Bone injury degree evaluation system based on artificial intelligence

Abstract

The invention discloses a bone injury degree evaluation system based on artificial intelligence. The bone injury degree evaluation system comprises a data set building module, an image enhancement module, a bone injury degree evaluation model building module and a bone injury degree evaluation module. The invention relates to the technical field of bone injury assessment, in particular to a bone injury degree assessment system based on artificial intelligence, which enhances the visibility of tiny injury features through self-adaptive adjustment of local scale factors, extraction of base information by Gaussian kernel convolution, enhancement of details by anti-base information and multi-scale structure tensor fusion. Interference of noise on injury identification is reduced, and the accuracy and reliability of a bone injury evaluation system are effectively improved; through dynamic convolution, local information expression is enhanced, variational distribution modeling is performed, a channel attention mechanism is introduced, features are projected to a Riemannian manifold space, and variational loss optimization model training is integrated, so that the accuracy, stability and generalization ability of bone injury degree evaluation are improved, and treatment decision making is more accurately assisted.

Description

Technical Field

[0001] The present invention relates to the technical field of bone injury assessment, and specifically refers to an artificial intelligence-based bone injury degree assessment system. Background Art

[0002] The artificial intelligence-based bone injury degree assessment system utilizes computer vision, deep learning, and medical image analysis technologies. By training medical image data, it can automatically interpret medical images, accurately assess the degree of bone injury, improve the efficiency of clinical decision-making, and reduce misdiagnosis and missed diagnosis. Traditional bone injury assessment schemes have problems such as unclear medical image features of bone injuries and blurred edges, lost details, and large noise interference in traditional image enhancement methods for bone injury detection; traditional bone injury assessment schemes have problems such as insufficient feature extraction, lost local information, inability to adaptively adjust channel weights, and limited assessment accuracy. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides an artificial intelligence-based bone injury degree assessment system. Aiming at the problems of unclear medical image features of bone injuries and blurred edges, lost details, and large noise interference in traditional image enhancement methods for bone injury detection in traditional bone injury assessment schemes, this solution improves the contrast and edge clarity of the bone injury area through adaptive adjustment of local scale factors, extraction of base information by Gaussian kernel convolution, enhancement of details by anti-base information, and fusion of multi-scale structure tensors. At the same time, it enhances the visibility of micro-injury features, reduces the interference of noise on injury recognition, and thus effectively improves the accuracy and reliability of the bone injury assessment system; aiming at the problems of insufficient feature extraction, lost local information, inability to adaptively adjust channel weights, and limited assessment accuracy in traditional bone injury assessment schemes, this solution improves the accuracy, robustness, and generalization ability of bone injury degree assessment through dynamic convolution to enhance local information expression, variational distribution modeling, introduction of channel attention mechanism, projection of features into the Riemannian manifold space, and incorporation of variational loss to optimize model training, so as to more accurately assist clinical diagnosis and treatment decisions.

[0004] The technical solution adopted by the present invention is as follows: The artificial intelligence-based bone injury degree assessment system provided by the present invention includes a data set construction module, an image enhancement module, a bone injury degree assessment model establishment module, and a bone injury degree assessment module;

[0005] The data set construction module collects historical CT images of bone injuries and the degree of bone injuries, annotates the CT images with the degree of bone injuries, and finally creates a model training data set and adds the annotated CT images to the model training data set;

[0006] The image enhancement module includes: a local scale determination unit, a base information acquisition unit, an information inversion unit, a bilateral weight generation unit, a detail enhancement unit, a local structure tensor construction unit, an adaptive weight generation unit, and a feature fusion unit;

[0007] The bone injury degree assessment model establishment module includes: a dynamic convolution unit, a variational distribution definition unit, a variational feature pooling unit, a channel attention enhancement unit, a feature projection unit, an evaluation unit, a loss calculation unit, and a parameter learning unit;

[0008] The bone injury degree assessment module collects the bone injury CT images of the user, inputs the images into the bone injury degree assessment model, and the model outputs the bone injury degree of the user.

[0009] Furthermore, the dataset construction module collects historical bone injury CT images and bone injury degrees, where the bone injury degrees include mild injury, moderate injury, and severe injury, annotates the CT images with the bone injury degree data, and finally creates a model training dataset and adds the annotated CT images to the model training dataset.

