Bone age evaluation method and device, bone age detection equipment and storage medium

Abstract

The invention provides a bone age evaluation method and device, bone age detection equipment and a storage medium. The bone age evaluation method comprises the following steps: acquiring an ultrasonic image sequence of a plurality of to-be-detected parts; inputting the real-time ultrasonic image of each to-be-detected part into the coarse positioning model, and outputting the score of each section ultrasonic image in real time; selecting a section ultrasonic image subsequence according to each section ultrasonic image and the corresponding score thereof, inputting the subsequence into a fine positioning model, and identifying a standard section ultrasonic image; inputting a standard section ultrasonic image into the segmentation model to obtain a cartilage region and an ossification center region; and calculating the bone maturity according to the maximum diameter of the cartilage area and the maximum diameter of the ossification center area. The section ultrasonic image is positioned and marked based on the coarse positioning model and the fine positioning model, the accuracy of the evaluation result can be improved, rapid, continuous and quantitative evaluation is achieved, and the accuracy of clinical evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bone age measurement and assessment, and in particular to a bone age assessment method, device, bone age detection equipment and storage medium. Background Art

[0002] Bone age plays an important role in the assessment of children's growth and development. It is an indicator that reflects the skeletal maturity of children. The most commonly used clinical evaluation method is based on X-ray measurement, but its evaluation is qualitative or semi-quantitative, and it is impossible to achieve quantitative evaluation of bone age. X-ray images are analyzed based on computer vision technology to achieve quantitative evaluation of bone age measurement, but this method has certain ionizing radiation and is not convenient for multiple evaluations in a short period of time. In related technologies, ultrasonic technology is used to determine bone age, mainly by ultrasonically measuring the thickness of the femoral head epiphysis to determine the bone age of children. However, the sensitivity of ultrasonic detection is low, and the quality of the acquired ultrasonic images is poor, resulting in inaccurate output results of the detection model. The existing ultrasonic bone age detection technology cannot be clinically promoted and applied. Summary of the invention

[0003] The present invention provides a bone age assessment method, device, bone age detection equipment and storage medium, which are used to solve the defects that traditional ultrasonic detection has low sensitivity, resulting in inaccurate output results of the detection model, and the existing ultrasonic bone age detection technology cannot be clinically promoted and applied.

[0004] The present invention provides a bone age assessment method, comprising: Acquire ultrasound image sequences of multiple parts to be tested; Input the real-time ultrasound image of each part to be tested into the coarse positioning model, and output the score of each section ultrasound image in real time; A section ultrasound image subsequence is selected according to each section ultrasound image and its corresponding score, and the section ultrasound image subsequence is input into a fine positioning model, wherein the fine positioning model identifies a standard section ultrasound image based on changes between different frames in the section ultrasound image subsequence; Inputting the standard section ultrasound image into the segmentation model to obtain the epiphyseal region and the ossification center region; The skeletal maturity is calculated based on the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region.

[0005] According to a bone age assessment method provided by the present invention, the selecting of a section ultrasound image subsequence according to each section ultrasound image and its corresponding score comprises: According to each section ultrasound image and its corresponding score, obtaining the section ultrasound image with the highest score; The highest-scoring section ultrasonic image and a continuous section ultrasonic image sequence within an adjacent preset time range are selected as a section ultrasonic image subsequence according to the output time of the highest-scoring section ultrasonic image.

[0006] According to a bone age assessment method provided by the present invention, the fine positioning model includes an image feature extraction part and a sequence feature extraction part, and the fine positioning model training method includes: Inputting the slice ultrasound image training data into the image feature extraction part to extract the image features of each frame of the slice ultrasound image in the slice ultrasound image training data; Inputting the image features and corresponding scores of the slice ultrasound image training data into the sequence feature extraction part to obtain the degree of proximity between each frame of the slice ultrasound image in the slice ultrasound image training data and the standard slice; Outputting a predicted standard section ultrasound image according to the proximity between each frame of the section ultrasound image and the standard section; Calculating the loss value of the predicted standard section ultrasound image and the labeled standard section ultrasound image using a first loss function; The fine positioning model weights are updated according to the loss value until the training end condition is met.

