Bone age assessment method, device and computer-readable storage medium based on deep convolutional neural network
Through deep convolutional neural network, the wrist bone X-ray films are segmented and preprocessed, and the problem of subjectivity and error of manual interpretation in the existing bone age assessment methods is solved, and the efficiency and accuracy of automated bone age assessment is achieved. The error is 6 The accuracy rate within a month reaches 80%, and the accuracy rate within a year reaches 95%.
Patent Information
- Application Number
- CN202111451503.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-01
- Publication Date
- 2024-12-20
- Estimated Expiration
- 2041-12-01
AI Technical Summary
The existing bone age assessment methods rely on manual interpretation, which have problems such as large workload, long measurement cycle, strong subjectivity and large errors in fully automatic methods, which are difficult to meet clinical needs, especially the accuracy of the error control within 6 months. .
Using a deep convolutional neural network method, two deep convolutional neural networks are constructed through UNet and DenseNet algorithms, and image segmentation and preprocessing of wrist bone X-rays are performed, and the model is trained to achieve automated bone age assessment and improve the accuracy of the assessment results. .
The automatic derivation of bone age assessment was achieved, and the defects of missing features in the fully automatic segmentation method were overcome. The accuracy rate of error within 6 months reached 80%, and the accuracy rate of error within 1 year reached 95%, meeting the requirements of clinical application.
Smart Images
Figure CN114255211B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical image processing, and in particular to a bone age assessment method, device and computer-readable storage medium based on a deep convolutional neural network. Background Art
[0002] Bone age is referred to as bone age. Bone age is an objective indicator for assessing the degree of bone development in adolescents and children and measuring biological age. Bone age assessment plays an important role in the growth and development, disease diagnosis, forensic identification, and sports competition of adolescents and children. In terms of growth and development, bone age can give the physiological age of children and adolescents, estimate height, and assess whether the growth and development are normal. In terms of disease diagnosis, if the bone age is higher than the living age, it is an early development state. The doctor will consider whether the individual suffers from hyperthyroidism, adrenal hyperplasia and other diseases. On the contrary, it is a case of delayed individual development, and the doctor will consider whether the individual suffers from growth hormone deficiency, hypothyroidism and other diseases. In terms of forensic identification, bone age can determine the true age of the relevant personnel, which is conducive to judicial justice. In terms of sports competition, youth competitions have strict requirements on the age of participants, and bone age assessment helps to ensure the fairness of the competition.
[0003] The commonly used bone age assessment standards at present are the atlas method and the scoring method. In China, bone age is mainly assessed by taking bone age films, and then manually assessed by bone age experts. However, manual interpretation has the disadvantages of large workload, long measurement cycle, and strong subjectivity. Different doctors may give different results for the same bone age film, and the same doctor may have different judgments at different times. The existing number of doctors is not enough to meet the huge demand, and there is often a lack of sufficient measurement time. Therefore, it is urgent to establish a good automatic bone age assessment method.
[0004] The existing fully automatic methods are based on GP atlas and TW scoring methods, which require the extraction of wrist bones or characteristic bone blocks, and usually result in over-segmentation, which may miss some important features. The currently available fully automatic bone age assessment has a minimum error of less than 95% accuracy within one year, while clinically it is necessary to control the assessment error within 6 months.
[0005] Therefore, a new bone age assessment method based on deep convolutional neural network is proposed to improve the accuracy of bone age assessment. Summary of the invention
[0006] In order to overcome the above-mentioned technical defects, the purpose of the present invention is to provide a bone age assessment method, device and computer-readable storage medium based on deep convolutional neural network, so that the bone age assessment results can meet the requirements of clinical application.
[0007] The present invention discloses a bone age assessment method based on a deep convolutional neural network, comprising the following steps:
[0008] Wrist bone X-rays with calibrated bone age information were collected, and the first deep convolutional neural network was constructed based on the UNet algorithm to perform image segmentation on the wrist bone X-rays to remove the background information of the wrist bone X-rays to obtain a preprocessed image set, and some images in the preprocessed image set were selected as training set images, and the remaining images were selected as test set images;
[0009] Constructing a second deep convolutional neural network based on the DenseNet algorithm, and inputting the training set images into the second deep convolutional neural network to train parameters of the second deep convolutional neural network, thereby obtaining a trained second deep convolutional neural network, wherein the trained second deep convolutional neural network has learned network weight parameters;
[0010] Part or all of the test set images are input into the trained second deep convolutional neural network, and after iteration of the trained second deep convolutional neural network, the bone age assessment results of the wrist bone X-ray films of the subjects in the test set images are output.
