Fracture type classification method for femoral trochanter

Abstract

The invention provides a femoral trochanter fracture type classification method. The method comprises the following steps: preprocessing an image data set; performing feature extraction on the image data by using an SIFT algorithm; performing feature extraction on the image data by using a DenseNet model; fusing the key point features and the deep features; and classifying the fused features. The method has the beneficial effects that the SIFT algorithm is combined with the DenseNet deep learning model, SIFT focuses on extracting local features in the image and has strong rotation and zoom invariance, and DenseNet captures more complex global features through dense connection characteristics, so that the overall structure and context information of the image are understood; according to the weighted fusion strategy, the model can dynamically adjust the weights of different features according to data characteristics and training stages, the adaptability of the model to different types of fracture features is enhanced, the model can make more accurate classification judgment through reasonable combination of local and global features no matter simple fracture or complex fracture, and the classification accuracy is improved. And the generalization ability of the model is improved.

Description

Technical Field

[0001] The invention belongs to the field of medical image processing, and in particular relates to a method for classifying fracture types of trochanteric femur. Background Art

[0002] Trochanteric fractures are common in the elderly and seriously affect patients' daily lives. Currently, radiologists mainly diagnose and classify fracture types based on X-ray images using the AO / OTA classification criteria (such as 31 / A1, 31 / A2, and 31 / A3). However, manual analysis of imaging data is not only time-consuming and laborious, but may also lead to inconsistent diagnostic results due to differences in physician experience and professional background. Therefore, the development of an accurate and efficient computer-aided diagnosis system is of great clinical significance.

[0003] Existing technologies have certain limitations in the classification of femoral trochanteric fractures. Some methods rely on only a single feature for classification, which cannot fully and accurately characterize the fracture characteristics; some models fail to fully utilize local and global information during feature extraction, resulting in low classification accuracy; in addition, the traditional diagnostic process is inefficient and difficult to meet the needs of clinical rapid diagnosis. Therefore, a femoral trochanteric fracture classification method that solves the above problems is urgently needed. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for classifying femoral trochanter fracture types, which is particularly suitable for accurately classifying femoral trochanter fracture types through a model.

[0005] The technical solution adopted by the present invention is: in a first aspect, a method for classifying the fracture types of the femoral trochanter is provided, comprising:

[0006] Preprocess the image dataset;

[0007] Use SIFT algorithm to extract features from image data;

[0008] Use the DenseNet model to extract features from image data;

[0009] Through equation F final =w 1 F SIFT +w 2 F DenseNet The key point feature F SIFT and deep features F DenseNet Fusion, where F final is the fused feature vector, w 1 and w 2 is the dynamically adjusted weight coefficient;

[0010] Classify the fused features.

[0011] Furthermore, preprocessing the image dataset includes the following steps:

[0012] Get the image data set D = {I i}, where i=1,2,…,N,I i is the corresponding image;

[0013] By equation to I i Perform denoising to obtain the denoised image

[0014] By equation Normalization is performed, where is the normalized image, μ i is the mean of the denoised image, σ i is the standard deviation of the i-th denoised image.

[0015] Furthermore, the feature extraction of the image using the SIFT algorithm includes the following steps:

[0016] Conduct multi-scale spatial extreme point search;

[0017] Identify and screen key points;

[0018] Generate a descriptor.

[0019] Furthermore, the multi-scale spatial extreme point search includes the following steps:

[0020] Perform Gaussian blur processing on the preprocessed image;

[0021] By equation Generate multiple Gaussian filters of different scales, obtain images with different blur levels by gradually increasing σ, and identify multi-scale spatial extreme points by comparing images between adjacent scales, where G(x, y, σ) is a Gaussian function with a scale of σ, and x and y are The coordinate position of the image.

[0022] Furthermore, the identification and screening of key points includes the following steps:

[0023] The difference space method D(x, y, σ) = L(x, y, kσ) - L(x, y, σ) is used to determine whether the multi-scale space extreme point is a key point, where L(x, y, σ) represents the Gaussian blur result after weighted averaging, and k = 2 1 / 2 ;

[0024] Through the Hessian matrix By determining the eigenvalues ​​in the Hessian matrix, unreliable key points that cause positioning errors are screened out.

[0025] Further, generating the descriptor comprises the following steps:

[0026] For each confirmed key point, a small window centered on it is constructed by equation Calculate the gradient direction θ and magnitude M within the window, where G x and G y are the gradients of the pixels in the window in the x and y directions respectively;

[0027] Divide the small window into several sub-regions, create a direction histogram based on the gradient information in each sub-region, and normalize each direction using the equation D'=D / ||D||, where D' represents the descriptor vector after normalization;

[0028] Generate a descriptor D(p,q,k)=(d 11 ,d 12 ,...,d mn ), where p, q, k represent position indices, d ij Represents the feature vector at the corresponding index.