[0007] According to a bone age assessment method provided by the present invention, the coarse positioning model training method comprises: Perform standard section annotation on the ultrasound image to obtain the annotation score of the section ultrasound image; Inputting the slice ultrasound image training data into the coarse positioning model to obtain a prediction score for each slice ultrasound image; Calculating a loss value between the predicted score of the cross-sectional ultrasound image and the labeled score of the cross-sectional ultrasound image using a second loss function; The coarse positioning model weights are updated according to the loss value until the training end condition is met.

[0008] According to a bone age assessment method provided by the present invention, the parts to be measured include a left radius, a left ulna, and a left femur, and the segmentation model training method includes: Segmentation and annotation are performed on the standard section, and the annotation data includes the radial epiphysis area, the ulna epiphysis area, the femoral epiphysis area and their respective ossification center areas; The cross-sectional ultrasound image training data is input into the segmentation model to obtain the predicted radial epiphysis region, ulna epiphysis region, femoral epiphysis region and their respective ossification center regions; Calculate the loss value between the radial epiphysis region, the ulna epiphysis region, the femoral epiphysis region and their respective ossification center regions and the corresponding predicted radial epiphysis region, the ulna epiphysis region, the femoral epiphysis region and their respective ossification center regions in the labeled data using a third loss function; The segmentation model weights are updated according to the loss value until the training end condition is met.

[0009] According to a bone age assessment method provided by the present invention, the bone maturity is calculated according to the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region, including: The radius ossification rate is calculated based on the maximum diameter of the epiphyseal area and the maximum diameter of the ossification center area. , ulna ossification rate , femoral ossification rate ; According to the skeletal maturity formula , output the bone age assessment results.

[0010] According to a bone age assessment method provided by the present invention, after acquiring ultrasound image sequences of multiple parts to be measured, the method further includes: Preprocessing is performed on a plurality of ultrasound images of the part to be measured, wherein the preprocessing includes: at least one of image standardization, resampling, cropping, rotation, scaling and noise reduction.

[0011] The present invention also provides a bone age assessment device, comprising: An acquisition module, used for acquiring ultrasound image sequences of multiple parts to be tested; A coarse positioning module is used to input the real-time ultrasound image of each part to be tested into the coarse positioning model and output the score of each section ultrasound image in real time; A fine positioning module, configured to select a section ultrasound image subsequence according to each section ultrasound image and its corresponding score, and input the section ultrasound image subsequence into a fine positioning model, wherein the fine positioning model identifies a standard section ultrasound image based on changes between different frames in the section ultrasound image subsequence; A segmentation module, used for inputting the standard section ultrasound image into a segmentation model to obtain an epiphyseal region and an ossification center region; The calculation module is used to calculate the bone maturity according to the maximum diameter of the epiphyseal area and the maximum diameter of the ossification center area.

[0012] The present invention also provides a bone age detection device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above-described bone age assessment methods when executing the program.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the bone age assessment method described in any one of the above is implemented.

[0014] The present invention provides a bone age assessment method, device, bone age detection equipment and storage medium, which obtain ultrasound image sequences of multiple parts to be tested; input the real-time ultrasound image of each part to be tested into a coarse positioning model, and output the score of each section ultrasound image in real time; select a section ultrasound image subsequence according to each section ultrasound image and its corresponding score, input the section ultrasound image subsequence into a fine positioning model, and the fine positioning model identifies a standard section ultrasound image based on the changes between different frames in the section ultrasound image sequence; input the standard section ultrasound image into a segmentation model to obtain an epiphyseal region and an ossification center region; calculate the bone maturity based on the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region, and use ultrasound as a complete method for bone age assessment, which is non-invasive and radiation-free, obtains the section image score through the coarse positioning model, and then obtains the standard section ultrasound image based on the section ultrasound image changes between different frames in the section ultrasound image sequence according to the fine positioning model, thereby improving the accuracy of the assessment result, realizing rapid, continuous and quantitative assessment, being suitable for clinical use, and being conducive to improving the speed and accuracy of clinical assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 It is a schematic diagram of the flow chart of the bone age assessment method provided by the present invention; Figure 2 It is a structural schematic diagram of the bone age assessment device provided by the present invention; Figure 3 This is one of the structural schematic diagrams of the bone age detection device provided by the present invention; Figure 4 This is the second structural schematic diagram of the bone age detection device provided by the present invention. DETAILED DESCRIPTION

