A child bone age image acquisition and positioning device and its detection method

Abstract

The present invention discloses a positioning device and a detection method for collecting bone age images of children, belonging to the technical field of bone age detection, specifically including: setting a standard inspection period, collecting bone age images of children every period and obtaining inspection records written by medical staff; these records are then input into a preset BERT model to extract bone age feature keywords; an auxiliary detection model is constructed using a convolutional neural network and a recurrent neural network and trained through a labeled historical data set; the child's age and keywords are converted into vectors and input into the trained auxiliary inspection model, and the bone age difference between the child's actual age and bone age is output; every n inspection periods, the system will statistically analyze the bone age difference sequence of the child, calculate the inspection period adjustment coefficient according to the index and through a specific formula, and then adjust the number of days of the next n inspection periods; the present invention improves the popularization rate and effectiveness of bone age detection.

Description

Technical Field

[0001] The present invention relates to the technical field of bone age detection, and particularly relates to a device and a detection method for collecting and positioning images of children's bone age. Background Art

[0002] In the medical field, especially in the pediatric field, the assessment of bone age is of great significance for judging the growth and development status of children. Traditional bone age assessment methods rely on doctors' observation and experience judgment of wrist X-ray films. This method is not only time-consuming and laborious, but also greatly affected by doctors' subjective factors, and it is difficult to guarantee the accuracy. With the development of deep learning and artificial intelligence technologies, bone age assessment methods based on image data have gradually emerged. These methods automatically assess bone age by analyzing X-ray images of children's wrists. Although the efficiency and objectivity of the assessment have been improved, new problems have also arisen.

[0003] Firstly, the existing bone age examinations are either that parents actively plan the examination time for children, or they passively follow the doctor's instructions for subsequent examinations. They cannot scientifically plan the examination time according to the actual bone age characteristics of children. Moreover, bone age assessment methods based on image data require a large amount of labeled data for training, and the acquisition of these data often requires professional medical equipment and technical personnel, with high difficulty. Secondly, image data processing and analysis require high computing resources, especially when processing high-resolution X-ray images, the computing power cost increases significantly. In addition, the quality of image data directly affects the accuracy of the assessment, and the acquisition of high-quality images further increases the cost.

[0004] Therefore, how to reduce the cost of bone age assessment while ensuring the assessment accuracy has become an important technical problem. The present invention proposes a method for judging through the text of bone age examination reports, aiming to utilize the existing medical record text information and combine it with a deep learning model to achieve efficient and low-cost bone age assessment. This method can not only reduce the dependence on expensive medical equipment, but also reduce the requirements for professional medical personnel, thereby improving the popularity and convenience of bone age assessment. Summary of the Invention

[0005] The purpose of the present invention is to provide a device and a detection method for collecting and positioning images of children's bone age, and solve the following technical problems:

[0006] The existing time interval of bone age examinations cannot be scientifically planned according to the actual situation of children, and bone age assessment methods based on image data require a large amount of labeled data for training, with high difficulty and high computing resources.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A method for collecting and detecting children's bone age images, comprising the following steps:

[0009] Set a standard inspection period d, where d represents the number of days. Collect children's bone age images every inspection period d, and obtain the inspection records written by medical staff based on the bone age images;

[0010] Input the inspection records into a preset BERT model to extract the keyword of the bone age feature;

[0011] Construct an auxiliary detection model based on a convolutional neural network and a recurrent neural network. Train the model with a labeled historical data set, extract the child's age and keywords from the inspection records, convert the child's age and keywords into vectors and input them into the trained auxiliary inspection model to output the bone age difference between the child's actual age and bone age;

[0012] Statistically analyze the bone age difference sequence of children every n inspection periods d, and calculate the mean value Y of the bone age difference sequence a , the proportion P of the bone age difference exceeding the preset threshold d in the sequence, and draw a line chart of the bone age difference changing with time to obtain the slope k between any two adjacent coordinate points; Calculate the inspection period adjustment coefficient γ according to the following formula:

[0013]

[0014] where i ∈ [1, n - 1], k i represents the i-th slope value in the line chart. According to the calculated inspection period adjustment coefficient γ, adjust the number of days of the next n inspection periods to d' = d / γ.

[0015] As a further solution of the present invention: Statistically analyze the indicators corresponding to the bone age difference sequence every n inspection periods, and the indicators include the mean value Y a , the proportion P and the slope k, and calculate the number of days of a single inspection period of the next n inspection periods according to the indicators and the formula.