[0010] Furthermore, the image enhancement module specifically includes the following units:

[0011] The local scale determination unit first sets the size of the local window, calculates the variance of the pixels within the local window, and finally calculates the local scale unit;

[0012] The base information acquisition unit performs Gaussian kernel convolution using the local scale factor to obtain the base information;

[0013] The information inversion unit subtracts the base information from the original image to obtain the anti-base information containing the bone injury edges and fine textures;

[0014] The bilateral weight generation unit calculates the bilateral weights within the local window while considering the spatial distance and the anti-base information difference;

[0015] The detail enhancement unit magnifies the edge and texture features of the bone injury in the detail layer through local bilateral weighting while retaining the edge information;

[0016] The local structure tensor construction unit captures the local structure information and edge direction features by calculating the image gradient and performing Gaussian smoothing;

[0017] The adaptive weight generation unit;

[0018] The feature fusion unit uses the adaptive weights to fuse the features of the detail layer and the structure layer, and comprehensively depicts the bone injury details and local structures.

[0019] Furthermore, the module for establishing the bone injury degree evaluation model specifically includes the following units:

[0020] The dynamic convolution unit introduces a depth adjustment matrix to modulate the weights of the convolution kernels, dynamically responds to convolution, and reconstructs local information;

[0021] The variational distribution definition unit constructs a variational distribution using the statistical mean and variance of local region features;

[0022] The variational feature pooling unit performs a pooling operation using the method of variational sampling to pool local feature information;

[0023] The channel attention enhancement unit enhances important channel features by generating channel attention weights;

[0024] The feature projection unit projects features onto the Riemannian manifold space;

[0025] The evaluation unit evaluates the bone injury degree category;

[0026] The loss calculation unit measures the model classification error using cross-entropy loss. Meanwhile, a variational distribution regularization term is introduced to encourage the model to learn a stable and information-rich feature distribution;

[0027] The parameter learning unit calculates the gradient of the loss function with respect to the model parameters and sets a learning rate for parameter update.

[0028] Furthermore, the bone injury degree evaluation module collects the bone injury CT images of the user, inputs the images into the bone injury degree evaluation model, and the model outputs the bone injury degree of the user.

[0029] The beneficial effects achieved by the present invention using the above solution are as follows:

[0030] (1) Aiming at the problems existing in the traditional bone injury evaluation scheme, such as unclear medical image features of bone injuries and edge blurring, detail loss, and large noise interference in traditional image enhancement methods for bone injury detection, this solution improves the contrast and edge sharpness of the bone injury area through local scale factor adaptive adjustment, Gaussian kernel convolution to extract base information, anti-base information to enhance details, and multi-scale structure tensor fusion. At the same time, it enhances the visibility of micro-injury features and reduces the interference of noise on injury recognition, thereby effectively improving the accuracy and reliability of the bone injury evaluation system.

[0031] (2) Aiming at the problems of insufficient feature extraction, local information loss, inability to adaptively adjust channel weights, and limited evaluation accuracy in traditional bone injury assessment schemes, this solution enhances the expression of local information through dynamic convolution, variational distribution modeling, introducing a channel attention mechanism, projecting features into the Riemannian manifold space, and integrating variational loss to optimize model training, improving the accuracy, robustness, and generalization ability of bone injury degree assessment, thus more accurately assisting clinical diagnosis and treatment decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 FIG. is a schematic diagram of an artificial intelligence-based bone injury degree assessment system provided by the present invention;

[0033] Figure 2 FIG. is a schematic diagram of an image enhancement module;

[0034] Figure 3 FIG. is a schematic diagram of a module for establishing a bone injury degree assessment model.

[0035] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0037] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention.