[0011] Preferably, collecting wrist bone X-rays with calibrated bone age information, performing image segmentation on the wrist bone X-rays based on a first deep convolutional neural network to remove background information of the wrist bone X-rays to obtain a preprocessed image set, and selecting some images in the preprocessed image set as training set images and the remaining images as test set images include the following steps:
[0012] Wrist bone X-rays with bone age information labeled were collected from a radiology examination database;
[0013] A left hand frontal film with complete data is selected from the wrist bone X-ray film, and the left hand frontal film is preprocessed so that the brightness of the background area of the left hand frontal film is lower than the brightness of the wrist bone area;
[0014] According to the bone age information, wrist bone X-rays were categorized with a step length of m months, where 1≤m≤6;
[0015] Digitally sampling the wrist bone X-ray film so that the size of the wrist bone X-ray film is 512×512, and constructing a first deep convolutional neural network based on the UNet algorithm to perform image segmentation on the wrist bone X-ray film to remove background information of the wrist bone X-ray film to obtain a preprocessed image set, wherein the first deep convolutional neural network includes a feature extraction area, a sampling area, and a jump connection area, wherein the feature extraction area includes a plurality of alternating convolutional layers and pooling layers, and the sampling area includes a plurality of alternating convolutional layers and deconvolution layers;
[0016] 90% of the images are randomly selected from the preprocessed image set as training set images, data augmentation is performed on the training set images, the remaining 10% of the images are used as test set images, and some of the test set images are selected as wrist bone X-rays of the subjects.
[0017] Preferably, the step of classifying the wrist bone X-ray film in steps of m months according to the bone age information comprises:
[0018] The bone age difference is obtained by subtracting the minimum age in months from all wrist bone X-rays collected.
[0019] Divide the bone age difference by m to obtain categorical data.
[0020] Preferably, 90% of the images are randomly selected from the preprocessed image set as training set images, and the step of performing data augmentation on the training set images includes:
[0021] Perform any one or more of size transformation, displacement transformation, and angle rotation transformation on the training set images, so that the data volume of the training set images after augmentation is 4-8 times the data volume of the training set images before augmentation.
[0022] Preferably, a second deep convolutional neural network is constructed based on the DenseNet algorithm, and the training set image is input into the second deep convolutional neural network to train the parameters of the second deep convolutional neural network, thereby obtaining a trained second deep convolutional neural network, wherein the trained second deep convolutional neural network has learned network weight parameters. The steps include:
[0023] A second deep convolutional neural network is constructed based on the DenseNet algorithm. The second deep convolutional neural network includes alternating convolutional layers, pooling layers, and activation layers, and connections between different layers of the convolutional layers, pooling layers, and activation layers are established.
[0024] Inputting the training set images into the second deep convolutional neural network to train the parameters of the second deep convolutional neural network, so as to reduce the loss function value of the second deep convolutional neural network through the Adam optimization algorithm, and updating the network weight parameters of the second deep convolutional neural network;
[0025] After repeatedly training the parameters of the second deep convolutional neural network, the learned network weight parameters are obtained, wherein the training set images are classified according to the gender of the subjects in the bone age information.
[0026] Preferably, the step of inputting part or all of the test set images into the trained second deep convolutional neural network, and outputting the bone age assessment result of the wrist bone X-ray of the subject in the test set image after iteration of the trained second deep convolutional neural network includes:
[0027] Input part or all of the test set images into the trained second deep convolutional neural network, iterate the trained second deep convolutional neural network, and train the learned network weight parameters using the training set images classified based on the gender of the subject in the bone age information, and output category data;
[0028] The bone age assessment results of the subject's wrist bone X-ray are output based on the category data.