[0017] Figure 1 A flowchart of a bone age assessment method provided by an embodiment of the present invention, such as Figure 1 As shown, the bone age assessment method provided by the embodiment of the present invention includes: Step 101, obtaining ultrasound image sequences of multiple parts to be measured; Step 102: input the real-time ultrasound image of each part to be tested into the coarse positioning model, and output the score of each section ultrasound image in real time; Step 103: select a section ultrasound image subsequence according to each section ultrasound image and its corresponding score, input the section ultrasound image subsequence into a fine positioning model, and the fine positioning model identifies the standard section ultrasound image based on the changes between different frames in the section ultrasound image subsequence.

[0018] Step 104, inputting the standard section ultrasound image into the segmentation model to obtain the epiphyseal region and the ossification center region; Step 105: Calculate the bone maturity according to the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region.

[0019] Traditionally, ultrasound technology is used to determine bone age, and ultrasound is used to measure the thickness of the femoral head epiphysis to determine the bone age of children. However, the sensitivity of ultrasound detection is low, and the quality of the acquired ultrasound images is poor, resulting in inaccurate output results of the detection model. The existing ultrasound bone age detection technology cannot be promoted and applied clinically.

[0020] The present invention provides a bone age assessment method, which obtains ultrasound image sequences of multiple parts to be tested; inputs the real-time ultrasound image of each part to be tested into a coarse positioning model, and outputs the score of each section ultrasound image in real time; selects a section ultrasound image subsequence according to each section ultrasound image and its corresponding score, and inputs the section ultrasound image subsequence into a fine positioning model, and the fine positioning model identifies a standard section ultrasound image based on the changes between different frames in the section ultrasound image subsequence; inputs the standard section ultrasound image into a segmentation model to obtain an epiphyseal region and an ossification center region; and calculates bone maturity based on the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region. Ultrasound is used as a complete method for bone age assessment, which is non-invasive and radiation-free. The section image score is obtained through the coarse positioning model, and then the standard section ultrasound image is obtained based on the changes between different frames in the section ultrasound image sequence according to the fine positioning model, thereby improving the accuracy of the assessment result, realizing rapid, continuous and quantitative assessment, being suitable for clinical use, and being conducive to improving the speed and accuracy of clinical assessment.

[0021] Based on any of the above embodiments, the coarse positioning model training method includes: Step 201, annotate the ultrasound image with standard sections to obtain an annotation score for the section ultrasound image; Step 202: input the section ultrasound image training data into the coarse positioning model to obtain a prediction score for each section ultrasound image; Step 203: Calculate the loss value between the predicted score of the cross-sectional ultrasound image and the labeled score of the cross-sectional ultrasound image using the second loss function; Step 204: Update the coarse positioning model weights according to the loss value until the training end condition is met.

[0022] In an embodiment of the present invention, a double verification method is used for standard section annotation of ultrasound images. The annotation results are scored by two senior and experienced physicians according to the information displayed by the ultrasound images. When there are differences in the measurement results, the standard section annotation is verified by another one or more authoritative experts. In an embodiment of the present invention, the score reflects the confidence value of the section as a standard section. A real-time scoring model for ultrasound images based on deep learning is built, and a comprehensive loss function, such as a cross-entropy loss function, is set. The training process of the model uses the ultrasound image and the corresponding score as input to train the real-time scoring model to obtain a real-time scoring model training weight file. Through the back propagation of the training process, the weight of each layer of the real-time scoring model is continuously updated in a way that minimizes the loss function so that the score output by the model approaches the true value, until the loss value between the predicted value of the forward propagation of the real-time scoring model and the actual score reference value converges, thereby obtaining an optimized real-time scoring model weight file, and the optimized real-time scoring model weight file is used as the optimized parameter of the real-time scoring model.