[0038] Example 1, referring to Figure 1 , the artificial intelligence-based bone injury degree assessment system provided by the present invention includes a data set construction module, an image enhancement module, a bone injury degree assessment model establishment module, and a bone injury degree assessment module;

[0039] The dataset construction module collects historical CT images of bone injuries and the degrees of bone injuries, where the degrees of bone injuries include mild injuries, moderate injuries, and severe injuries. The CT images are labeled with the data of the degrees of bone injuries, and finally a model training dataset is created, and the labeled CT images are added to the model training dataset;

[0040] The image enhancement module receives the data sent by the image enhancement module, performs image enhancement through the local scale unit determination unit, base information acquisition unit, information inversion unit, bilateral weight generation unit, detail enhancement unit, local structure tensor construction unit, adaptive weight generation unit, and feature fusion unit, and sends the data to the bone injury degree evaluation model establishment module;

[0041] The bone injury degree evaluation model establishment module receives the data sent by the image enhancement module, and establishes a bone injury degree evaluation model through the dynamic convolution unit, variational distribution definition unit, variational feature pooling unit, channel attention enhancement unit, feature projection unit, evaluation unit, loss calculation unit, and parameter learning unit, and sends the data to the bone injury degree evaluation module;

[0042] The bone injury degree evaluation module receives the data sent by the bone injury degree evaluation model establishment module, and uses the bone injury degree evaluation model to evaluate the user's bone injury degree.

[0043] Embodiment 2, refer to Figure 1 and Figure 2 , based on the above embodiment, the image enhancement module specifically includes the following units:

[0044] The local scale unit determination unit first sets the size of the local window to 5 by 5 pixels, calculates the variance of the pixels within the local window, and finally calculates the local scale unit, which is expressed as follows:

[0045] ;

[0046] where x and y respectively represent the abscissa index and ordinate index of the pixels of the image, represents the local scale factor of the pixel at position , represents the exponential function with the natural constant as the base, represents the local window, represents the pixel variance within the local window with the pixel at position as the center, represents the sensitivity coefficient;

[0047] The base information acquisition unit performs Gaussian kernel convolution using the local scale factor to obtain the base information, which is expressed as follows:

[0048] ;

[0049] Among them, represents the base information of the pixel at position . represents pi, u and v respectively represent the abscissa index and ordinate index of the pixel in the local window, represents the pixel value of the original image at the local window position .

[0050] Information reverse unit, subtracting the base information from the original image to obtain the anti-base information containing the bone injury edge and fine texture, which is expressed as follows:

[0051] ;

[0052] Among them, represents the anti-base information at position , represents the pixel value of the original image at position .

[0053] Generate bilateral weight unit, considering both spatial distance and anti-base information difference, calculate the bilateral weight within the local window, which is expressed as follows:

[0054] ;

[0055] Among them, represents the bilateral weight centered on the pixel at position at the local window position , represents the spatial smoothing coefficient, represents the pixel similarity coefficient, represents the anti-base information at the local window position .

[0056] Detail enhancement unit, through local bilateral weighting enhancement, amplify the edge and texture features of the bone injury in the detail layer while retaining the edge information, which is expressed as follows:

[0057] ;

[0058] Among them, represents the pixel value of the image after detail enhancement at position , represents the detail enhancement coefficient, represents the coordinate within the local window, represents taking the absolute value;

[0059] Construct a local structure tensor unit. By calculating the image gradient and performing Gaussian smoothing, local structure information and edge direction features are captured, which is expressed as follows:

[0060] ;

[0061] Among them, represents the local structure tensor, represents a Gaussian smoothing kernel with a standard deviation of , represents the value of the original image at position after taking the partial derivative in the x direction, represents the value of the original image at position after taking the partial derivative in the y direction;

[0062] Generate an adaptive weight unit, which is expressed as follows:

[0063] ;

[0064] Among them, and represent the adaptive weights, represents the detail threshold;

[0065] Feature fusion unit. Using the adaptive weights, the features of the detail layer and the structure layer are fused to comprehensively depict the bone injury details and local structures, which is expressed as follows:

[0066] ;

[0067] Among them, represents the value of the image after feature fusion at position , represents the local structure tensor 's maximum eigenvalue.

[0068] By performing the above operations, aiming at the problems existing in the traditional bone injury assessment scheme, such as unclear medical image features of bone injuries and edge blurring, detail loss, and large noise interference in the traditional image enhancement method for bone injury detection, this scheme improves the contrast and edge clarity of the bone injury area through local scale factor adaptive adjustment, Gaussian kernel convolution to extract base information, inverse base information to enhance details, and multi-scale structure tensor fusion. At the same time, it enhances the visibility of micro-injury features, reduces the interference of noise on injury recognition, and thus effectively improves the accuracy and reliability of the bone injury assessment system.