[0016] As a further solution of the present invention: The process of extracting the keyword of the bone age feature is as follows:

[0017] Divide the labeled standard bone age feature dictionary into a training set and a validation set, map the training set into vectors and input them into the BERT model after encoding to extract the content features in the text. Update the BERT model through backpropagation using the validation set until convergence. Input the inspection records of the children to be tested into the BERT model after encoding. The inspection records include the actual age and bone age feature description text of the child. Extract the bone age features in the bone age feature description text, cluster to generate candidate keywords, and use the candidate keywords to update the model through backpropagation to obtain the candidate keywords after being updated repeatedly for several times, which are marked as keywords.

[0018] As a further solution of the present invention: the training process of the auxiliary diagnosis model is as follows:

[0019] The historical data set includes standard bone age feature texts corresponding to children of different ages, and non-standard bone age feature texts of children of different ages annotated in historical examination records. All bone age feature texts are sorted according to age to generate a time series of bone age features. The matrix formed by the time series of bone age features is regarded as an image and input into the convolutional neural network in sequence. The convolutional neural network includes several convolutional neural layers, and each convolutional neural layer includes several filters. Whenever the matrix image passes through a convolutional neural layer, the filter slides and scans the matrix and generates a multi-dimensional array, reducing the size of the matrix image and increasing the dimension of the matrix image. The increased dimension is the same as the number of filters in the convolutional neural network. Each convolutional neural layer reduces the image size in sequence and increases the dimension size of the matrix image. The multi-dimensional array finally generated by the convolutional neural network is flattened and converted into a vector, and this vector is used as the input of the subsequent recurrent neural network.

[0020] As a further solution of the present invention: the training process of the auxiliary diagnosis model further includes:

[0021] Keep the dimension of the matrix image output by the convolutional neural network unchanged, flatten the matrix image in the time dimension and the feature vector, generate a new matrix from the final multi-dimensional arrays of all convolutional neural layers, use the new matrix vector as the input of the recurrent neural network LSTM, use the flattened vector generated by flattening as the time dimension of the LSTM, and train the new matrix with the time dimension and the feature vector, and output the corresponding relationship between the bone age feature and the child's age.

[0022] As a further solution of the present invention: the recurrent neural network further includes several additional Transformer layers. All Transformer layers except the penultimate Transformer layer adopt the return sequence mode and generate a vector sequence corresponding to the continuous iteration of the recurrent neural network. The vector sequence is grouped according to the time series and used as the input of the penultimate Transformer layer.

[0023] As a further solution of the present invention: the examination records of the currently examined children are also stored in the historical data set.

[0024] The present invention also includes a device for collecting and positioning children's bone age images for implementing the above-mentioned method for collecting and detecting children's bone age images, which is characterized in that it includes:

[0025] A memory for setting a standard inspection period d, where d represents the number of days, and collecting the bone age images of children every inspection period d to obtain the inspection records written by medical staff based on the bone age images;

[0026] A text processor for inputting the inspection record into a preset BERT model to extract the keywords of bone age features;

[0027] A graphics processor for constructing an auxiliary detection model based on a convolutional neural network and a recurrent neural network, training the model through a labeled historical data set, extracting the child's age and keywords from the inspection record, converting the child's age and keywords into vectors and inputting them into the trained auxiliary inspection model to output the bone age difference between the actual age and bone age of the child;

[0028] A central processing unit for statistically analyzing the bone age difference sequence of children every n inspection periods d and calculating the mean Y of the bone age difference sequence a , the proportion P of the bone age difference exceeding the preset threshold d in the sequence, and plotting a line chart of the bone age difference changing with time to obtain the slope k between any two adjacent coordinate points; calculating the inspection period adjustment coefficient γ according to the following formula:

[0029]

[0030] where i ∈ [1, n - 1], k i represents the i-th slope value in the line chart, and adjusts the number of days of the next n inspection periods to d' = d / γ according to the calculated inspection period adjustment coefficient γ.

[0031] Advantages of the present invention:

[0032] The present invention provides a method for collecting and detecting children's bone age images. By setting a standard inspection period, regularly collecting children's bone age images and obtaining the inspection records written by medical staff. Using the BERT model to extract the keywords of bone age features, combining a convolutional neural network and a recurrent neural network to construct an auxiliary detection model, and training the historical data set to accurately output the bone age difference between the actual age and bone age of children. This method dynamically adjusts the inspection period by statistically analyzing the mean, proportion, and slope of the bone age difference sequence, improving the evaluation efficiency and accuracy. And by analyzing the inspection record text, it reduces the dependence on high-computing-power devices, reduces costs, and improves the popularity and effectiveness of bone age detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will be further described below with reference to the accompanying drawings.