[0023] Based on any of the above embodiments, the fine positioning model includes an image feature extraction part and a sequence feature extraction part, and the fine positioning model training method includes: Step 301: inputting the slice ultrasound image training data into the image feature extraction part, and extracting the image features of each frame of the slice ultrasound image in the slice ultrasound image training data; Step 302: input the image features of each frame of the ultrasonic slice image and its corresponding score into the sequence feature extraction part to obtain the degree of proximity between each frame of the ultrasonic slice image in the ultrasonic slice image training data and the standard slice; Step 303: outputting a predicted standard section ultrasound image according to the proximity between each frame of the section ultrasound image and the standard section; Step 304: Calculate the loss value of the predicted standard section ultrasound image and the labeled standard section ultrasound image using the first loss function; Step 305: Update the fine positioning model weights according to the loss value until the training end condition is met.

[0024] In an embodiment of the present invention, each frame of the slice ultrasound image in the slice ultrasound image sequence has a score, and these scores reflect the closeness of each slice to the standard slice; a neural network for processing and analyzing the image sequence is built, and the network includes two parts: an image feature extraction part and a sequence feature extraction part. Among them, the neural network of the image feature extraction part is responsible for extracting features from each frame of the image; the sequence feature extraction part adopts a sequence neural network (including but not limited to LSTM, GRU, integrated attention mechanism network, etc.). The model uses a comprehensive loss function (such as a cross entropy loss function) to optimize the training process. The network uses the ultrasound image sequence and its corresponding frame-by-frame score as input to obtain the corresponding training weight file; during the training process, the back propagation mechanism is used to continuously update the model weights, with the aim of minimizing the value of the loss function. This optimization process continues until the model shows convergence in the loss value between the prediction of different slices and the standard slice in the forward propagation, and obtains the optimized model weight file, which will be used as the key evidence for the model to judge and evaluate whether the slice in the image sequence is a standard slice.

[0025] Based on any of the above embodiments, the parts to be measured include a left radius, a left ulna, and a left femur, and the segmentation model training method includes: Step 401, segmenting and labeling on a standard section, wherein the labeling data includes the radial epiphysis region, the ulna epiphysis region, the femoral epiphysis region and the respective ossification center regions; Step 402, inputting the section ultrasound image training data into the segmentation model to obtain the predicted radial epiphysis region, ulna epiphysis region, femoral epiphysis region and their respective ossification center regions; Step 403, using a third loss function to calculate the loss value between the radial epiphysis region, the ulna epiphysis region, the femoral epiphysis region and their respective ossification center regions in the labeled data and the corresponding predicted radial epiphysis region, the ulna epiphysis region, the femoral epiphysis region and their respective ossification center regions; Step 404: Update the segmentation model weights according to the loss value until the training end condition is met.

[0026] In an embodiment of the present invention, after fine positioning to the standard section, based on medical experience, for example, the ossification center reference area and epiphysis reference area marked by senior and experienced physicians can be used as standard values ​​affecting the automatic segmentation model, and segmentation and annotation are performed on the standard section, including the epiphysis and ossification center of the left radius, left ulna, and left medial epicondyle of the femur. A convolutional neural network for image segmentation tasks is built, and a comprehensive loss function such as a cross entropy loss function is set. The training process of the model uses the standard section and the corresponding segmentation and annotation results as input to train the automatic cutting area model and obtain the corresponding training weight file; during the model training process, the model weights are continuously updated through the back propagation mechanism, and the goal is to minimize the loss function value. This optimization process is iteratively performed until the model shows obvious convergence in the loss value between the prediction of the epiphysis and ossification center cutting results and the actual segmentation annotation in the forward propagation, and an optimized model weight file is obtained. The automatic segmentation model weight file of the optimized model is used as the parameter of the automatic segmentation model in practical applications.