[0029] Furthermore, classifying the fused features includes the following steps:

[0030] Process the fused feature vector through the fully connected layer to obtain the score value z i and z j ;

[0031] By equation Map the eigenvectors, where P(y|F final ) represents the probability of fracture type, z i represents the score value of category i, z j represents the score value of category j, K represents the number of fracture types;

[0032] The probability values ​​of different fracture types are compared to determine the fracture type in the X-ray image.

[0033] In a second aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the femoral trochanter fracture type classification method provided in the present disclosure.

[0034] In a third aspect, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the method for classifying trochanteric fracture types provided by the present disclosure.

[0035] In a fourth aspect, a computer program product is provided, comprising a computer program / instruction, wherein the computer program / instruction, when executed by a processor, performs the method for classifying trochanteric fracture types provided in the present disclosure.

[0036] The advantages and positive effects of the present invention are as follows: due to the adoption of the above technical solution, the SIFT algorithm is combined with the DenseNet deep learning model, SIFT focuses on extracting local features in the image, has strong rotation and scaling invariance, and can effectively capture subtle structural changes in the femoral trochanter area, while DenseNet captures more complex global features through dense connection characteristics, thereby understanding the overall structure and contextual information of the image, and can more comprehensively and accurately characterize the characteristics of femoral trochanter fractures, thereby improving the accuracy of classification; the weighted fusion strategy enables the model to dynamically adjust the weights of different features according to data characteristics and training stages, thereby enhancing the adaptability of the model to different types of fracture features. Whether it is a simple fracture or a complex fracture, the model can make more accurate classification judgments by reasonably combining local and global features, thereby improving the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 FIG. 1 is a flow chart of a method for classifying trochanteric fracture types according to an embodiment of the present invention.

[0038] Figure 2 Schematic diagram of receiver operating characteristic curves of various models according to an embodiment of the present invention DETAILED DESCRIPTION

[0039] The present disclosure is described more fully below with reference to the accompanying drawings, in which exemplary embodiments of the present disclosure are described. The technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0040] like Figure 1 As shown, the present invention provides a method for classifying the fracture types of the femoral trochanter, comprising:

[0041] S100, preprocessing the image data set;

[0042] Collect a certain number of X-ray images of patients with femoral trochanteric fractures to form an image data set. Ensure that the image quality is clear and can accurately reflect the characteristics of the fracture site. Number the images for subsequent processing and analysis. Check the quality of the preprocessed images to ensure that the denoising and normalization effects are good, without obvious distortion or information loss. For some images with unsatisfactory preprocessing effects, they can be adjusted or reprocessed according to the specific situation to ensure the consistency and reliability of the data.

[0043] S200, extracting features from image data using SIFT algorithm;

[0044] The SIFT (Scale Invariant Feature Transform) algorithm is an algorithm model used to extract local invariant features from images, and is particularly suitable for medical image analysis. The SIFT image feature extraction algorithm is integrated into the DenseNet deep convolutional neural network to perform key point feature extraction on the image data. SIFT Extraction, or SIFT image feature extraction algorithm can be used alone to extract key point features of image data. SIFT extract.

[0045] S300, using the DenseNet model to perform deep feature extraction on image data;

[0046] The preprocessed image data is used as the input of the DenseNet model to extract deep features F from the image data. DenseNet Each layer of the DenseNet model takes the output of all previous layers as input, and the output of the lth layer is H l =H l-1 +F(H l-1 ), H l-1 is the output of the previous layer, and F( ) is the overall transformation of the input feature map composed of multiple convolution operations and activation functions, which can be extended to H l =H l-1 +H l-2 +...+H 0 +F(H l-1 ), realize the comprehensive integration from early low-level features to later high-level features, and effectively capture feature information at different levels. This method realizes effective information transfer and reuse, so as to better capture feature information at different levels. Introduce batch normalization and ReLU activation function in each convolution module to improve training dynamics, accelerate convergence, and improve final classification accuracy and robustness.

[0047] S400, through Formula F final =w 1 F SIFT +w 2 F DenseNet The key point feature F SIFTand deep features F DenseNet Fusion, where F final is the fused feature vector, w 1 and w 2 is the dynamically adjusted weight coefficient;

[0048] The weighted fusion method is used to combine the key point features extracted by the SIFT algorithm and the deep features output by the fully connected layer of the DenseNet model according to the dynamically adjusted weight coefficient w. 1 and w 2 The fusion is performed to obtain the final fused feature vector, balancing the contribution of the two features and strengthening the network's comprehensive understanding of local and global information. The weighted fusion strategy enables the model to dynamically adjust the weights of different features according to data characteristics and training stages, enhancing the model's adaptability to different types of fracture features. Whether it is a simple fracture or a complex fracture, the model can make more accurate classification judgments by reasonably combining local and global features, improving the generalization ability of the model.

