An image-assisted detection system for bone injuries based on data analysis

Abstract

The present invention discloses a bone injury image-assisted detection system based on data analysis, and relates to a bone injury image detection neighborhood. The bone injury image-assisted detection system based on data analysis includes: a data acquisition module for acquiring current bone image data, a region extraction module for performing region extraction on the preprocessed current bone image data to obtain current bone region data, a prediction module for obtaining predicted normal bone region data, a bone injury analysis module for performing feature extraction on the current and predicted normal bone region data respectively to obtain the texture, shape, and density features of the current and predicted normal bone regions, and performing comprehensive analysis to obtain the bone injury type corresponding to the current bone region data. The present invention uses accurate comparison of texture, shape, and density features, so that the system can not only identify the type of bone injury, but also finely analyze the severity of the injury. This in-depth analysis supports doctors to customize personalized treatment plans for each patient.

Description

[0001] Technology Neighborhood

[0002] The present invention relates to a bone injury imaging detection area, and in particular to a bone injury imaging auxiliary detection system based on data analysis. Background Art

[0003] Bone injury is one of the common clinical conditions, especially in cases of sports trauma, traffic accidents and falls of the elderly, fractures and other bone injuries occur frequently. Accurate and rapid diagnosis of the type and extent of bone injury is crucial for formulating appropriate treatment plans, improving treatment effects and shortening recovery time. Traditional bone imaging diagnosis mainly relies on the doctor's experience, and judges fractures, cracks and osteoporosis by observing images such as X-rays, CT and MRI. However, this method has certain subjectivity and limitations, and is easily affected by the doctor's personal experience and fatigue, resulting in inconsistent or incorrect diagnostic results. In recent years, with the rapid development of data analysis and artificial intelligence technology, a bone injury image-assisted detection system based on data analysis has emerged.

[0004] The limitations of existing technologies include at least the following problems. First, traditional methods often rely heavily on doctors' interpretation of images. Differences in experience and skill levels among different doctors may lead to inconsistent diagnostic results. Second, manual analysis of image data is time-consuming and inefficient, especially when large amounts of data need to be processed, such as in epidemiological studies or large clinical trials. Traditional imaging diagnostic methods are difficult to capture tiny fractures or early osteoporosis, especially when the image quality is poor, which can easily lead to misdiagnosis. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a bone injury image-assisted detection system based on data analysis, which solves the problem that traditional methods often rely heavily on doctors' interpretation of images, and differences in experience and skill levels of different doctors may lead to inconsistent diagnostic results. Secondly, manual analysis of image data is time-consuming and inefficient, especially when a large amount of data needs to be processed, such as in epidemiological studies or large clinical trials. In addition, traditional imaging diagnostic methods are difficult to capture tiny fractures or early osteoporosis, especially when the image quality is poor, which can easily lead to the occurrence of misjudgments.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a bone injury image-assisted detection system based on data analysis, comprising: a data acquisition module, a region extraction module, a prediction module, and a bone injury analysis module; the data acquisition module is used to acquire current bone image data and perform preprocessing; the region extraction module is used to perform region extraction on the preprocessed current bone image data to obtain current bone region data; the prediction module inputs the current bone region data into a pre-established prediction model for prediction analysis to obtain predicted normal bone region data; the bone injury analysis module is used to perform feature extraction on the current bone region data and the predicted normal bone region data, respectively, to obtain the texture, shape, and density features of the current bone region and the predicted normal bone region, and to perform comprehensive analysis to obtain the bone injury type corresponding to the current bone region data.

[0007] Furthermore, the current bone image data specifically refers to the pixel value and two-dimensional coordinates of each pixel point in the current bone image, the current bone region data specifically refers to the pixel value and two-dimensional coordinates of each actual bone pixel point, and the normal bone region data specifically refers to the pixel value and two-dimensional coordinates of each normal bone pixel point.

[0008] Furthermore, the specific steps of performing regional extraction on the preprocessed current bone image data to obtain the current bone region data are as follows: reading the pixel value and two-dimensional coordinates of each pixel point in the preprocessed current bone image, and detecting the initial edge pixel point set of the initial bone edge in the current bone image based on the edge detection algorithm, including several initial edge pixel points; using the regional growing algorithm to fill the initial bone edge to obtain the initial bone region pixel point set, including several bone pixel points within the edge; based on a preset pixel threshold, performing pixel division processing on the pixel value of each pixel point in the preprocessed current bone image to obtain the bone division pixel point set, including several bone division pixel points; merging the initial edge pixel point set, the initial bone region pixel point set, and the bone division pixel point set to obtain several actual bone pixel points, and the pixel value and two-dimensional coordinates corresponding to each actual bone pixel point.

[0009] Furthermore, the prediction model is specifically a conditional generative adversarial network, which includes a generator that receives damaged bone images as input and generates undamaged bone images as output, the generator includes convolutional layers, deconvolution layers and activation layers, a discriminator that distinguishes the generated images from real undamaged bone images, the discriminator includes convolutional layers, pooling layers and fully connected layers, a discriminative adversarial loss function, a generative adversarial loss function, and an optimizer.

[0010] Furthermore, the pre-establishment steps of the conditional generative adversarial network are as follows: obtain several groups of bone image data, each group of bone image data includes damaged bone images and corresponding undamaged normal bone images; pre-process each group of bone image data and divide it into an image training set and an image verification set; initialize the conditional generative adversarial network, specifically the weight initialization of the generator and the discriminator; based on the image training set, perform generator update and discriminator update processing on each training cycle in the set number of training cycles; and after each training cycle, perform evaluation analysis based on the image verification set, and adjust the parameters of the conditional generative adversarial network based on the evaluation analysis until the output result of the conditional generative adversarial network meets the expected standards.