[0069] Example 3. Refer to Figure 1 and Figure 3 . Based on the above example, the module for establishing a bone injury degree assessment model specifically includes the following units:

[0070] The dynamic convolution unit introduces a depth adjustment matrix to modulate the weights of the convolution kernel, dynamically respond to convolution, and reconstruct local information, which is expressed as follows:

[0071] ;

[0072] Among them, h and w respectively represent the horizontal coordinate index and vertical coordinate index of the feature map, and i and j respectively represent the position indices of the rows and columns of the convolution kernel. represents the depth adjustment matrix of the convolution kernel at the position of the feature map . represents the sigmoid function, c represents the number of channels of the feature map, represents the depth adjustment weight of channel c, represents the eigenvalue of the input feature map at the position and channel c, represents the depth adjustment bias of the convolution kernel at the position . represents the dynamic convolution output of the feature map at the position for channel c, k represents the size of the convolution kernel, represents the dynamic convolution weight of the convolution kernel at the position for channel c, represents the eigenvalue of the input feature map at the position and channel c;

[0073] Define the variational distribution unit, which constructs a variational distribution using the statistical mean and variance of local region features, and is expressed as follows:

[0074] ;

[0075] Among them, represents the mean of the variational distribution, represents the variance of the variational distribution, represents the variational distribution at the position of the feature map , represents the value of variational sampling at the position of the feature map , represents a normal distribution with a mean of and a variance of ;

[0076] The variational feature pooling unit uses the method of variational sampling to perform pooling operations and pool local feature information, which is expressed as follows:

[0077] ;

[0078] Among them, Denotes the variational feature pooling output of the feature map for channel c at position , denotes the variational distribution, denotes the value of variational sampling, denotes the expectation operation;

[0079] The channel attention enhancement unit enhances important channel features by generating channel attention weights, which is expressed as follows:

[0080] ;

[0081] Among them, denotes the importance score of channel c, where H and W respectively denote the maximum index of the abscissa and the maximum index of the ordinate of the feature map, denotes the attention weight of channel c, C denotes the maximum number of channels, and l denotes the index of the channel, denotes the importance score of channel l, Denotes the output of the attention enhancement of the feature map for channel c at position ;

[0082] The feature projection unit projects the feature into the Riemannian manifold space, which is expressed as follows:

[0083] ;

[0084] Among them, denotes the feature projection output of the feature map for channel c at position , denotes the logarithmic function, denotes taking the modulus;

[0085] The evaluation unit evaluates the category of bone injury degree, which is expressed as follows:

[0086] ;

[0087] Among them, denotes the output value of the evaluation unit, denotes the normalized exponential function, and respectively denote the evaluation weight and the evaluation bias, denotes the input value of the evaluation unit;

[0088] The loss calculation unit measures the classification error of the model using the cross-entropy loss, and at the same time introduces a variational distribution regularization term to encourage the model to learn a stable and information-rich feature distribution, which is expressed as follows:

[0089] ;

[0090] Among them, Represents the loss value of the bone injury degree assessment model, n represents the index of the data sample in the model training dataset, M represents the total number of data samples in the model training dataset, o represents the index of the category of bone injury degree, and O represents the total number of categories of bone injury degree. Represents the true label of the nth data sample in the oth category. Represents the probability that the model predicts the nth data sample as the oth category.

[0091] The parameter learning unit calculates the gradient of the loss function with respect to the model parameters and sets the learning rate for parameter update, which is expressed as follows:

[0092] ;

[0093] where t represents the number of times of parameter learning. Represents the model weight and bias parameters at the (t + 1)th parameter learning. Represents the model weight and bias parameters at the tth parameter learning. Represents the model parameter learning rate. Represents the gradient of the model's loss function with respect to the parameters.

[0094] By performing the above operations, aiming at the problems of insufficient feature extraction, local information loss, inability to adaptively adjust channel weights, and limited evaluation accuracy in the traditional bone injury assessment scheme, this scheme enhances local information expression through dynamic convolution, variational distribution modeling, introduces a channel attention mechanism, projects features into the Riemannian manifold space, and incorporates variational loss to optimize model training, improving the accuracy, robustness, and generalization ability of bone injury degree assessment, thus more accurately assisting clinical diagnosis and treatment decisions.