[0027] Based on any of the above embodiments, the bone age assessment method provided by the embodiment of the present invention includes: Step 501, obtaining ultrasound image sequences of multiple parts to be measured; Step 502: preprocessing ultrasound image sequences of multiple parts to be measured; In the embodiment of the present invention, the preprocessing includes but is not limited to image standardization, resampling, cropping, rotation, scaling and noise reduction to ensure the accuracy of subsequent input data.

[0028] Step 503: input the pre-processed ultrasound image sequence of each part to be measured into the coarse positioning model frame by frame, and output the score of each section ultrasound image in real time; Step 504: acquiring the highest-scoring section ultrasound image according to each section ultrasound image and its corresponding score; Step 505: Select the highest-scoring section ultrasound image and its adjacent continuous section ultrasound image sequence within a preset time range as a section ultrasound image subsequence according to the output time of the highest-scoring section ultrasound image.

[0029] Step 506: input the section ultrasound image subsequence into a fine positioning model, and the fine positioning model identifies a standard section ultrasound image based on changes between different frames in the section ultrasound image subsequence; Step 507, input the standard section ultrasound image into the segmentation model to obtain the epiphyseal region and the ossification center region; Step 508: Calculate the bone maturity according to the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region.

[0030] In an embodiment of the present invention, the bone maturity is calculated according to the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region, including: According to the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region

[0031] Radius ossification rate , ulna ossification rate , femoral ossification rate ; According to the skeletal maturity formula , output the bone age assessment results.

[0032] In the embodiment of the present invention, the diameters of the epiphyseal region (D) and the ossification center (d) in the standard section of the left radius, left ulna, and left femur are obtained according to the results output by the segmentation model. Calculate the ossification rate and use the formula , output the bone age assessment result. The bone age assessment result output by the embodiment of the present invention is compared with the gold standard X-ray bone age TW-3 method currently used in clinical practice. The evaluation results of the two are similar, the result deviation mean is small, the linear fit is high, and the sensitivity and specificity are extremely high (greater than 90-95%), indicating that the present invention has the potential to be used as a better new clinical examination method to replace traditional evaluation methods. In addition, the method is non-invasive and radiation-free, which reduces the error caused by the examiner's subjective identification, improves the accuracy of the result evaluation, and is helpful for grassroots and those who lack evaluation experience to carry out examinations, which is conducive to improving the speed and accuracy of clinical evaluation.

[0033] The bone age assessment method provided by the embodiment of the present invention collects ultrasonic sequence images of the three parts of the left radius, the left ulna, and the left medial epicondyle of the femur, and uses the trained rough positioning measurement model to respectively obtain the standard section ultrasonic images of the three parts of the left radius, the left ulna, and the left medial epicondyle of the femur; uses the trained fine measurement model to respectively obtain the standard section ultrasonic images of the three parts of the left radius, the left ulna, and the left medial epicondyle of the femur; uses the trained segmentation model to automatically segment the standard section images of the left radius, the left ulna, and the left medial epicondyle of the femur, and obtains the segmentation results of the epiphysis and the ossification center. The maximum diameter of the epiphysis area and the maximum diameter of the ossification center are automatically measured, and the ossification rates of the three parts of the left radius, the left ulna, and the left medial epicondyle of the femur are respectively calculated; the sum of the ossification rates of the three parts is calculated, and the final result of the ultrasonic bone age assessment is given, which can reduce the dependence of the bone age assessment on manual operation and improve the accuracy of the bone age assessment.

[0034] The bone age assessment device provided by the present invention is described below. The bone age assessment device described below and the bone age assessment method described above can be referenced to each other.