[0011] Furthermore, the specific steps of performing feature extraction on the current bone region data and the predicted normal bone region data to obtain the texture features of the current bone region and the predicted normal bone region are as follows:

[0012] Grayscale processing is performed on the current bone region data and the predicted normal bone region data respectively, and key point extraction is performed on the current bone region data after the grayscale processing to obtain a number of current bone key grayscale pixel points of the current bone region data, and key point correspondence matching is performed on the predicted normal bone region data based on the several current bone key grayscale pixel points of the current bone region data to obtain a number of normal bone key grayscale pixel points of the predicted normal bone region data, and each current bone key grayscale pixel point corresponds to each normal bone key grayscale pixel point one by one; the number of actual bone grayscale pixel points within a set neighborhood of each current bone key grayscale pixel point of the current bone region data is counted, and based on the corresponding The grayscale pixel values ​​are analyzed separately to obtain the LBP value of each current bone key grayscale pixel point of the current bone region data, which is the texture feature of the current bone region; the number of normal bone grayscale pixels in the set neighborhood of each normal bone key grayscale pixel point of the predicted normal bone region data is counted, and the LBP value of each normal bone key grayscale pixel point of the predicted normal bone region data is obtained based on the corresponding grayscale pixel values, which is the texture feature of the predicted normal bone region; wherein, the specific formula for calculating the LBP value of each current bone key grayscale pixel point of the current bone region data and the LBP value of each normal bone key grayscale pixel point of the predicted normal bone region data is as follows: Among them, LBP i is the LBP value of the i-th current bone key grayscale pixel point in the current bone region data, Xs i is the grayscale pixel value of the i-th current bone key grayscale pixel point in the current bone region data, Xs iais the grayscale pixel value of the ath actual bone grayscale pixel in the set neighborhood of the i-th current bone key grayscale pixel in the current bone region data, s(x) is the sign function, a=1, 2, 3, ..., A, A is the number of actual bone grayscale pixels in the set neighborhood of the current bone key grayscale pixel in the current bone region data, LBP i ′ is the LBP value of the i-th normal bone key grayscale pixel point in the predicted normal bone area data, Xs′ i To predict the grayscale pixel value of the i-th normal bone key grayscale pixel point in the normal bone area data, Xs′ ib To predict the grayscale pixel value of the bth normal bone grayscale pixel point within the set neighborhood of the i-th normal bone key grayscale pixel point of the normal bone area data, b=1, 2, 3, ..., B, B is the number of normal bone grayscale pixels within the set neighborhood of the normal bone key grayscale pixel point for predicting the normal bone area data, i=1, 2, 3, ..., N, N is the number of current bone key grayscale pixels of the current bone area data, and is also the number of normal bone key grayscale pixels for predicting the normal bone area data.

[0013] Furthermore, the features of the current bone region data and the predicted normal bone region data are extracted respectively, and the specific steps of obtaining the shape features of the current bone region and the predicted normal bone region are as follows: read several current bone key grayscale pixel points of the current bone region data for closed connection to obtain the current bone polygon region, and analyze the perimeter value and area value of the current bone polygon region and the compactness index of the current bone region, which is the shape feature of the current bone region; read several normal bone key grayscale pixel points of the predicted normal bone region data for closed connection to obtain the normal bone polygon region, and analyze the perimeter value and area value of the normal bone polygon region and the compactness index of the normal bone region, which is the shape feature of the predicted normal bone region; wherein, the specific formula for calculating the perimeter value, area value, and compactness index of the current bone polygon region is as follows: Among them, Dzc is the perimeter value of the current bone polygon area, Dmj is the area value of the current bone polygon area, (x i ,y i ,) is the i-th current bone key grayscale pixel point of the current bone region data, (x i+1 ,y i+1 ) means that when i takes the last current bone key grayscale pixel point, it is reset to the coordinates of the first current bone key grayscale pixel point, Djc is the current bone area compactness index, π is a natural constant, i = 1, 2, 3, ..., N, N is the number of current bone key grayscale pixels in the current bone area data.

[0014] Furthermore, the specific steps for obtaining the density characteristics of the current bone region and the predicted normal bone region are as follows: read the number of actual bone grayscale pixels within the set neighborhood of each current bone key grayscale pixel of the current bone region data, and analyze them separately based on the corresponding grayscale pixel values ​​to obtain the average grayscale density of the neighborhood corresponding to each current bone key grayscale pixel of the current bone region data, which is the density characteristic of the current bone region; read the number of normal bone grayscale pixels within the set neighborhood of each normal bone key grayscale pixel of the predicted normal bone region data, and analyze them separately based on the corresponding grayscale pixel values ​​to obtain the average grayscale density of the neighborhood corresponding to each normal bone key grayscale pixel of the predicted normal bone region data, which is the density characteristic of the predicted normal bone region; wherein, the specific formula for calculating the average grayscale density of the neighborhood corresponding to each current bone key grayscale pixel of the current bone region data and the average grayscale density of the neighborhood corresponding to each normal bone key grayscale pixel of the predicted normal bone region data is as follows: Among them, Dmd i is the average grayscale density of the neighborhood corresponding to the i-th current bone key grayscale pixel point in the current bone region data, Xs i is the grayscale pixel value of the i-th current bone key grayscale pixel point in the current bone region data, Xs ia is the grayscale pixel value of the ath actual bone grayscale pixel in the set neighborhood of the ith current bone key grayscale pixel in the current bone region data, a=1, 2, 3, ..., A, A is the number of actual bone grayscale pixels in the set neighborhood of the current bone key grayscale pixel in the current bone region data, Zmd i To predict the average gray density of the neighborhood corresponding to the i-th normal bone key gray pixel point in the normal bone area data, Xs′ i To predict the grayscale pixel value of the i-th normal bone key grayscale pixel point in the normal bone area data, Xs′ ib To predict the grayscale pixel value of the bth normal bone grayscale pixel point within the set neighborhood of the i-th normal bone key grayscale pixel point of the normal bone area data, b=1, 2, 3, ..., B, B is the number of normal bone grayscale pixels within the set neighborhood of the normal bone key grayscale pixel point for predicting the normal bone area data, i=1, 2, 3, ..., N, N is the number of current bone key grayscale pixels of the current bone area data, and is also the number of normal bone key grayscale pixels for predicting the normal bone area data.

[0015] Furthermore, the specific steps to obtain the bone injury type corresponding to the current bone area data are as follows: perform corresponding difference analysis on the texture, shape, and density features of the current bone area with the texture, shape, and density features of the predicted normal bone area to obtain texture differences, shape differences, and density differences; perform discriminant analysis on the texture differences, shape differences, and density differences in combination with preset type judgment rules to obtain the bone injury type corresponding to the current bone area data.

[0016] The present invention has the following beneficial effects:

[0017] (1) This bone injury image-assisted detection system based on data analysis can not only identify the type of bone injury through precise comparison of texture, shape, and density features, but also analyze the severity of the injury in detail. This in-depth analysis supports doctors to customize personalized treatment plans for each patient, such as selecting the most suitable surgical method, drug treatment, or rehabilitation plan. In addition, accurate judgment of the severity of the injury can also help predict the treatment effect and the patient's recovery timeline, providing patients and their families with more accurate medical information and expectations.

[0018] (2) This bone injury image-assisted detection system based on data analysis greatly reduces the uncertainty of relying on doctors' personal experience and subjective judgment through the application of automated feature extraction and prediction models. For example, the local binary pattern (LBP) is used to analyze texture differences and the conditional generative adversarial network (CGAN) is used to predict and compare normal bone areas, ensuring the accuracy and consistency of the diagnosis results. This method can ensure that even between different medical personnel, consistent diagnosis results can be obtained for the same patient data.