[0029] The present invention also discloses a bone age assessment device based on a deep convolutional neural network, comprising:
[0030] The processing module collects wrist bone X-rays that have been calibrated with bone age information, constructs a first deep convolutional neural network based on the UNet algorithm, performs image segmentation on the wrist bone X-rays to remove background information of the wrist bone X-rays to obtain a preprocessed image set, and selects some images in the preprocessed image set as training set images, and the remaining images as test set images;
[0031] A training module, constructing a second deep convolutional neural network based on a DenseNet algorithm, and inputting a training set image into the second deep convolutional neural network to train parameters of the second deep convolutional neural network, thereby obtaining a trained second deep convolutional neural network, wherein the trained second deep convolutional neural network has learned network weight parameters;
[0032] The assessment module inputs part or all of the test set images into the assessment module having a trained second deep convolutional neural network, and after iteration of the trained second deep convolutional neural network, outputs the bone age assessment result of the wrist bone X-ray of the subject in the test set images.
[0033] The present invention further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the bone age assessment method as described above are implemented.
[0034] Compared with the prior art, the above technical solution has the following beneficial effects:
[0035] 1. Based on two deep convolutional networks, the assessment results can be automatically exported, improving the efficiency of bone age assessment;
[0036] 2. A deep convolutional neural network is applied in the preprocessing process, which can overcome the defect of the existing fully automatic segmentation method that over-segments the wrist bone image and causes the loss of key features;
[0037] 3. The accuracy rate can reach 80% within 6 months of error and 95% within 1 year of error. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1A schematic diagram of a flow chart of a bone age assessment method in accordance with a preferred embodiment of the present invention;
[0039] Figure 2 This is an original image of a wrist bone X-ray film in accordance with a preferred embodiment of the present invention. DETAILED DESCRIPTION
[0040] The advantages of the present invention are further described below in conjunction with the accompanying drawings and specific embodiments.
[0041] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0042] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the disclosure. The singular forms of "a", "said" and "the" used in this disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0043] It should be understood that although the terms first, second, third, etc. may be used in the present disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present disclosure, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".
[0044] In the description of the present invention, it is necessary to understand that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.
[0045] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.
[0046] In the following description, the suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present invention, and have no specific meanings. Therefore, "module" and "component" can be used interchangeably.
[0047] See also Figure 1 , is a flow chart of a bone age assessment method based on a deep convolutional neural network in accordance with a preferred embodiment of the present invention. In this embodiment, the bone age assessment method includes the following steps:
[0048] S100: collect wrist bone X-rays that have been calibrated with bone age information, build a first deep convolutional neural network based on the UNet algorithm, perform image segmentation on the wrist bone X-rays to remove background information of the wrist bone X-rays to obtain a preprocessed image set, and select some images in the preprocessed image set as training set images, and the remaining images as test set images;
[0049] The network of the Unet algorithm consists of two parts. The first half is used for feature extraction (the basic model is MobileNet), and the second half is for upsampling.
[0050] Bone age information is wrist bone X-rays that have been calibrated with the subject's bone age, gender and other information. These wrist bone X-rays are collected as the basis for machine learning bone age information assessment. At the same time, the first deep convolutional neural network is used to segment and preprocess the wrist bone X-rays to deepen the characteristic parts of the wrist bone X-rays.
[0051] S200: constructing a second deep convolutional neural network based on the DenseNet algorithm, and inputting the training set image into the second deep convolutional neural network to train parameters of the second deep convolutional neural network, thereby obtaining a trained second deep convolutional neural network, wherein the trained second deep convolutional neural network has learned network weight parameters;
[0052] DenseNet is another convolutional neural network with deeper layers. It has the following advantages: (1) It has fewer parameters than ResNet; (2) The bypass enhances feature reuse; (3) The network is easier to train and has a certain regularization effect; (4) It alleviates the problems of gradient vanishing and model degradation.
[0053] Through the training of the second deep convolutional neural network, the learned network weight parameters can be obtained to serve as the evaluation criteria for subsequent wrist bone X-rays to be evaluated.
[0054] S300: Input part or all of the test set images into the trained second deep convolutional neural network, and after iteration of the trained second deep convolutional neural network, output the bone age assessment result of the wrist bone X-ray of the subject in the test set images.
[0055] Through the two established deep convolutional neural networks, the evaluation results can be exported fully automatically, which is convenient and fast. At the same time, after the pre-processed image is formed, the defect of the fully automatic segmentation method in the prior art that the wrist bone image is over-segmented and the key features are lost can be overcome.