[0035] Figure 2 A schematic diagram of a bone age assessment device provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, the bone age assessment device provided by the embodiment of the present invention includes: An acquisition module 201 is used to acquire ultrasound image sequences of multiple parts to be measured; A coarse positioning module 202 is used to input the real-time ultrasound image of each part to be measured into the coarse positioning model and output the score of each section ultrasound image in real time; A fine positioning module 203 is used to select a section ultrasound image subsequence according to each section ultrasound image and its corresponding score, and input the section ultrasound image subsequence into a fine positioning model, wherein the fine positioning model identifies a standard section ultrasound image based on changes between different frames in the section ultrasound image sequence; The segmentation module 204 is used to input the standard section ultrasound image into the segmentation model to obtain the epiphyseal region and the ossification center region; The calculation module 205 is used to calculate the bone maturity according to the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region.

[0036] The present invention provides a bone age assessment device, which obtains ultrasound image sequences of multiple parts to be tested; inputs the real-time ultrasound image of each part to be tested into a coarse positioning model, and outputs the score of each section ultrasound image in real time; selects a section ultrasound image subsequence according to each section ultrasound image and its corresponding score, and inputs the section ultrasound image subsequence into a fine positioning model, wherein the fine positioning model identifies a standard section ultrasound image based on the section ultrasound image changes between different frames in the section ultrasound image subsequence; inputs the standard section ultrasound image into a segmentation model to obtain an epiphyseal region and an ossification center region; calculates bone maturity based on the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region, and uses ultrasound as a complete method for bone age assessment, which is non-invasive and radiation-free, obtains section image scores through a coarse positioning model, and then obtains a standard section ultrasound image based on the section ultrasound image changes between different frames in the section ultrasound image sequence according to the fine positioning model, thereby improving the accuracy of the assessment result, realizing rapid, continuous and quantitative assessment, being suitable for clinical use, and being conducive to improving the speed and accuracy of clinical assessment.

[0037] The embodiment of the present invention further provides a bone age detection device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the bone age assessment method described in any of the above embodiments when executing the program. Figure 3As shown, the bone age detection device is a semi-automatic bone age detection device. The semi-automatic bone age detection device includes: a main unit 10, a detection base 20, an ultrasonic imaging transducer 30, a deformable water bag 40, and a leveling base 50. When using the semi-automatic bone age detection device, first use the leveling base 50 to adjust the detection device to a horizontal level, and then apply a medical ultrasonic coupling agent to the part to be tested for acoustic coupling between human tissue and the deformable water bag 40, to reduce the reflection of ultrasonic waves at the interface between the deformable water bag 40 and human skin, so that more ultrasonic waves can enter the human body. Then, the part to be tested is placed on the deformable water bag 40, and the part to be tested is placed between the ultrasonic imaging transducer 30. The deformable water bag 40 can be conformed to the actual part to be tested and the shape of the different testers, so as to better achieve the sound conduction effect between the deformable water bag 40. Preferably, the deformable water bag 40 is bonded to the acoustic window opening of the ultrasonic imaging transducer 30 on the bottom 20 of the detection machine by waterproof glue, the deformable water bag 40 is filled with degassed water, and the ultrasonic imaging transducer 30 is immersed therein, and the deformable water bag 40 can be designed to be U-shaped so as to better fit the part to be tested, such as the wrist, knee, etc. The operator realizes ultrasonic imaging of the part to be tested by operating the host 10, and according to the needs, the patient is allowed to rotate the part to be tested respectively, so as to realize continuous ultrasonic imaging of the three parts of the left radius, the left ulna, and the left medial epicondyle of the femur. Then the obtained ultrasonic image sequence is transmitted to the host 10, and the ultrasonic image is processed by using artificial intelligence-assisted automatic section recognition, measurement and calculation methods, and the standard section ultrasonic image of the three parts of the left radius, the left ulna, and the left medial epicondyle of the femur is quickly and accurately positioned, without manual recognition and selection, and the bone maturity can be quickly and accurately obtained, which reduces the dependence on manual operation, avoids the error of manual operation, and improves the speed and accuracy of ultrasonic bone age assessment. Preferably, the artificial intelligence-assisted automated section recognition system can instantly display the credibility of the imaged section and the theoretical standard section image during the continuous imaging process of the patient rotating the measured part, which can further help inexperienced operators to obtain accurate images more accurately, while reducing the operational complexity of ultrasonic assessment of bone age.

