model tends to converge, stop training the model, obtain the trained model, and save the trained model.

[0086] After the multi-objective segmentation model is trained, it can be applied to the ossification center and cartilage area extraction steps of bone age detection. The server can perform end-to-end reasoning on a large number of ultrasound images, thereby achieving rapid segmentation of the ossification center and cartilage area of ​​the radius in ultrasound images, and evaluating the application efficiency of the quality control segmentation model through intersection-over-union ratio, Dice coefficient, accuracy, etc. The segmented ossification center and cartilage area flow to the next detection node - the bone age prediction model.

[0087] In a possible embodiment, Figure 7 The multi-target feature fusion of the ossification center area and the cartilage area includes:

[0088] (1) According to the ISBI standard, high-throughput image features of the ossification center area and the cartilage area segmented in the previous step are respectively extracted, wherein the image features include at least one or more of intensity features, shape features, texture features and wavelet features.

[0089] (2) Using variance analysis to eliminate features with constant values ​​for the image features, so as to eliminate irrelevant features; then using univariate significance analysis to eliminate features with a P value greater than 0.05, thereby eliminating redundant features and obtaining relevant image features after initial screening;

[0090] (3) The clinical information such as the gender, actual age and family genetic history of the examined person is fused with the relevant imaging features after the initial screening to obtain the fused multi-target features.

[0091] It is understandable that the fused multi-objective features will be used as the input of the prediction model to evaluate the final bone age measurement value.

[0092] In a possible embodiment, Figure 7 As shown, XGBoost regression analysis is performed on the fused multi-objective features and clinical information to obtain bone age measurements, including:

[0093] As mentioned above, senior and experienced physicians mark the bone age reference value of each frame of ultrasound image based on experience, input the bone age reference value and the fused multi-target features into the trained XGBoost prediction model, and use regression analysis to fit the fused multi-target features and the bone age reference value, and obtain the bone age measurement value based on the fitting result. The architecture of the XGBoost prediction model is prior art and will not be described in detail in this patent.

[0094] like Figure 8 As shown, this embodiment also provides a bone age measurement system based on ultrasound imaging, including an image acquisition and preprocessing module, a quality control module, a multi-target segmentation module and a prediction module, wherein:

[0095] An image acquisition and preprocessing module, used for acquiring ultrasound images and preprocessing them;

[0096] A quality control module, used for performing quality screening on the ultrasound images using a trained quality control model based on deep learning;

[0097] The multi-objective segmentation module is used to extract the ossification center area and cartilage area from the ultrasound images that have passed the quality screening by using the trained multi-objective segmentation model based on deep learning;

[0098] The prediction module is used to perform multi-target feature fusion on the ossification center area, cartilage area and clinical information, and perform XGBoost regression analysis on the fused multi-target features and clinical information to obtain bone age measurement values.

[0099] It can be understood that the bone age measurement system based on ultrasound imaging provided by the present invention corresponds to the bone age measurement method provided in the aforementioned embodiments. The relevant technical features of the bone age measurement system based on ultrasound imaging can refer to the relevant technical features of the bone age measurement method, which will not be repeated here.

[0100] See also Fig. 9 , Fig. 9 Schematic diagram of an electronic device provided by an embodiment of the present invention. Fig. 9 As shown, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0101] Acquire and pre-process ultrasound images;

[0102] Using a trained deep learning-based quality control model to perform quality screening on the ultrasound images;

[0103] The trained multi-objective segmentation model based on deep learning is used to extract the ossification center information and cartilage area information from the ultrasound images that have passed the quality screening.

[0104] The ossification center information, cartilage area information and clinical information are subjected to multi-target feature fusion, and the fused multi-target features are subjected to XGBoost regression analysis to obtain bone age measurement values.

[0105] See also Fig.10 , Fig.10 Schematic diagram of an embodiment of a computer-readable storage medium provided by the present invention. Fig.10 As shown, this embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0106] Acquire and pre-process ultrasound images;

[0107] Using a trained deep learning-based quality control model to perform quality screening on the ultrasound images;

[0108] The trained multi-objective segmentation model based on deep learning is used to extract the ossification center information and cartilage area information from the ultrasound images that have passed the quality screening.

[0109] The ossification center information, cartilage area information and clinical information are subjected to multi-target feature fusion, and the fused multi-target features are subjected to XGBoost regression analysis to obtain bone age measurement values.

[0110] The bone age measurement method and system provided by the present invention collects ultrasonic images of the radius of the wrist of the subject as the basis for detection, avoiding the radiation problem to the human body caused by traditional X-ray detection. Filtering and image expansion are performed through image preprocessing to improve the robustness of each subsequent deep learning model; the quality of the collected ultrasonic image samples is screened by the quality control module to improve the accuracy of subsequent detection and achieve rapid evaluation of the image; the information of the target area to be measured is extracted by the multi-target segmentation module, the image features of the target area to be measured are extracted and multi-target feature fusion is performed, and the multi-target features after fusion are regressed and analyzed by the prediction model to obtain the bone age measurement value. Compared with the traditional detection method, the bone age measurement method of the present invention has the advantages of high safety, simple operation, and accurate measurement results.