[0056] In a preferred embodiment, step S100 specifically includes:
[0057] S110: collecting wrist bone X-rays labeled with bone age information from a radiology examination database;
[0058] Wrist bone X-ray data with calibrated bone age information can be collected from the radiological detection database of one or more hospitals, such as Figure 2 These wrist bone X-rays with calibrated bone age information may include, for example, 6,500 pieces of data on men and women of all ages.
[0059] S120: selecting a left hand frontal film with complete data from the wrist bone X-ray films, and preprocessing the left hand frontal film so that the brightness of the background area of the left hand frontal film is lower than the brightness of the wrist bone area;
[0060] When collecting wrist bone X-rays, both hands of the subject are collected. In order to save the subsequent data processing, the left hand AP film with complete data will be selected from all wrist bone X-rays (this can be achieved based on the position of the thumb). After preprocessing, the feature parts will be enhanced and uniformly processed into a result with brighter foreground and brighter background.
[0061] S130: classify wrist bone X-rays according to bone age information with a step length of m months, where 1≤m≤6;
[0062] For all wrist bone X-rays, the categories are marked according to the age given in the bone age information. For example, if the bone age information is F4Y6M, it means that the person is a female of 4 years and 6 months. It can be understood that the step length of m months determines the accuracy of the bone age assessment results. The smaller the m value, the more categories there are, and the more accurate the bone age assessment results will be, but the amount of data processing will also be greater.
[0063] S140: digitally sampling the wrist bone X-ray film so that the size of the wrist bone X-ray film is 512×512, and performing image segmentation on the wrist bone X-ray film based on a first deep convolutional neural network to remove background information of the wrist bone X-ray film to obtain a preprocessed image set, wherein the first deep convolutional neural network includes a convolution layer, a pooling layer, and an activation layer that are alternately operated;
[0064] To facilitate image processing, all wrist bone X-rays will be uniformly digitally sampled with a size of 512×512 (wrist bone features must not be removed), and unnecessary background information in the image will be removed through the first deep convolutional neural network segmentation.
[0065] S150: Randomly select 90% of the images from the preprocessed image set as training set images, perform data augmentation on the training set images, use the remaining 10% of the images as test set images, and select part of the test set images as wrist bone X-rays of the subject.
[0066] It is understandable that the more data the training set images have, the better the training results of the second deep convolutional neural network will be. The test set images are selected to verify the second deep convolutional neural network.
[0067] Furthermore, the above step S130 includes:
[0068] S131: subtract the minimum age in all wrist bone X-rays collected from the bone age information to obtain the bone age difference;
[0069] Still taking the above subject whose bone age information is marked as F4Y6M as an example, if all the data initially collected refers to a 2-18 year old adolescent, it means that the youngest month in all wrist bone X-rays is 24 months old, and the bone age information of the subject F4Y6M is 54 months minus 24 months, and the bone age difference is 30 months.
[0070] S132: Divide the bone age difference by m to obtain category data.
[0071] Take m=3 as an example, that is, all bone age information is classified every 3 months, and the above bone age difference is 30 months divided by 3, and the quotient is 10, which means that the category data of the subject labeled as F4Y6M is 10. The above processing method can, on the one hand, represent the evaluation results of the subsequent subjects in the form of category information, and output the results faster. On the other hand, the accuracy of the evaluation results can be controlled and maintained within the range of m.
[0072] Preferably, the above step S150 includes:
[0073] S151: Perform any one or more of size transformation, displacement transformation, and angle rotation transformation on the training set images, so that the data volume of the training set images after augmentation is 4-8 times the data volume of the training set images before augmentation.
[0074] The larger the amount of data in the training set images, the richer the learning foundation of the second deep convolutional neural network will be, and the more accurate the evaluation results of the subjects will be.
[0075] Preferably or optionally, step S200 includes:
[0076] S210: constructing a second deep convolutional neural network based on the DenseNet algorithm, wherein the second deep convolutional neural network includes convolutional layers, pooling layers, and activation layers that are alternately operated, and establishing connections between different layers of the convolutional layers, the pooling layers, and the activation layers;
[0077] S220: Inputting the training set image into the second deep convolutional neural network to train the parameters of the second deep convolutional neural network, so as to reduce the loss function value of the second deep convolutional neural network through the Adam optimization algorithm, and update the network weight parameters of the second deep convolutional neural network;
[0078] During training, the gender of the subject will be referenced for identification and classification.