[0038] like Figure 4As shown, the bone age detection device is a fully automatic bone age detection device. The fully automatic bone age detection device includes: a main unit 10, a detection base 20, an ultrasonic imaging transducer 30, a drive motor 40, a leveling base 50, an arched hard plastic shell 60, a silicone gasket 70, a deformable water bag 80, and an arched guide rail 90. When using the fully automatic bone age detection device, first use the leveling base 50 to adjust the detection device to a horizontal level, and then apply a medical ultrasonic coupling agent to the part to be tested for acoustic coupling between human tissue and the deformable water bag 80. Then, the part to be tested is placed on the silicone gasket 70 and fits tightly with the deformable water bag 80. The deformable water bag 80 can be conformally fitted according to the actual conditions and shapes of the parts to be tested of different testers, so as to better achieve the sound conduction effect between the deformable water bag 80. The deformable water bag 80 is bonded to the lower surface of the arched hard plastic shell 60 by waterproof glue. The deformable water bag 80 is filled with degassed water, and the ultrasonic imaging transducer 30 is immersed therein. The deformable water bag 80 ensures that the ultrasonic imaging transducer 30 can better conform to the part to be measured, such as the wrist, knee, etc. The operator realizes ultrasonic imaging of the measured part by operating the host 10, and can adjust the driving motor 40 to move back and forth along the arched guide rail 90 as needed, so as to realize continuous scanning and imaging of different bones in the measured part, thereby obtaining continuous ultrasonic images of the three parts of the left radius, the left ulna, and the left medial epicondyle of the femur, and then transmit the obtained ultrasonic image sequence to the host 10, and use artificial intelligence to assist in automatic section recognition, measurement and calculation methods to process the ultrasonic images, and quickly and accurately obtain standard section ultrasonic images of the three parts of the left radius, the left ulna, and the left medial epicondyle of the femur. Without manual identification and selection, the bone maturity can be quickly and accurately obtained, which reduces the dependence on manual operation, avoids the error of manual operation, and improves the speed and accuracy of ultrasonic assessment of bone age. Furthermore, the fully automatic ultrasound bone age assessment system does not require the tester to rotate the tested part according to the doctor's needs, does not rely on the operator's experience, does not require the operator to manually scan the image, and does not require the operator to identify the standard section of the selected part to be tested. It can achieve automatic, accurate, and rapid assessment of ultrasound bone age, improve the accuracy and speed of ultrasound bone age assessment, reduce manual operation errors, and reduce the operational complexity of ultrasound bone age assessment.

[0039] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the bone age assessment method provided by the above-mentioned methods, the method comprising: obtaining ultrasound image sequences of multiple parts to be measured; inputting the real-time ultrasound image of each part to be measured into a coarse positioning model, and outputting the score of each section ultrasound image in real time; selecting a section ultrasound image subsequence according to each section ultrasound image and its corresponding score, and inputting the section ultrasound image subsequence into a fine positioning model, the fine positioning model identifying a standard section ultrasound image based on the changes between different frames in the section ultrasound image sequence; inputting the standard section ultrasound image into a segmentation model to obtain an epiphyseal region and an ossification center region; and calculating the bone maturity based on the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region.

[0040] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0041] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing bone age, characterized in that: include: Acquire ultrasound image sequences of multiple parts to be tested; Input the real-time ultrasound image of each part to be tested into the coarse positioning model, and output the score of each section ultrasound image in real time; A section ultrasound image subsequence is selected according to each section ultrasound image and its corresponding score, and the section ultrasound image subsequence is input into a fine positioning model, wherein the fine positioning model identifies a standard section ultrasound image based on changes between different frames in the section ultrasound image subsequence; Inputting the standard section ultrasound image into the segmentation model to obtain the epiphyseal region and the ossification center region; The skeletal maturity is calculated based on the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region.