[0019] (3) This bone injury image-assisted detection system based on data analysis significantly improves the analysis speed through the system's automated data processing and feature extraction process, reducing the time from data acquisition to final diagnosis. Especially in emergency situations, such as fractures caused by traffic accidents, rapid and accurate diagnosis can immediately initiate appropriate treatment, thereby improving the patient's chance of recovery and survival rate. This is particularly important in hospital emergency rooms that need to deal with a large number of patients, as it can effectively allocate medical resources and reduce the workload of medical staff.

[0020] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a block diagram of a bone injury image-assisted detection system based on data analysis in the present invention.

[0022] Figure 2This is a flowchart of the specific steps for obtaining current bone region data in a bone injury image-assisted detection system based on data analysis of the present invention.

[0023] Figure 3 This is a flow chart of the pre-establishment steps of a conditional generative adversarial network in a bone injury image-assisted detection system based on data analysis of the present invention. DETAILED DESCRIPTION

[0024] The embodiment of the present application solves the problem that traditional methods often rely heavily on doctors' interpretation of images, and differences in experience and skill levels of different doctors may lead to inconsistent diagnostic results. Secondly, manual analysis of image data is time-consuming and inefficient, especially when large amounts of data need to be processed, such as in epidemiological studies or large clinical trials. In addition, traditional imaging diagnostic methods have difficulty capturing tiny fractures or early osteoporosis, especially when the image quality is poor, which can easily lead to misjudgment.

[0025] The overall idea of ​​the problem in the embodiment of this application is as follows:

[0026] First, the system collects the current bone image data through the data acquisition module and performs necessary preprocessing, such as noise removal and contrast adjustment, to prepare for subsequent image analysis. The region extraction module is used to perform edge detection and region growing on the preprocessed images to accurately calibrate the affected bone areas. Then, key features are extracted from the image data of these specific areas, including texture (through LBP value), shape (through area and perimeter) and density (through grayscale analysis). Through the prediction module, the conditional generative adversarial network is used to generate images of undamaged normal bone areas for comparative analysis. Next, the features of the damaged bone area are analyzed for differences with those of the normal bone area to identify the specific differences in texture, shape and density. Finally, these differences are comprehensively analyzed according to the preset type judgment rules to determine the specific type of bone injury.

[0027] See also Figure 1The embodiment of the present invention provides a technical solution: a bone injury image-assisted detection system based on data analysis, comprising: a data acquisition module, a region extraction module, a prediction module, and a bone injury analysis module; the data acquisition module is used to acquire current bone image data and perform preprocessing; the region extraction module is used to perform region extraction on the preprocessed current bone image data to obtain current bone region data; the prediction module inputs the current bone region data into a pre-established prediction model for prediction analysis to obtain predicted normal bone region data; the bone injury analysis module is used to perform feature extraction on the current bone region data and the predicted normal bone region data respectively to obtain the texture, shape, and density features of the current bone region and the predicted normal bone region, and perform comprehensive analysis to obtain the bone injury type corresponding to the current bone region data.

[0028] The current bone image data specifically refers to the pixel value and two-dimensional coordinates of each pixel point in the current bone image, the current bone area data specifically refers to the pixel value and two-dimensional coordinates of each actual bone pixel point, and the normal bone area data specifically refers to the pixel value and two-dimensional coordinates of each normal bone pixel point.

[0029] Specifically, Figure 2 As shown, the specific steps of performing region extraction on the preprocessed current bone image data to obtain the current bone region data are as follows: reading the pixel value and two-dimensional coordinates of each pixel point in the preprocessed current bone image, and detecting the initial edge pixel point set of the initial bone edge in the current bone image based on the edge detection algorithm Canny, including several initial edge pixel points; using the regional growing algorithm to fill the initial bone edge to obtain the initial bone region pixel point set, including several bone pixel points within the edge; based on the preset pixel threshold, performing pixel division processing on the pixel value of each pixel point in the preprocessed current bone image to obtain the bone division pixel point set, including several bone division pixel points, which is specifically completed automatically by the Otsu method, and the method calculates the optimal threshold to maximize the between-class variance; merging the initial edge pixel point set, the initial bone region pixel point set, and the bone division pixel point set to obtain several actual bone pixels, and the pixel value and two-dimensional coordinates corresponding to each actual bone pixel point.

[0030] In this implementation, the current bone image is read, and preprocessing steps such as noise filtering and contrast adjustment are performed. This step can improve the quality of the image, make subsequent edge detection more accurate, and reduce the risk of misdiagnosis. The Canny edge detection algorithm is applied to identify the outline of the bone. In this step, the Canny algorithm can effectively identify clear edges and help doctors accurately locate the boundaries of the bone structure. This step ensures that fractures or other structural abnormalities can be accurately marked, which is crucial for diagnosing minor fractures or early lesions. The regional growing algorithm is applied to the edge pixels determined by the Canny algorithm to fill in all the pixels inside the bone edge. This step can ensure that all areas belonging to the same bone structure are completely captured through the regional growing algorithm, especially in areas with blurred edges. This step enhances the integrity of the image, makes the quantitative analysis of the bone area more accurate, and is conducive to further medical analysis and processing. , the optimal threshold is automatically calculated by the Otsu method to divide the bone and non-bone tissues. The Otsu method in this step can maximize the inter-class variance, automatically determine the threshold, reduce manual intervention, and improve processing speed and accuracy. This method provides an objective way to distinguish bones from other tissues, and is particularly suitable for processing situations where image quality is uneven or bone density varies greatly. The edge pixel set, the bone area pixel set, and the pixel set separated by the Otsu method are merged to identify the actual bone pixel points. The integrated data in this step can provide a complete bone image, ensuring that all relevant pixels are taken into account, and providing complete data support for subsequent diagnosis. Through this method, a detailed bone image can be finally obtained, making the judgment of the type of bone injury more dependent on comprehensive data analysis rather than a single image judgment, thereby improving the accuracy and scientificity of the diagnosis.

[0031] Specifically, the prediction model is a conditional generative adversarial network, which includes a generator that receives damaged bone images as input and generates undamaged bone images as output. The generator includes convolutional layers, deconvolutional layers and activation layers, and a discriminator that distinguishes the generated images from the real undamaged bone images. The discriminator includes convolutional layers, pooling layers and fully connected layers, a discriminative adversarial loss function, a generative adversarial loss function, and an optimizer Adam or RMSprop.