[0111] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0112] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0113] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0114] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0115] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0116] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0117] The above are only specific embodiments of the present invention. It should be pointed out that any changes or substitutions that can be easily thought of by any technician familiar with the field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. The rest not described in detail are prior arts.

Claims

1. A method for measuring bone age, Features: The following steps are involved: Step S1: Acquire ultrasonic images and pre-process them; Step S2: using the trained quality control model based on deep learning to perform quality screening on the ultrasound image; wherein the training process of the quality control model based on deep learning includes: Based on experience, a corresponding quality level reference value is marked on each frame of ultrasound image; A quality control model based on a convolutional neural network structure is constructed, and the loss function is set to a classification cross loss function; the ultrasound image and its quality grade reference value are used as input to train the quality control model to obtain a quality control model training weight file; Through the back propagation of the training process, the weight of each layer of the quality control model is continuously updated in a way that minimizes the loss function so that the output of the quality control model iteratively approaches the true value, until the classification cross loss value between the predicted value of the forward propagation of the quality control model and the labeled quality grade reference value tends to converge, and an optimized quality control model weight file is obtained, and the optimized quality control model weight file is used as the input of the quality control model; Step S3: extracting the ossification center area and the cartilage area from the ultrasound images that have passed the quality screening using the trained multi-objective segmentation model based on deep learning; wherein the training process of the multi-objective segmentation model based on deep learning includes: Based on experience, each frame of ultrasound image that has passed the quality screening is labeled with the ossification center reference area and the cartilage reference area; A multi-objective segmentation model based on a convolutional neural network structure was built, and the loss function was set to the Dice coefficient. The ultrasound images that passed the quality screening and their ossification center reference area and cartilage reference area were used as input to train the multi-objective segmentation model, and the training weight file of the multi-objective segmentation model was obtained. Through the back propagation of the training process, the weight of each layer of the multi-objective segmentation model is continuously updated in a way that minimizes the loss function so that the output of the multi-objective segmentation model iteratively approaches the true value, until the Dice coefficient between the predicted segmentation result of the forward propagation of the multi-objective segmentation model and the annotated ossification center reference area and cartilage reference area tends to converge, and an optimized multi-objective segmentation model weight file is obtained; the optimized multi-objective segmentation model weight file is used as the input of the multi-objective segmentation model; Step S4: performing multi-target feature fusion on the ossification center area, cartilage area and clinical information, performing XGBoost regression analysis on the fused multi-target features, and obtaining bone age measurement values.

2. The bone age measurement method according to claim 1, Features: In step S1, the ultrasound image is preprocessed, including: 1) The collected ultrasound images are denoised using the spatial mean filtering method; 2) Flip, scale, rotate and / or adjust the brightness of the denoised ultrasound image, and retain the ultrasound image after each type of operation to achieve the expansion of the number of ultrasound images.

3. The bone age measurement method according to claim 1 or 2, Features: In the step S4, multi-target feature fusion is performed on the ossification center region, cartilage region and clinical information, including: respectively extracting high-throughput imaging features of the ossification center region and the cartilage region; The image features are subjected to variance analysis to eliminate features with constant values, so as to eliminate irrelevant features; and then the redundant features are deleted by single factor significance analysis to obtain relevant image features after initial screening; The clinical information of the examined object is fused with the relevant imaging features after the initial screening to obtain the fused multi-target features; The clinical information includes at least gender, actual age and family history of genetic diseases.

4. The bone age measurement method according to claim 3, Features: The image features include at least one or more of intensity features, shape features, texture features and wavelet features.

5. The bone age measurement method according to claim 3, Features: In step S4, the XGBoost regression analysis is performed on the fused multi-target features to obtain the bone age measurement value, including: The bone age reference value of each frame of ultrasound image is annotated based on experience, the bone age reference value and the fused multi-target features are input into the trained XGBoost prediction model, the fused multi-target features are fitted with the bone age reference value using regression analysis, and the bone age measurement value is obtained based on the fitting result.

6. A system for implementing the bone age measurement method according to any one of claims 1 to 5, Features: It includes image acquisition and preprocessing module, quality control module, multi-target segmentation module and prediction module, among which: An image acquisition and preprocessing module, used for acquiring ultrasound images and preprocessing them; A quality control module, used for performing quality screening on the ultrasound images using a trained quality control model based on deep learning; The multi-objective segmentation module is used to extract the ossification center area and cartilage area from the ultrasound images that have passed the quality screening by using the trained multi-objective segmentation model based on deep learning; The prediction module is used to perform multi-target feature fusion on the ossification center area, cartilage area and clinical information, and perform XGBoost regression analysis on the fused multi-target features to obtain bone age measurement values.

7. An electronic device, Features: include: Memory for storing computer software programs; A processor is used to read and execute the computer software program to implement the bone age measurement method described in any one of claims 1 to 5.

8. A computer-readable storage medium, Features: The computer-readable storage medium stores a computer software program for implementing the bone age measurement method described in any one of claims 1 to 5.