[0034] Figure 1 is a schematic flowchart of a method for collecting and detecting children's bone age images according to the present invention. Specific Embodiments

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the protection scope of the present invention.

[0036] Please refer to Figure 1 As shown, the present invention is a method for collecting and detecting children's bone age images, including the following steps:

[0037] Set the inspection cycle: First, set a standard inspection cycle d (in days). Every this inspection cycle, collect bone age images of children once. These images are then analyzed by professional medical staff and detailed inspection records are written.

[0038] Text feature extraction: Input the inspection records written by medical staff into a preset BERT model. The BERT model is an advanced natural language processing model that can deeply understand the semantic information of the text. Through the BERT model, we can extract keywords related to bone age features from the inspection records, and these keywords are crucial for subsequent bone age assessment.

[0039] Construct an auxiliary detection model: Based on the convolutional neural network (CNN) and the recurrent neural network (RNN), we construct an auxiliary detection model. This model is trained using the labeled historical data set to learn how to extract keywords of children's age and bone age features from the inspection records and convert this information into vector form. These vectors are then input into the trained auxiliary detection model, and the model outputs the difference between the actual age and bone age of the child.

[0040] Dynamically adjust the inspection cycle: Every n inspection cycles, we statistically analyze the bone age difference sequence of children. Specifically, we calculate the mean value Ya of the bone age difference sequence, and the proportion P of the bone age differences exceeding the preset threshold d in the sequence. At the same time, we also draw a line chart of the bone age difference changing with time and obtain the slope k between any two adjacent coordinate points through calculation. Based on these statistical data and slope information, we use a specific formula to calculate the adjustment coefficient γ of the inspection cycle. This adjustment coefficient is used to dynamically adjust the number of days of the next n inspection cycles, and the formula is:

[0041]

[0042] where i ∈ [1, n - 1], k iDenote the \(i\)-th slope value in the line chart. According to the calculated inspection period adjustment coefficient \(\gamma\), adjust the number of days in the next \(n\) inspection periods to \(d' = d / \gamma\).

[0043] In a preferred embodiment of the present invention, the indicators of the corresponding bone age difference sequence are statistically analyzed every \(n\) inspection periods. The indicators include the mean value \(Y\) a , the proportion \(P\), and the slope \(k\). Calculate the number of days in a single inspection period in the next \(n\) inspection periods according to the indicators and the formula.

[0044] In another preferred embodiment of the present invention, the process of extracting the keywords of bone age features is as follows:

[0045] 1. Data preparation: First, we prepare a labeled standard bone age feature dictionary, which contains a rich vocabulary of bone age features and their corresponding annotation information. We divide this dictionary into two parts: the training set and the validation set. The training set is used to train the model, while the validation set is used to evaluate the performance of the model and make necessary adjustments.

[0046] 2. Vector mapping and encoding: Next, we map the data in the training set into vector form and perform encoding processing. These encoded vectors are then input into a pre-trained BERT model. The BERT model is a powerful natural language processing model that can deeply understand the semantic information of the text and extract useful features from it.

[0047] 3. Feature extraction and model training: In the BERT model, we train the model with the training set so that it learns how to extract content features related to bone age from the text. At the same time, we also use the validation set to evaluate the model and update the model through the backpropagation algorithm until the model converges, that is, reaches a stable performance state.

[0048] 4. Preprocessing of the data to be measured: After the model training is completed, we preprocess the inspection records of the children to be measured. These inspection records contain the actual age of the children and detailed text descriptions of bone age features. We perform the same encoding processing on the data to be measured and input it into the trained BERT model.

[0049] 5. Keyword extraction and clustering: In the BERT model, we deeply analyze the data to be measured and extract content features related to the text description of bone age features from it. These features are then used to generate tentative keywords. To further improve the accuracy and reliability of the keywords, we also perform clustering analysis on these tentative keywords to group similar or related keywords together.

[0050] 6. Model Iterative Update: Finally, we use the pending keywords generated by clustering to perform backpropagation updates on the BERT model. This is an iterative process where we perform multiple updates repeatedly until we obtain satisfactory keywords. These keywords that have been repeatedly updated and optimized are finally marked as keywords related to bone age characteristics and are used in the subsequent bone age assessment process.

[0051] In another preferred embodiment of the present invention, the training process of the auxiliary diagnosis model is as follows:

[0052] 1. Historical Dataset Preparation:

[0053] Collect the standard bone age characteristic texts corresponding to children of different ages, as well as the non-standard bone age characteristic texts of children of different ages annotated in the historical examination records. These data form the basis for our training model.