[0032] In this implementation scheme, the generator is designed to include convolutional layers and deconvolutional layers, through which damaged bone images are converted into undamaged images. Activation layers such as ReLU or LeakyReLU are used to increase nonlinear processing capabilities to help the network learn more complex features. The generator can create high-quality images similar to real bone images for intuitive comparison and further analysis. The generator is trained to generate undamaged bone images, which provides a powerful tool for evaluating and understanding the extent and nature of bone injuries. The discriminator is composed of convolutional layers, pooling layers, and fully connected layers, and is responsible for distinguishing the generated bone images from the real undamaged bone images. The discriminator promotes the generator to produce more and more realistic images by improving the judgment accuracy, thereby making The quality of the generated images is continuously improved. The discriminator helps ensure that the generated images are of sufficient quality for medical diagnosis and reduce the possibility of misdiagnosis. The network is trained using the generative adversarial loss function and the discriminative adversarial loss function. Optimizers such as Adam or RMSprop are used to adjust the neural network parameters and optimize the learning process. This adversarial training can significantly improve the authenticity and reliability of the generated images, making them closer to real-world medical images. The adversarial training mechanism enables the model to maintain efficient learning while generating high-quality images. The choice of optimizer further improves the stability and efficiency of the training process. CGAN can generate undamaged bone images, which can provide doctors with sufficient comparison materials even in the case of limited medical imaging data. This is particularly important in the field of medical research where data is scarce. By comparing with the generated undamaged bone images, doctors can more accurately identify and locate bone injuries, especially in the detection of subtle changes such as fractures and cracks. This directly improves the timeliness and effectiveness of treatment. Through the generated images, doctors can more carefully evaluate the specific injuries of each patient, thereby providing more personalized treatment recommendations. This method can ensure that patients receive the most appropriate treatment plan and reduce unnecessary medical interventions.

[0033] Specifically, Figure 3 As shown, the pre-establishment steps of the conditional generative adversarial network are as follows: obtain several groups of bone image data, each group of bone image data includes damaged bone images and corresponding undamaged normal bone images; pre-process each group of bone image data and divide it into an image training set and an image verification set; initialize the conditional generative adversarial network, specifically the weight initialization of the generator and the discriminator; based on the image training set, update the generator and the discriminator for each training cycle in the set number of training cycles; and after each training cycle, perform evaluation and analysis based on the image verification set, and adjust the parameters of the conditional generative adversarial network based on the evaluation and analysis until the output result of the conditional generative adversarial network meets the expected standards.

[0034] Among them, the specific implementation steps of training conditional generative adversarial networks are as follows:

[0035] At the beginning of training: the weights of the generator and discriminator are initialized, and the number of training cycles is set. Usually multiple epochs are required to ensure that the quality of the generated images gradually improves.

[0036] Training loop: For each epoch, the following steps are performed:

[0037] For each batch, the following steps are performed: extract a batch of damaged bone images and their corresponding undamaged normal bone images from the training set;

[0038] Generator update: Use the current generator to generate undamaged bone images, send the generated images and the real undamaged images together to the discriminator, calculate the generation adversarial loss (for example, using cross entropy loss), which reflects the difference between the generated image and the real image in the eyes of the discriminator, and use backpropagation to update the weights of the generator to reduce the generation adversarial loss.

[0039] Discriminator update: Provide the generated images and real images to the discriminator, calculate the discriminant adversarial loss, evaluate the discriminator's classification accuracy for real and generated images, and use backpropagation to update the discriminator's weights to improve its ability to distinguish between real and fake images.

[0040] Verification process: After each epoch, use the validation set to evaluate the model performance, calculate and record the loss and other key indicators on the validation set (such as image quality evaluation indicators), monitor the validation loss, and check whether the model is overfitting or still improving.

[0041] Adjust the learning rate: As the training progresses, adjust the learning rate accordingly. This can be done manually or automatically by setting up a learning rate scheduler.

[0042] Model saving and checkpointing: Save the state of the model periodically during training so that training can be resumed from the most recent checkpoint later.

[0043] By repeating the above steps, the predicted undamaged normal bone image corresponding to the damaged bone image output by the conditional generative adversarial network is consistent with the actual undamaged normal bone image.

[0044] In addition, the current bone region data is input into a pre-established prediction model for prediction analysis, and the specific steps for obtaining predicted normal bone region data are as follows:

[0045] The damaged current bone region data is fed into the generator of cGAN. The task of the generator is to generate predicted undamaged bone images based on the damaged input data.

[0046] The convolutional layers in the generator help extract features of the input data.

[0047] The activation function (such as ReLU) in the activation layer is applied to the output of the convolutional layer, introducing non-linearity and helping to capture complex data patterns.

[0048] The deconvolution layer is used to upsample the extracted features, gradually reconstruct the image details and structure, and gradually form a complete and undamaged bone image.

[0049] The generated undamaged bone images and the real undamaged bone images (from the training dataset) are fed into the discriminator simultaneously.

[0050] The convolutional and pooling layers help the discriminator understand and analyze the features and details of the image, and distinguish between generated images and real images.

[0051] The output of the convolution and pooling layers is converted into the final classification result, that is, whether the image is generated or real.

[0052] The discriminator calculates the difference between the generated image and the real image through the adversarial loss function to optimize its own accuracy.

[0053] Once the generator and discriminator are optimized to sufficient accuracy after several iterations of training, the generator will be able to output high-quality predicted normal bone region data.

[0054] The generated images may need to go through post-processing steps, such as thresholding, image sharpening, etc., to further improve the quality and usefulness of the output data.