[0049] S500: classify the fused features.

[0050] The fused features are processed through several fully connected layers to improve the nonlinear expression ability, and the activation function is introduced to enhance the network's ability to learn complex patterns. The fused and optimized features are mapped to different categories, and the final category judgment is determined by comparing the scores to achieve effective processing of multimodal information and provide support for the diagnosis of femoral trochanteric fractures. The SIFT algorithm is combined with the DenseNet deep learning model. SIFT focuses on extracting local features in images, has strong rotation and scaling invariance, and can effectively capture subtle structural changes in the femoral trochanter area. DenseNet captures more complex global features through dense connection characteristics, thereby understanding the overall structure and contextual information of the image, and can more comprehensively and accurately characterize the characteristics of femoral trochanteric fractures, thereby improving the accuracy of classification.

[0051] In one embodiment, preprocessing the image data set includes the following steps:

[0052] Get the image data set D = {I i}, where i=1,2,…,N,I i is the corresponding image;

[0053] By equation to I i Perform denoising to obtain the denoised image

[0054] By equation Normalization is performed, where is the normalized image, μi is the mean of the denoised image, σ i is the standard deviation of the i-th denoised image.

[0055] In one embodiment, using the SIFT algorithm to extract key point features from an image includes the following steps:

[0056] Conduct multi-scale spatial extreme point search;

[0057] Identify and screen key points;

[0058] Generate a descriptor.

[0059] In one embodiment, performing multi-scale spatial extreme point search includes the following steps:

[0060] Perform Gaussian blur processing on the preprocessed image;

[0061] By equation Generate multiple Gaussian filters of different scales, obtain images with different blur levels by gradually increasing σ, and identify multi-scale spatial extreme points by comparing images between adjacent scales, where G(x, y, σ) is a Gaussian function with a scale of σ, and x and y are The coordinate position of the image.

[0062] In one embodiment, the identification and screening of key points includes the following steps:

[0063] The difference space method D(x, y, σ) = L(x, y, kσ) - L(x, y, σ) is used to determine whether the multi-scale space extreme point is a key point, where L(x, y, σ) represents the Gaussian blur result after weighted averaging, and k = 2 1 / 2 ;

[0064] Through the Hessian matrix By determining the eigenvalues ​​in the Hessian matrix, unreliable key points that cause positioning errors are screened out.

[0065] In one embodiment, for each confirmed key point, a small window centered on it is constructed, and the equation Calculate the gradient direction θ and magnitude M within the window, where G x and G y are the gradients of the pixels in the window in the x and y directions respectively;

[0066] Calculate the gradient of the pixels in the window in the x and y directions, G x =I(x+1,y)-I(x-1,y), G y =I(x,y+1)-I(x,y-1), and then we get the gradient amplitude M and direction θ.

[0067] Divide the small window into several sub-regions, create a direction histogram based on the gradient information in each sub-region, and normalize each direction using the equation D'=D / ||D||, where D' represents the descriptor vector after normalization;

[0068] Generate a descriptor D(p,q,k)=(d 11 ,d 12 ,...,d mn ), where p, q, k represent position indices, d ij Represents the feature vector at the corresponding index.

[0069] In one embodiment, classifying the fused features includes the following steps:

[0070] Process the fused feature vector through the fully connected layer to obtain the score value z i and z j ;

[0071] By equation Map the eigenvectors, where P(y|F final ) represents the probability of fracture type, z i represents the score value of category i, z j represents the score value of category j, K represents the number of fracture types;

[0072] The probability values ​​of different fracture types are compared to determine the fracture type in the X-ray image.

[0073] The classification standards for femoral trochanteric fractures include AO / OTA classification, Evans classification, and Jensen modified Evans classification. The AO / OTA classification standard includes three fracture types: 31 / A1, 31 / A2, and 31 / A3. The probability values ​​of different fracture types are compared. The category with the largest probability value is the fracture type in the X-ray image, thereby achieving accurate classification of femoral trochanteric fractures. Record the classification results and compare them with the known real fracture types. Calculate evaluation indicators such as classification accuracy and recall rate to evaluate model performance. Optimize and adjust the model according to the evaluation results. If the accuracy does not meet the requirements, the model can be continuously optimized and the classification performance can be improved by adjusting the weight coefficient, increasing the amount of training data, adjusting the DenseNet model parameters, or improving the feature extraction method until a satisfactory effect is achieved.