[0054] 2. Time Series Generation:

[0055] Sort all the bone age characteristic texts according to the age of the children to generate a time series of bone age characteristics. This time series reflects the changing trend of children's bone age with age growth.

[0056] 3. Matrix Image Construction:

[0057] Regard the matrix formed by the time series of bone age characteristics as a special "image" and input it into the convolutional neural network (CNN) in sequence. This representation method enables the CNN to capture the changing patterns of bone age characteristics in the time dimension.

[0058] 4. Convolutional Neural Network Processing:

[0059] The CNN includes several convolutional neural layers, and each convolutional neural layer contains multiple filters. Whenever the matrix image passes through a convolutional neural layer, the filters slide and scan the matrix to generate a multi-dimensional array. This process reduces the size of the matrix image while increasing the dimension of the matrix image, and the increased dimension is the same as the number of filters in this convolutional neural network.

[0060] Each convolutional neural layer sequentially reduces the image size and increases the dimension size of the matrix image. Finally, the multi-dimensional array generated by the CNN is flattened and converted into a vector as the input of the subsequent recurrent neural network (RNN).

[0061] 5. Recurrent Neural Network LSTM Training:

[0062] In the preferred embodiment, we keep the dimension of the matrix image output by the convolutional neural network unchanged and flatten the matrix image in the time dimension and the feature vector. In this way, the final multi-dimensional arrays of all convolutional neural layers generate a new matrix.

[0063] This new matrix vector serves as the input to the recurrent neural network LSTM. We flatten the resulting flattened vector as the time dimension of the LSTM, train the new matrix with the time dimension and the feature vector, and output the correspondence between the bone age feature and the child's age.

[0064] 6. Enhancement of Transformer layer:

[0065] In another preferred embodiment, we add several additional Transformer layers to the recurrent neural network. Except for the last Transformer layer, all Transformer layers adopt the return sequence mode and generate a sequence of vectors corresponding to the consecutive iterations of the recurrent neural network.

[0066] These sequences of vectors are grouped according to the time series and used as the input to the last Transformer layer. This structure further enhances the model's ability to process time series data and improves the accuracy and stability of bone age assessment.

[0067] It should be noted that the examination records of the currently examined child are also stored in the historical dataset.

[0068] The present invention also includes a device for collecting and positioning child bone age images for implementing the above-mentioned method for collecting and detecting child bone age images, which is characterized by comprising:

[0069] A memory for setting a standard examination period d, where d represents the number of days, collecting the child's bone age images every examination period d, and obtaining the examination records written by medical staff based on the bone age images;

[0070] A text processor for inputting the examination records into a preset BERT model to extract the keywords of the bone age features;

[0071] A graphics processor for constructing an auxiliary detection model based on a convolutional neural network and a recurrent neural network, training the model through a labeled historical dataset, extracting the child's age and keywords from the examination records, converting the child's age and keywords into vectors and inputting them into the trained auxiliary examination model, and outputting the bone age difference between the child's actual age and the bone age;

[0072] A central processing unit for statistically analyzing the sequence of bone age differences of the child every n examination periods d, calculating the mean Y of the sequence of bone age differences a , the proportion P of the bone age differences exceeding the preset threshold d in the sequence, and plotting a line graph of the bone age difference changing with time to obtain the slope k between any two adjacent coordinate points; calculating the examination period adjustment coefficient γ according to the following formula:

[0073]

[0074] where \(i\in[1,n - 1]\), \(k\) i represents the \(i\)-th slope value in the line graph. According to the calculated inspection period adjustment coefficient \(\gamma\), the number of days for the next \(n\) inspection periods is adjusted to \(d'=d / \gamma\).

[0075] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. Any equivalent changes and improvements made within the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.