[0055] In this implementation scheme, several groups of bone image data are collected, each group includes damaged bone images and corresponding undamaged normal bone images. This step ensures the diversity and coverage of the data, which helps the model learn more comprehensive features. This step provides sufficient training data to support the generalization ability of the model in various bone injury situations. The bone image data is preprocessed (such as normalization, denoising, etc.) and then divided into a training set and a validation set. The standardized data processing step in this step improves the stability and effect of model training. This step ensures that the performance evaluation of the model during the training process is more reliable through reasonable data division. The weights of the generator and the discriminator are initialized, and the number of training cycles (epoch number) is set. Good weight initialization in this step can accelerate training convergence and improve the performance of the final model. This step provides a good starting point, making the subsequent training process more efficient. A batch of damaged bone images and their corresponding undamaged normal bone images are extracted from the training set. The generator generates an undamaged bone image, and sends the generated image and the real undamaged image to the discriminator together, calculates the generation adversarial loss (such as cross entropy loss), and uses back propagation to update the weight of the generator. This step minimizes the generation adversarial loss so that the generated image gradually approaches the real image, thereby improving the quality of image generation. In this step, the generator can continuously improve during the training process and generate higher quality undamaged bone images to provide strong support for diagnosis. The generated image and the real image are provided to the discriminator, the discriminative adversarial loss is calculated, the classification accuracy of the discriminator for the real and generated images is evaluated, and the weight of the discriminator is updated using back propagation. This step improves the discriminator's ability to distinguish between real and generated images, so that the quality of the images generated by the generator continues to improve. The continuous optimization of the discriminator in this step ensures that the model is effective in distinguishing between real and generated images. High precision in true and false images, thereby improving the reliability of the entire system. After each epoch, the validation set is used to evaluate the model performance, the loss and other key indicators on the validation set (such as image quality evaluation indicators) are calculated and recorded, and the validation loss is monitored. In this step, the overfitting or underfitting problems of the model can be discovered and corrected in a timely manner through the evaluation of the validation set, ensuring the consistency of the model in training and practical applications. This step provides a real-time feedback mechanism for model performance, ensuring the efficiency of the training process and the reliability of the results. According to the progress of training, the learning rate is adjusted accordingly, which can be done manually or automatically by setting the learning rate scheduler. The learning rate adjustment in this step can accelerate the convergence process and prevent oscillation and instability during training. This step optimizes the training process of the model and improves the final model performance and training efficiency.

[0056] Specifically, feature extraction is performed on the current bone region data and the predicted normal bone region data respectively, and the specific steps for obtaining the texture features of the current bone region and the predicted normal bone region are as follows: grayscale processing is performed on the current bone region data and the predicted normal bone region data respectively, and key point extraction is performed on the current bone region data after the grayscale processing to obtain several current bone key grayscale pixel points of the current bone region data, and key point correspondence matching is performed on the predicted normal bone region data based on the several current bone key grayscale pixel points of the current bone region data to obtain several normal bone key grayscale pixel points of the predicted normal bone region data, and each current bone key grayscale pixel point corresponds to each normal bone key grayscale pixel point one by one, so the number is the same, and the specific process is: use SIFT or ORB algorithm to process the two images to detect significant points on the edge as key points, and record these key points. The coordinates of the points are counted, and these coordinates will be used for subsequent feature extraction; the number of actual bone grayscale pixels within the set neighborhood of each current bone key grayscale pixel of the current bone region data is counted, and the corresponding grayscale pixel values ​​are analyzed separately to obtain the LBP value of each current bone key grayscale pixel of the current bone region data, which is the texture feature of the current bone region; the number of normal bone grayscale pixels within the set neighborhood of each normal bone key grayscale pixel of the predicted normal bone region data is counted, and the corresponding grayscale pixel values ​​are analyzed separately to obtain the LBP value of each normal bone key grayscale pixel of the predicted normal bone region data, which is the texture feature of the predicted normal bone region; wherein, the specific formula for calculating the LBP value of each current bone key grayscale pixel of the current bone region data and the LBP value of each normal bone key grayscale pixel of the predicted normal bone region data is as follows: Among them, LBP i is the LBP value of the i-th current bone key grayscale pixel point in the current bone region data, Xs i is the grayscale pixel value of the i-th current bone key grayscale pixel point in the current bone region data, Xs ia is the grayscale pixel value of the ath actual bone grayscale pixel in the set neighborhood of the i-th current bone key grayscale pixel in the current bone region data, S(x) is the sign function, a=1, 2, 3, ..., A, A is the number of actual bone grayscale pixels in the set neighborhood of the current bone key grayscale pixel in the current bone region data, LBP i ′ is the LBP value of the i-th normal bone key grayscale pixel point in the predicted normal bone area data, Xs′ i To predict the grayscale pixel value of the i-th normal bone key grayscale pixel point in the normal bone area data, Xs′ ibTo predict the grayscale pixel value of the bth normal bone grayscale pixel point within the set neighborhood of the i-th normal bone key grayscale pixel point of the normal bone area data, b=1, 2, 3, ..., B, B is the number of normal bone grayscale pixels within the set neighborhood of the normal bone key grayscale pixel point for predicting the normal bone area data, i=1, 2, 3, ..., N, N is the number of current bone key grayscale pixels of the current bone area data, and is also the number of normal bone key grayscale pixels for predicting the normal bone area data.

[0057] It should be explained that the definition of s(x) is: if x≥0, then the value of s(x) is 1, and if x<0, then the value of s(x) is 0.

[0058] In this implementation, the current bone region data and the predicted normal bone region data are converted into grayscale images respectively. This step simplifies image processing, reduces the amount of data, and makes the feature extraction process more efficient. This step ensures the consistency and repeatability of image analysis by unifying grayscale image processing, and reduces the impact of differences between different images on the results. The two grayscale images are processed using SIFT or ORB algorithms to detect significant points on the edges as key points. These algorithms in this step can effectively identify important feature points in the image, ensuring that the extracted key points have a high degree of stability and robustness. This step improves the accuracy and efficiency of key point detection, and provides a reliable foundation for subsequent polygon construction and feature extraction. Several current bone key grayscale images based on the current bone region data are used. Pixel points, key point correspondence matching is performed on the predicted normal bone area data to obtain several normal bone key grayscale pixel points of the predicted normal bone area data. This step ensures that in the feature extraction and analysis process, the bone features at the same position are compared to improve the accuracy of the comparison results. This step solves the problem of inaccurate feature comparison between different images in the traditional method through precise key point matching. The number of actual bone grayscale pixel points within the set neighborhood of each current bone key grayscale pixel point is counted, and each analysis is performed based on the corresponding grayscale pixel value. In this step, the grayscale value changes around each key point are analyzed in detail to capture subtle structural differences. This step provides rich local texture information through statistical analysis of the neighborhood grayscale values, which helps to refine the changes in bone structure.

[0059] Specifically, the features of the current bone region data and the predicted normal bone region data are extracted respectively, and the specific steps of obtaining the shape features of the current bone region and the predicted normal bone region are as follows: read several current bone key grayscale pixel points of the current bone region data for closed connection to obtain the current bone polygon region, and analyze the perimeter value and area value of the current bone polygon region and the compactness index of the current bone region, which is the shape feature of the current bone region; read several normal bone key grayscale pixel points of the predicted normal bone region data for closed connection to obtain the normal bone polygon region, and analyze the perimeter value and area value of the normal bone polygon region and the compactness index of the normal bone region, which is the shape feature of the predicted normal bone region; wherein, the specific formula for calculating the perimeter value, area value, and compactness index of the current bone polygon region is as follows: Among them, Dzc is the perimeter value of the current bone polygon area, Dmj is the area value of the current bone polygon area, (x i ,y i ) is the i-th current bone key grayscale pixel point of the current bone region data, (x i+1 ,y i+1 ) means that when i takes the last current bone key grayscale pixel point, it is reset to the coordinates of the first current bone key grayscale pixel point, Djc is the current bone area compactness index, π is a natural constant, usually 3.14, i = 1, 2, 3, ..., N, N is the number of current bone key grayscale pixels in the current bone area data.