2. A bone age assessment method according to claim 1, characterized in that: The selecting a section ultrasound image subsequence according to each section ultrasound image and its corresponding score comprises: According to each section ultrasound image and its corresponding score, obtaining the section ultrasound image with the highest score; The highest-scoring section ultrasonic image and a continuous section ultrasonic image sequence within an adjacent preset time range are selected as a section ultrasonic image subsequence according to the output time of the highest-scoring section ultrasonic image.

3. A bone age assessment method according to claim 1, characterized in that: The fine positioning model includes an image feature extraction part and a sequence feature extraction part, and the fine positioning model training method includes: Inputting the slice ultrasound image training data into the image feature extraction part to extract the image features of each frame of the slice ultrasound image in the slice ultrasound image training data; Inputting the image features and corresponding scores of the slice ultrasound image training data into the sequence feature extraction part to obtain the degree of proximity between each frame of the slice ultrasound image in the slice ultrasound image training data and the standard slice; Outputting a predicted standard section ultrasound image according to the proximity between each frame of the section ultrasound image and the standard section; Calculating the loss value of the predicted standard section ultrasound image and the labeled standard section ultrasound image using a first loss function; The fine positioning model weights are updated according to the loss value until the training end condition is met.

4. A bone age assessment method according to claim 1, characterized in that: The coarse positioning model training method comprises: Perform standard section annotation on the ultrasound image to obtain the annotation score of the section ultrasound image; Inputting the slice ultrasound image training data into the coarse positioning model to obtain a prediction score for each slice ultrasound image; Calculating a loss value between the predicted score of the cross-sectional ultrasound image and the labeled score of the cross-sectional ultrasound image using a second loss function; The coarse positioning model weights are updated according to the loss value until the training end condition is met.

5. A bone age assessment method according to claim 1, characterized in that: The parts to be measured include a left radius, a left ulna, and a left femur, and the segmentation model training method includes: Segmentation and annotation are performed on the standard section, and the annotation data includes the radial epiphysis area, the ulna epiphysis area, the femoral epiphysis area and their respective ossification center areas; The cross-sectional ultrasound image training data is input into the segmentation model to obtain the predicted radial epiphysis region, ulna epiphysis region, femoral epiphysis region and their respective ossification center regions; Calculate the loss value between the radial epiphysis region, the ulna epiphysis region, the femoral epiphysis region and the ossification center region in the labeled data and the corresponding predicted radial epiphysis region, the ulna epiphysis region, the femoral epiphysis region and the ossification center region by using a third loss function; The segmentation model weights are updated according to the loss value until the training end condition is met.

6. A bone age assessment method according to claim 5, characterized in that: The step of calculating the skeletal maturity according to the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region comprises: According to the maximum diameter of the epiphyseal region and the maximum diameter of the ossification center region Radius ossification rate , ulna ossification rate , femoral ossification rate ; According to the skeletal maturity formula , output the bone age assessment results.

7. A bone age assessment method according to claim 1, characterized in that: After obtaining the ultrasound image sequences of the multiple parts to be measured, the method further includes: Preprocessing is performed on a plurality of ultrasound images of the part to be measured, wherein the preprocessing includes: at least one of image standardization, resampling, cropping, rotation, scaling and noise reduction.

8. A bone age assessment device, characterized in that: include: An acquisition module, used for acquiring ultrasound image sequences of multiple parts to be tested; A coarse positioning module is used to input the real-time ultrasound image of each part to be tested into the coarse positioning model and output the score of each section ultrasound image in real time; A fine positioning module, configured to select a section ultrasound image subsequence according to each section ultrasound image and its corresponding score, and input the section ultrasound image subsequence into a fine positioning model, wherein the fine positioning model obtains a standard section ultrasound image based on changes between different frames in the section ultrasound image subsequence; A segmentation module, used for inputting the standard section ultrasound image into a segmentation model to obtain an epiphyseal region and an ossification center region; The calculation module is used to calculate the bone maturity according to the maximum diameter of the epiphyseal area and the maximum diameter of the ossification center area.

9. A bone age detection device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the bone age assessment method as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the bone age assessment method according to any one of claims 1 to 7 is implemented.

Images