Claims

1. A method for collecting and detecting children's bone age images, characterized in that, The following steps are involved: A standard inspection cycle d is set, where d represents the number of days, and a bone age image of the child is collected every inspection cycle d, and an inspection record written by a medical staff based on the bone age image is obtained; Inputting the inspection record into a preset BERT model to extract keywords of bone age characteristics; An auxiliary detection model is constructed based on a convolutional neural network and a recurrent neural network, the model is trained through annotated historical data sets, the child's age and keywords in the examination record are extracted, the child's age and keywords are converted into vectors and input into the trained auxiliary detection model, and the bone age difference between the child's actual age and bone age is output; Statistically analyze the bone age difference sequence of children every n inspection cycles d, and calculate the mean value Y of the bone age difference sequence a , the proportion P of the bone age difference exceeding the preset threshold d in the sequence, and draw a line chart of the bone age difference changing with time to obtain the slope k between any two adjacent coordinate points; The inspection cycle adjustment factor γ is calculated according to the following formula: ； where \(i\in[1,n - 1]\), \(k\) i represents the \(i\)-th slope value in the line graph. According to the calculated inspection period adjustment coefficient \(\gamma\), the number of days for the next \(n\) inspection periods is adjusted to \(d'=d / \gamma\); The training process of the auxiliary detection model is: The historical data set includes standard bone age feature texts corresponding to children of different ages, and non-standard bone age feature texts of children of different ages marked in historical examination records. All bone age feature texts are sorted according to age to generate a time series of bone age features. The matrix formed by the time series of bone age features is regarded as an image, and is sequentially input into the convolutional neural network. The convolutional neural network includes a plurality of convolutional neural layers, each of which includes a plurality of filters. Whenever a matrix image passes through a convolutional neural layer, the filter will slide and scan the matrix and generate a multidimensional array, reduce the size of the matrix image, and increase the dimension of the matrix image. The increased dimension is the same as the number of filters in the convolutional neural network. Each convolutional neural layer sequentially reduces the image size and increases the dimension size of the matrix image. The multidimensional array finally generated by the convolutional neural network is flattened and converted into a vector, and the vector is used as the input of a subsequent recurrent neural network. The training process of the auxiliary detection model also includes: The dimension of the matrix image output by the convolutional neural network remains unchanged, and the matrix image is flattened in the time dimension and eigenvector. The final multidimensional arrays of all convolutional neural layers generate a new matrix, and the new matrix vector is used as the input of the recurrent neural network LSTM. The flattened vector is used as the time dimension of the LSTM. The new matrix is trained by the time dimension and eigenvector, and the corresponding relationship between bone age characteristics and children's age is output.

2. The method for collecting and detecting children's bone age images according to claim 1, wherein Statistically analyze the indicators of the corresponding bone age difference sequence every n inspection cycles, and the indicators include the mean value Y a , the proportion P and the slope k, and calculate the number of days in a single inspection cycle for the next n inspection cycles according to the indicators and the formula.

3. A method for collecting and detecting children's bone age images according to claim 1, characterized in that, The process of extracting keywords for bone age characteristics is: The annotated standard bone age feature dictionary is divided into a training set and a validation set. The training set is mapped into a vector and input into the BERT model after encoding. The content features in the text are extracted. The BERT model is back-propagated and updated through the validation set until convergence. The examination records of the children to be tested are encoded and input into the BERT model. The examination records include the actual age of the child and a bone age feature description text. The bone age features in the bone age feature description text are extracted, clustered to generate pending keywords, and the pending keywords are used to back-propagate and update the model. The pending keywords are obtained after repeated updates for several times and marked as keywords.

4. The method for collecting and detecting children's bone age images according to claim 3, characterized in that, The recurrent neural network further includes a number of additional Transformer layers. All Transformer layers except the penultimate one adopt the return sequence mode and generate a sequence of vectors corresponding to the continuous iterations of the recurrent neural network. The sequence of vectors is grouped by time series and used as the input of the penultimate Transformer layer.

5. A method for collecting and detecting children's bone age images according to claim 1, characterized in that, The examination records of the currently examined child are also stored in the historical dataset.

6. A child bone age image acquisition and positioning device for implementing the child bone age image acquisition and detection method according to any one of claims 1-5, characterized in that, including: A memory for setting a standard examination period d, where d represents the number of days. The bone age images of children are collected every examination period d, and the examination records written by medical staff based on the bone age images are obtained; A text processor for inputting the examination records into a preset BERT model to extract the keywords of bone age features; A graphics processor for constructing an auxiliary detection model based on a convolutional neural network and a recurrent neural network, training the model through a labeled historical dataset, extracting the child's age and keywords from the examination records, converting the child's age and keywords into vectors and inputting them into the trained auxiliary detection model, and outputting the bone age difference between the actual age and the bone age of the child; A central processing unit, configured to statistically calculate a bone age difference sequence of a child every n inspection cycles d, and calculate an average value Y of the bone age difference sequence a , a proportion P of bone age differences exceeding a preset threshold d in the sequence, and plot a line graph of the bone age difference changing with time to obtain a slope k between any two adjacent coordinate points; Calculate the examination period adjustment coefficient γ according to the following formula: ； where \(i\in[1,n - 1]\), \(k\) i represents the \(i\)-th slope value in the line graph. According to the calculated inspection period adjustment coefficient \(\gamma\), adjust the number of days for the next \(n\) inspection periods to \(d'=d / \gamma\).

Images