[0060] It needs to be explained that the specific formula for calculating the perimeter value and area value of the normal bone polygon area and the compactness index of the normal bone area is consistent with the logic used in the specific formula for calculating the perimeter value, area value and compactness index of the current bone polygon area.

[0061] The area value of the current bone polygon region is calculated using the shoelace formula.

[0062] In this embodiment, the key grayscale pixel points of the extracted current bone area are closed and connected in sequence to form the current bone polygon area. In the same way, the key points of the predicted normal bone area are closed and connected to form a normal bone polygon area. This step can ensure that all key points are correctly connected to form a complete bone outline through closed connection. This step provides an effective way to construct the geometric structure of the bone and ensure the accuracy of the analysis through automated feature extraction and geometric analysis, reducing the subjectivity and inconsistency of manual operations. The use of standardized calculation methods, such as perimeter, area and compactness index, ensures the objectivity and accuracy of the diagnosis. These standardized calculation methods make the diagnostic results more reliable and help doctors make more accurate diagnoses.

[0063] Specifically, the specific steps for obtaining the density characteristics of the current bone region and the predicted normal bone region are as follows: read the number of actual bone grayscale pixels within the set neighborhood of each current bone key grayscale pixel of the current bone region data, and analyze them separately based on the corresponding grayscale pixel values ​​to obtain the average grayscale density of the neighborhood corresponding to each current bone key grayscale pixel of the current bone region data, which is the density characteristic of the current bone region; read the number of normal bone grayscale pixels within the set neighborhood of each normal bone key grayscale pixel of the predicted normal bone region data, and analyze them separately based on the corresponding grayscale pixel values ​​to obtain the average grayscale density of the neighborhood corresponding to each normal bone key grayscale pixel of the predicted normal bone region data, which is the density characteristic of the predicted normal bone region; wherein, the specific formulas for calculating the average grayscale density of the neighborhood corresponding to each current bone key grayscale pixel of the current bone region data and the average grayscale density of the neighborhood corresponding to each normal bone key grayscale pixel of the predicted normal bone region data are as follows: Among them, Dmd i is the average grayscale density of the neighborhood corresponding to the i-th current bone key grayscale pixel point in the current bone region data, Xs i is the grayscale pixel value of the i-th current bone key grayscale pixel point in the current bone region data, Xs ia is the grayscale pixel value of the ath actual bone grayscale pixel in the set neighborhood of the ith current bone key grayscale pixel in the current bone region data, a=1, 2, 3, ..., A, A is the number of actual bone grayscale pixels in the set neighborhood of the current bone key grayscale pixel in the current bone region data, Zmd i To predict the average gray density of the neighborhood corresponding to the i-th normal bone key gray pixel point in the normal bone area data, Xs′ i To predict the grayscale pixel value of the i-th normal bone key grayscale pixel point in the normal bone area data, Xs′ ib To predict the grayscale pixel value of the bth normal bone grayscale pixel point within the set neighborhood of the i-th normal bone key grayscale pixel point of the normal bone area data, b=1, 2, 3, ..., B, B is the number of normal bone grayscale pixels within the set neighborhood of the normal bone key grayscale pixel point for predicting the normal bone area data, i=1, 2, 3, ..., N, N is the number of current bone key grayscale pixels of the current bone area data, and is also the number of normal bone key grayscale pixels for predicting the normal bone area data.

[0064] In this embodiment, the number of actual bone grayscale pixel points within the set neighborhood of each key grayscale pixel point in the current bone area data is extracted, and separate analyses are performed based on these grayscale pixel values. This step can capture local structural changes of the bone by accurately calculating the number of grayscale pixel values ​​in the neighborhood, and provide detailed density analysis. This step provides an accurate calculation method for local grayscale density, improves the perception of bone density changes, extracts the number of normal bone grayscale pixel points within the set neighborhood of each key grayscale pixel point in the predicted normal bone area data, and separate analyses are performed based on these grayscale pixel values. This step can provide a comparison benchmark by accurately calculating the number of grayscale pixel values ​​in the normal bone neighborhood, so as to more accurately assess the abnormality of the damaged bone. This step provides reference data for normal bone density and enhances a comprehensive understanding of bone health.

[0065] Specifically, the specific steps to obtain the bone injury type corresponding to the current bone area data are as follows: perform corresponding difference analysis on the texture, shape, and density features of the current bone area with the texture, shape, and density features of the predicted normal bone area to obtain texture differences, shape differences, and density differences; perform discriminant analysis on the texture differences, shape differences, and density differences in combination with preset type judgment rules to obtain the bone injury type corresponding to the current bone area data.

[0066] Among them, it needs to be explained that for texture differences: by comparing the differences in LBP values, the degree of local structural changes can be quantitatively understood; for shape differences: shape difference analysis (area, perimeter, compactness) can reveal physical deformation or loss of structural integrity; for density differences: comparison of density characteristics can reveal changes in bone density, which is very critical in diagnosing osteoporosis or other similar diseases.

[0067] The default type determination rules are defined as follows:

[0068] Rules for texture differences:

[0069] Mild changes: If the difference in LBP values ​​is less than the predetermined threshold A, it indicates that the surface changes of the bone are minor and may be related to early osteoporosis or minor injuries.

[0070] Significant change: If the difference in LBP values ​​is greater than threshold A but less than threshold B, this may indicate a moderate change in bone structure, such as more severe osteoporosis.

[0071] Severe changes: If the difference in LBP values ​​exceeds threshold B, it indicates severe changes in bone structure, which may be related to severe fractures or bone destruction.

[0072] Rules for shape differences:

[0073] Mild deformation: Small changes in area or perimeter, little change in compactness index, may indicate a mild fracture or crack.

[0074] Significant deformation: A significant change in area or perimeter, or a large change in the compactness index, may indicate a significant crack or partial fracture of the bone.

[0075] Severe deformation: The area or circumference changes greatly, and the compactness index is significantly reduced, which may be a complete fracture or severe bone deformation.

[0076] Rules for density differences:

[0077] Mild changes: The difference in density is small, indicating a slight decrease in bone density, which may be an early sign of osteoporosis.

[0078] Significant change: A moderate difference in density, indicating a significant decrease in bone density and a possible stage in the development of osteoporosis.