[0079] S230: After repeatedly training the parameters of the second deep convolutional neural network, the learned network weight parameters are obtained, wherein the training set images are classified according to the gender of the subject in the bone age information
[0080] Through continuous iteration, the learned network weight parameters will serve as the basis for subsequent identification of wrist bone X-rays of the examinees. It is understandable that when there are new wrist bone X-rays of the examinees to be evaluated, research can be used instead of training. In other words, after the evaluation results of the examinees are output, the doctor can manually evaluate to form his own manual evaluation results, and input the manual evaluation results into the second deep convolutional neural network again as another iterative process. In other words, with the continuous use of the entire second deep convolutional neural network, it will continue to learn and evolve.
[0081] Preferably or optionally, step S300 includes:
[0082] S310: input part or all of the test set images into the trained second deep convolutional neural network, iterate the trained second deep convolutional neural network, and use the trained set images classified based on the gender of the subject in the bone age information to train the learned network weight parameters, and output category data;
[0083] When selecting the test set image, a data with a size of 512 × 512 can be randomly selected. The category data output in this step is the category value mentioned in the above embodiment.
[0084] S320: Output the bone age assessment result of the subject's wrist bone X-ray according to the category data.
[0085] For example, when the category value output is n, it can be converted to n*m and then added with the bone age difference to obtain the bone age assessment result of the subject.
[0086] The present invention also discloses a bone age assessment device based on a deep convolutional neural network, comprising: a processing module, which collects wrist bone X-rays that have been calibrated with bone age information, constructs a first deep convolutional neural network based on the UNet algorithm, performs image segmentation on the wrist bone X-rays to remove background information of the wrist bone X-rays and obtain a preprocessed image set, and selects some images in the preprocessed image set as training set images, and the remaining images as test set images; a training module, which constructs a second deep convolutional neural network based on the DenseNet algorithm, and inputs the training set images into the second deep convolutional neural network to train the parameters of the second deep convolutional neural network, thereby obtaining a trained second deep convolutional neural network, and the trained second deep convolutional neural network has learned network weight parameters; an assessment module, which inputs part or all of the test set images into an assessment module having the trained second deep convolutional neural network, and after iteration of the trained second deep convolutional neural network, outputs the bone age assessment results of the wrist bone X-rays of the subject in the test set images.
[0087] The present invention further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the bone age assessment method as described above are implemented.
[0088] It should be noted that the embodiments of the present invention have better practicability and do not impose any form of limitation on the present invention. Any technician familiar with the field may use the technical content disclosed above to change or modify it into an equivalent effective embodiment. However, any modification or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.
Claims
1. A bone age assessment method based on deep convolutional neural network, characterized in that: The steps include: Wrist bone X-ray films that have been calibrated with bone age information are collected, a first deep convolutional neural network is constructed based on the UNet algorithm, image segmentation is performed on the wrist bone X-ray films to remove background information of the wrist bone X-ray films to obtain a preprocessed image set, and some images in the preprocessed image set are randomly selected as training set images, and the remaining images are used as test set images; a second deep convolutional neural network is constructed based on the DenseNet algorithm, and the training set images are input into the second deep convolutional neural network to train parameters of the second deep convolutional neural network, thereby obtaining a trained second deep convolutional neural network, wherein the trained second deep convolutional neural network has learned network weight parameters; A randomly selected part of the images or all the images in the test set are input into the trained second deep convolutional neural network, and after iteration of the trained second deep convolutional neural network, the bone age assessment results of the wrist bone X-rays of the subjects in the test set images are output.