[0095] Example 4, refer to Figure 1 Based on the above example, the bone injury degree assessment module collects the user's bone injury CT image, inputs the image into the bone injury degree assessment model, and the model outputs the user's bone injury degree.

[0096] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0097] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0098] The above description of the present invention and its embodiments is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In general, if those of ordinary skill in the art are inspired by it and, without departing from the purpose of the present invention, design similar structural forms and embodiments to this technical solution without creative efforts, they shall fall within the protection scope of the present invention.

Claims

1. The bone injury degree assessment system based on artificial intelligence is characterized by: It includes a data set construction module, an image enhancement module, a bone injury degree assessment model establishment module and a bone injury degree assessment module; The data set building module collects historical CT images of bone injuries and the degree of bone injuries, annotates the CT images with the degree of bone injuries, and then creates a model training data set, and adds the annotated CT images to the model training data set; The image enhancement module separates the base and high-frequency details through local scale adaptive Gaussian smoothing, generates adaptive weights using bilateral filtering and structure tensors, and fuses the detail layer with the structure layer. Specifically, it includes: determining a local scale unit, obtaining a base information unit, an information inversion unit, generating a bilateral weight unit, a detail enhancement unit, constructing a local structure tensor unit, generating an adaptive weight unit, and a feature fusion unit. The bone injury degree assessment model module is established to classify and assess the degree of bone injury through dynamic convolution and depth adjustment, feature pooling based on variational distribution, channel attention and Riemann manifold feature projection, and specifically includes: a dynamic convolution unit, a variational distribution definition unit, a variational feature pooling unit, a channel attention enhancement unit, a feature projection unit, an assessment unit, a loss calculation unit and a parameter learning unit; The bone injury degree assessment module collects the user's bone injury CT image, inputs the image into the bone injury degree assessment model, and the model outputs the user's bone injury degree.

2. The artificial intelligence-based bone injury assessment system according to claim 1, characterized in that: The image enhancement module specifically includes the following units: Determine the local scale unit, first set the size of the local window, calculate the variance of the pixels in the local window, and finally calculate the local scale unit; Obtain the basis information unit, perform Gaussian kernel convolution using the local scale factor to obtain the basis information; The information inversion unit uses the original image to subtract the base information to obtain the reverse base information including the bone damage edge and subtle texture; Generate bilateral weight units, taking into account both spatial distance and anti-basis information differences, and calculate the bilateral weights within the local window; The detail enhancement unit amplifies the edge and texture features of bone damage in the detail layer through local bilateral weighted enhancement while retaining the edge information; Construct a local structure tensor unit to capture local structure information and edge direction features by calculating image gradients and performing Gaussian smoothing; generating an adaptive weight unit; The feature fusion unit uses adaptive weights to fuse the features of the detail layer and the structure layer to comprehensively characterize the bone injury details and local structures.

3. The artificial intelligence-based bone injury assessment system according to claim 1, characterized in that: The module for establishing a bone injury degree assessment model specifically includes the following units: Dynamic convolution unit, which introduces a depth adjustment matrix to modulate the weight of the convolution kernel, dynamically respond to convolution and reconstruct local information; Define the variational distribution unit and construct the variational distribution using the statistical mean and variance of local area features; The variational feature pooling unit uses the variational sampling method to perform pooling operations and pool local feature information; Channel attention enhancement unit, which enhances important channel features by generating channel attention weights; Feature projection unit, which projects features into the Riemann manifold space; Assessment unit, assessing the degree of bone damage; The loss calculation unit uses cross entropy loss to measure the model classification error and introduces a variational distribution regularization term to encourage the model to learn stable and informative feature distributions. The parameter learning unit calculates the gradient of the loss function with respect to the model parameters and sets the learning rate for parameter update.

4. The artificial intelligence-based bone injury assessment system according to claim 1, characterized in that: The dataset construction module collects historical CT images of bone injuries and the degree of bone injuries, wherein the degree of bone injuries includes mild injury, moderate injury and severe injury, annotates the CT images with the bone injury degree data, and finally creates a model training dataset and adds the annotated CT images to the model training dataset.

5. The artificial intelligence-based bone injury assessment system according to claim 1, characterized in that: The bone injury degree assessment module collects the user's bone injury CT image, inputs the image into the bone injury degree assessment model, and the model outputs the user's bone injury degree.

Images