[0079] Severe changes: A large difference in density indicates a severe decrease in bone density, which may be severe osteoporosis or bone loss.

[0080] In this embodiment, the LBP value of the current bone region is compared with the LBP value of the predicted normal bone region, and the difference is calculated. This step can quantitatively understand the local structural changes on the bone surface, such as osteoporosis or minor bone damage. This step provides an objective way to evaluate the texture changes of the bone through the difference analysis of the LBP value, thereby improving the accuracy of diagnosis. The area, perimeter and compactness index of the current bone region are compared with the corresponding values ​​of the predicted normal bone region, and the difference is calculated. This step can identify the physical deformation of the bone or the loss of structural integrity, such as fractures or cracks. This step can provide detailed geometric change information through the difference analysis of shape features, which is helpful for diagnosing the type and severity of fractures. The average grayscale density of the current bone region is compared with the average grayscale density of the predicted normal bone region, and the difference is calculated. This step can reveal the changes in bone density, such as the degree of osteoporosis. This step provides a quantitative way to evaluate the changes in bone density through density difference analysis, which is very critical for the diagnosis of diseases such as osteoporosis. Through clear type judgment rules and difference analysis, the influence of subjective judgment is reduced, and the consistency and reliability of the diagnostic results are improved. The use of standardized difference analysis methods, such as differences in LBP values, geometric features, and grayscale density, ensures the objectivity and accuracy of the diagnosis. These standardized judgment rules make the diagnosis more reliable and help doctors make more accurate diagnoses.

[0081] And the specific implementation example of obtaining the bone injury type corresponding to the current bone region data is as follows: Assuming that the current bone region data of a patient is compared with the predicted normal bone region data, we obtain the following results:

[0082] Texture difference: LBP difference is 25% (threshold A=10%, B=20%).

[0083] Shape differences: 15% less area, 10% less perimeter, 30% less compactness index.

[0084] Density difference: Average grayscale density is 40% lower.

[0085] That is, according to the preset rules:

[0086] Differences in texture indicate severe structural changes.

[0087] The shape differences indicate significant to severe deformation.

[0088] The density difference indicates severe bone density loss.

[0089] Combining this information, it can be inferred that the patient may have suffered a severe fracture or osteoporosis.

[0090] In summary, this application has at least the following effects:

[0091] Through precise comparison of texture, shape, and density features, the system can not only identify the type of bone injury, but also finely analyze the severity of the injury. This in-depth analysis supports doctors to customize personalized treatment plans for each patient, such as choosing the most suitable surgical method, drug treatment, or rehabilitation plan. In addition, accurate judgment of the severity of the injury can also help predict the treatment effect and the patient's recovery timeline, providing patients and their families with more accurate medical information and expectations.

[0092] Through the application of automated feature extraction and prediction models, the uncertainty of relying on doctors' personal experience and subjective judgment is greatly reduced. For example, the use of local binary patterns (LBP) to analyze texture differences and conditional generative adversarial networks (CGAN) to predict and compare normal bone areas ensures the accuracy and consistency of the diagnosis results. This method can ensure that even between different medical personnel, consistent diagnosis results can be obtained for the same patient data.

[0093] The system's automated data processing and feature extraction processes significantly improve the speed of analysis and reduce the time from data acquisition to final diagnosis. Especially in emergency situations, such as fractures caused by traffic accidents, rapid and accurate diagnosis can immediately initiate appropriate treatment, thereby improving the patient's chance of recovery and survival rate. This is especially important in hospital emergency rooms that need to deal with a large number of patients, which can effectively allocate medical resources and reduce the workload of medical staff.

[0094] Although the preferred embodiments of the present invention have been described, other changes and modifications may be made to these embodiments once the basic creative concept is known to those skilled in the art. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0095] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A bone injury image-assisted detection system based on data analysis, characterized in that: include: Data acquisition module, region extraction module, prediction module, bone injury analysis module; The data acquisition module is used to acquire current bone image data and perform preprocessing; The region extraction module is used to perform region extraction on the preprocessed current bone image data to obtain current bone region data; The prediction module inputs the current bone region data into a pre-established prediction model for prediction analysis to obtain predicted normal bone region data; The bone injury analysis module is used to extract features from the current bone region data and the predicted normal bone region data, respectively, to obtain the texture, shape, and density features of the current bone region and the predicted normal bone region, and to perform comprehensive analysis to obtain the bone injury type corresponding to the current bone region data; The prediction model is specifically a conditional generative adversarial network, which includes a generator that receives a damaged bone image as input and generates an undamaged bone image as output; The specific steps to obtain the density characteristics of the current bone area and predict the normal bone area are as follows: Read the number of actual bone grayscale pixels within the set neighborhood of each current bone key grayscale pixel of the current bone region data, and analyze them separately based on the corresponding grayscale pixel values ​​to obtain the average grayscale density of the neighborhood corresponding to each current bone key grayscale pixel of the current bone region data, which is the density feature of the current bone region; The number of normal bone grayscale pixels in the set neighborhood of each normal bone key grayscale pixel of the predicted normal bone area data is read, and the corresponding grayscale pixel values ​​are analyzed respectively to obtain the average grayscale density of the neighborhood corresponding to each normal bone key grayscale pixel of the predicted normal bone area data, which is the density feature of the predicted normal bone area; The specific formulas for calculating the average grayscale density of the neighborhood corresponding to each current bone key grayscale pixel point of the current bone region data and predicting the average grayscale density of the neighborhood corresponding to each normal bone key grayscale pixel point of the normal bone region data are as follows: ； in, The current bone region data The average gray density of the neighborhood corresponding to the current key gray pixel of the skeleton, The current bone region data The grayscale pixel value of the current key grayscale pixel of the skeleton, The current bone region data The first pixel in the set neighborhood of the current key grayscale pixel of the skeleton The grayscale pixel value of the actual bone grayscale pixel point, , The number of actual bone grayscale pixels within the set neighborhood of the current bone key grayscale pixel of the current bone region data. To predict the normal bone area data The average gray density of the neighborhood corresponding to the key gray pixel points of normal bones, To predict the normal bone area data The grayscale pixel value of the key grayscale pixel of a normal bone, To predict the normal bone area data The first normal bone key grayscale pixel in the set neighborhood The grayscale pixel value of a normal bone grayscale pixel, , To predict the number of normal bone grayscale pixels in the set neighborhood of the normal bone key grayscale pixels of the normal bone region data, , is the number of key grayscale pixels of the current bone in the current bone region data, and is also the number of key grayscale pixels of the normal bone in the predicted normal bone region data; The specific steps to obtain the bone injury type corresponding to the current bone area data are as follows: The texture, shape, and density features of the current bone region are respectively analyzed with the texture, shape, and density features of the predicted normal bone region to obtain texture differences, shape differences, and density differences; Combined with the preset type judgment rules, the texture differences, shape differences, and density differences are discriminated and analyzed to obtain the bone injury type corresponding to the current bone area data.