2. The bone age assessment method according to claim 1, characterized in that: The steps of collecting wrist bone X-ray films marked with bone age information, constructing a first deep convolutional neural network based on the UNet algorithm, performing image segmentation on the wrist bone X-ray films to remove background information of the wrist bone X-ray films to obtain a preprocessed image set, and selecting some images in the preprocessed image set as training set images and the remaining images as test set images include: Wrist bone X-rays with bone age information labeled were collected from a radiology examination database; Selecting a left hand frontal film with complete data from the wrist bone X-ray film, and preprocessing the left hand frontal film so that the brightness of the background area of the left hand frontal film is lower than the brightness of the wrist bone area; According to the bone age information, the wrist bone X-ray film is categorized with a step length of m months, where 1≤m≤6; Digitally sampling the wrist bone X-ray film so that the size of the wrist bone X-ray film is 512×512, and performing image segmentation on the wrist bone X-ray film based on a first deep convolutional neural network to remove background information of the wrist bone X-ray film to obtain a preprocessed image set, wherein the first deep convolutional neural network includes a feature extraction area, a sampling area, and a jump connection area, wherein the feature extraction area includes a plurality of alternating convolutional layers and pooling layers, and the sampling area includes a plurality of alternating convolutional layers and deconvolution layers; 90% of the images are randomly selected from the preprocessed image set as training set images, data augmentation is performed on the training set images, the remaining 10% of the images are used as test set images, and a portion of the test set images is selected as wrist bone X-rays of the subjects.
3. The bone age assessment method according to claim 2, characterized in that: The step of classifying the wrist bone X-ray film in steps of m months according to the bone age information comprises: The bone age information is subtracted from the minimum age in months in all wrist bone X-rays collected to obtain the bone age difference; Divide the bone age difference by m to obtain category data.
4. The bone age assessment method according to claim 2, characterized in that: 90% of the images are randomly selected from the pre-processed image set as training set images, and the step of performing data augmentation on the training set images comprises: Perform any one or more of size transformation, displacement transformation, and angle rotation transformation on the training set images, so that the data volume of the training set images after augmentation is 4-8 times the data volume of the training set images before augmentation.
5. The bone age assessment method according to claim 1, characterized in that: The steps of constructing a second deep convolutional neural network based on the DenseNet algorithm, and inputting the training set images into the second deep convolutional neural network to train the parameters of the second deep convolutional neural network, thereby obtaining a trained second deep convolutional neural network, wherein the trained second deep convolutional neural network has learned network weight parameters include: Constructing a second deep convolutional neural network based on the DenseNet algorithm, wherein the second deep convolutional neural network includes alternating convolutional layers, pooling layers, and activation layers, and establishing connections between different layers of the convolutional layers, pooling layers, and activation layers; inputting the training set images into the second deep convolutional neural network to train the parameters of the second deep convolutional neural network, so as to reduce the loss function value of the second deep convolutional neural network through the Adam optimization algorithm, and updating the network weight parameters of the second deep convolutional neural network; After repeatedly training the parameters of the second deep convolutional neural network, the learned network weight parameters are obtained, wherein the training set images are classified according to the gender of the subjects in the bone age information.
6. The bone age assessment method according to claim 1, characterized in that: The steps of inputting part or all of the test set images into the trained second deep convolutional neural network, and outputting the bone age assessment result of the wrist bone X-ray of the subject in the test set images after iteration of the trained second deep convolutional neural network include: Input part or all of the test set images into the trained second deep convolutional neural network, iterate the trained second deep convolutional neural network, and train the learned network weight parameters using the training set images classified based on the gender of the subject in the bone age information, and output category data; The bone age assessment result of the subject's wrist bone X-ray is output based on the said category data.
7. A bone age assessment device based on deep convolutional neural network, characterized in that: include: A processing module collects wrist bone X-rays that have been calibrated with bone age information, constructs a first deep convolutional neural network based on the UNet algorithm, performs image segmentation on the wrist bone X-rays to remove background information of the wrist bone X-rays to obtain a preprocessed image set, and randomly selects some images in the preprocessed image set as training set images, and the remaining images as test set images; A training module, constructing a second deep convolutional neural network based on a DenseNet algorithm, and inputting a training set image into the second deep convolutional neural network to train parameters of the second deep convolutional neural network, thereby obtaining a trained second deep convolutional neural network, wherein the trained second deep convolutional neural network has learned network weight parameters; The assessment module inputs some or all images randomly selected from the test set images into the assessment module having a trained second deep convolutional neural network, and after iteration of the trained second deep convolutional neural network, outputs the bone age assessment results of the wrist bone X-ray films of the subjects in the test set images.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the bone age assessment method as described in any one of claims 1-6 are implemented.