2. The bone injury image-assisted detection system based on data analysis according to claim 1 is characterized in that: The current bone image data specifically refers to the pixel value and two-dimensional coordinates of each pixel point in the current bone image, the current bone area data specifically refers to the pixel value and two-dimensional coordinates of each actual bone pixel point, and the normal bone area data specifically refers to the pixel value and two-dimensional coordinates of each normal bone pixel point.

3. The bone injury image-assisted detection system based on data analysis according to claim 2 is characterized in that: The specific steps of performing region extraction on the preprocessed current bone image data to obtain the current bone region data are as follows: Read the pixel value and two-dimensional coordinates of each pixel point in the preprocessed current bone image, and detect an initial edge pixel point set of an initial bone edge in the current bone image based on an edge detection algorithm, including a plurality of initial edge pixel points; The initial bone edge is filled by using the region growing algorithm to obtain the initial bone region pixel point set, including several bone pixel points within the edge; Based on a preset pixel threshold, pixel division processing is performed on the pixel value of each pixel point in the preprocessed current bone image to obtain a bone division pixel point set, including a plurality of bone division pixel points; The initial edge pixel point set, the initial skeleton region pixel point set, and the skeleton partition pixel point set are merged to obtain a number of actual skeleton pixel points, and the pixel value and two-dimensional coordinates corresponding to each actual skeleton pixel point.

4. The bone injury image-assisted detection system based on data analysis according to claim 1 is characterized in that: The prediction model is specifically a conditional generative adversarial network, which includes a generator that receives damaged bone images as input and generates undamaged bone images as output. The generator includes convolutional layers, deconvolutional layers and activation layers, and a discriminator that distinguishes the generated images from the real undamaged bone images. The discriminator includes convolutional layers, pooling layers and fully connected layers, a discriminative adversarial loss function, a generative adversarial loss function, and an optimizer.

5. The bone injury image-assisted detection system based on data analysis according to claim 4 is characterized in that: The pre-establishment steps of the conditional generative adversarial network are as follows: Acquire a plurality of sets of bone image data, each set of bone image data including a damaged bone image and a corresponding undamaged normal bone image; Each set of bone image data is preprocessed and divided into an image training set and an image verification set; Initialize the conditional generative adversarial network, specifically the weight initialization of the generator and discriminator; Based on the image training set, each training cycle in the set number of training cycles is processed to update the generator and the discriminator respectively; After each training cycle, an evaluation analysis is performed based on the image verification set, and the parameters of the conditional generative adversarial network are adjusted based on the evaluation analysis until the output results of the conditional generative adversarial network meet the expected standards.

6. The bone injury image-assisted detection system based on data analysis according to claim 3 is characterized in that: The specific steps of extracting features from the current bone region data and the predicted normal bone region data to obtain the texture features of the current bone region and the predicted normal bone region are as follows: Grayscale processing is performed on the current bone region data and the predicted normal bone region data respectively, and key point extraction is performed on the current bone region data after the grayscale processing to obtain a number of current bone key grayscale pixel points of the current bone region data, and key point correspondence matching is performed on the predicted normal bone region data based on the several current bone key grayscale pixel points of the current bone region data to obtain a number of normal bone key grayscale pixel points of the predicted normal bone region data, and each current bone key grayscale pixel point corresponds to each normal bone key grayscale pixel point one by one; Count the number of actual bone grayscale pixels within the set neighborhood of each current bone key grayscale pixel of the current bone region data, and analyze them separately based on the corresponding grayscale pixel values ​​to obtain the LBP value of each current bone key grayscale pixel of the current bone region data, which is the texture feature of the current bone region; The number of normal bone grayscale pixel points within the set neighborhood of each normal bone key grayscale pixel point of the predicted normal bone area data is counted, and the LBP value of each normal bone key grayscale pixel point of the predicted normal bone area data is obtained based on the corresponding grayscale pixel values, which is the texture feature of the predicted normal bone area; The specific formula for calculating the LBP value of each current bone key grayscale pixel point of the current bone region data and predicting the LBP value of each normal bone key grayscale pixel point of the normal bone region data is as follows: ； in, The current bone region data The LBP value of the current key grayscale pixel of the skeleton, The current bone region data The grayscale pixel value of the current key grayscale pixel of the skeleton, The current bone region data The first pixel in the set neighborhood of the current key grayscale pixel of the skeleton The grayscale pixel value of the actual bone grayscale pixel point, is the symbolic function, , The number of actual bone grayscale pixels within the set neighborhood of the current bone key grayscale pixel of the current bone region data. To predict the normal bone area data The LBP value of the key grayscale pixel of a normal bone, To predict the normal bone area data The grayscale pixel value of the key grayscale pixel of a normal bone, To predict the normal bone area data The first normal bone key grayscale pixel in the set neighborhood The grayscale pixel value of a normal bone grayscale pixel, , To predict the number of normal bone grayscale pixels in the set neighborhood of the normal bone key grayscale pixels of the normal bone region data, , It is the number of key grayscale pixels of the current bone in the current bone region data, and is also the number of key grayscale pixels of the normal bone in the predicted normal bone region data.

7. The bone injury image-assisted detection system based on data analysis according to claim 6 is characterized in that: The specific steps of extracting features from the current bone region data and the predicted normal bone region data to obtain shape features of the current bone region and the predicted normal bone region are as follows: Read several current bone key grayscale pixel points of the current bone region data for closed connection to obtain the current bone polygon region, and analyze the perimeter value and area value of the current bone polygon region and the compactness index of the current bone region, which is the shape feature of the current bone region; Read several normal bone key grayscale pixel points of the predicted normal bone area data and connect them in a closed manner to obtain a normal bone polygon area, and analyze the perimeter value and area value of the normal bone polygon area and the normal bone area compactness index, that is, to predict the shape characteristics of the normal bone area; The specific formulas for calculating the perimeter value, area value, and compactness index of the current bone polygon area are as follows: ； in, is the perimeter value of the current bone polygon area, is the area value of the current bone polygon area, The current bone region data The current key grayscale pixel points of the skeleton, express When the last key grayscale pixel of the current bone is obtained, it is reset to the coordinates of the first key grayscale pixel of the current bone. is the compactness index of the current bone region, is a natural constant, , The number of key grayscale pixels of the current bone in the current bone region data.

Images