[0074] In order to demonstrate the superiority of the model proposed in this application in its ability to classify fracture types, it was compared with other models that can be applied to fracture classification.

[0075] like Figure 2As shown, in one embodiment, the performance of multiple models in classification tasks is evaluated, including ResNet, DenseNet, BEiT, ConViT, ReXNet and the model proposed by the present invention (Ours).

[0076] From the results, the accuracy of the model of the present invention reached 89.54%, which is better than other models.

[0077]

[0078] Specifically, the accuracy of DenseNet and ConViT is 88.24% and 88.83% respectively, while the accuracy of ResNet is 87.58%. In terms of sensitivity, the proposed model also performs well, reaching 89.59%. In comparison, the sensitivity of DenseNet is 89.22%, while that of BEiT is 85.71%. In terms of specificity, the performance of the model of the present invention is also relatively good, reaching 89.65%, which is higher than other models, especially 89.60% of ReXNet. Finally, in terms of F1-score indicator, Ours model leads other models with 75.76%, showing its advantage in balancing precision and recall. In addition, the experiment also compares the specific receiver operating characteristic curves, and the results show that the model of the present invention can achieve an AUC score of 0.913. These results show that the model proposed in the present invention performs well in all indicators and has strong classification ability.

[0079] Based on the embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0080] An electronic device includes at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the femoral trochanter fracture type classification method provided in the present disclosure.

[0081] Electronic devices are intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0082] A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the femoral trochanter fracture type classification method provided by the present disclosure.

[0083] Various embodiments of the present disclosure may be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0084] A computer program product includes a computer program / instruction, and the computer program / instruction is executed by a processor to implement the method for classifying trochanteric fracture types provided by the present disclosure.

[0085] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0086] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0087] The embodiments of the present invention are described in detail above, but the contents described are only preferred embodiments of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for classifying trochanteric fracture types, characterized in that: include: Preprocess the image dataset; Use SIFT algorithm to extract features from image data; Use the DenseNet model to extract features from image data; Through equation F final =w1F SIFT +w2F DenseNet The key point feature F SIFT and deep features F DenseNet Fusion, where F final is the fused feature vector, w1 and w2 are dynamically adjusted weight coefficients; Classify the fused features.

2. The method for classifying trochanteric fracture types according to claim 1, characterized in that: Preprocessing the image dataset includes the following steps: Get the image data set D = {I i }, where i=1,2,…,N,I i is the corresponding image; By equation to I i Perform denoising to obtain the denoised image By equation Normalization is performed, where is the normalized image, μ i is the mean of the denoised image, σ i is the standard deviation of the i-th denoised image.

3. The method for classifying trochanteric fracture types according to claim 1, characterized in that: Using SIFT algorithm to extract features from images includes the following steps: Conduct multi-scale spatial extreme point search; Identify and screen key points; Generate a descriptor.

4. The method for classifying trochanteric fracture types according to claim 3, characterized in that: The multi-scale spatial extreme point search includes the following steps: Perform Gaussian blur processing on the preprocessed image; By equation Generate multiple Gaussian filters of different scales, obtain images with different blur levels by gradually increasing σ, and identify multi-scale spatial extreme points by comparing images between adjacent scales, where G(x, y, σ) is a Gaussian function with a scale of σ, and x and y are The coordinate position of the image.

5. The method for classifying trochanteric fracture types according to claim 3, characterized in that: The identification and screening of key points includes the following steps: The difference space method D(x, y, σ) = L(x, y, kσ) - L(x, y, σ) is used to determine whether the multi-scale space extreme point is a key point, where L(x, y, σ) represents the Gaussian blur result after weighted averaging, and k = 2 1 / 2 ; Through the Hessian matrix By determining the eigenvalues ​​in the Hessian matrix, unreliable key points that cause positioning errors are screened out.

6. The method for classifying trochanteric fracture types according to claim 3, characterized in that: Generating a descriptor consists of the following steps: For each confirmed key point, a small window centered on it is constructed by equation Calculate the gradient direction θ and magnitude M within the window, where G x and G y are the gradients of the pixels in the window in the x and y directions respectively; Divide the small window into several sub-regions, create a direction histogram based on the gradient information in each sub-region, and normalize each direction using the equation D'=D / ||D||, where D' represents the descriptor vector after normalization; Generate a descriptor D(p,q,k)=(d 11 ,d 12 ,...,d mn ), where p, q, k represent position indices, d ij Represents the feature vector at the corresponding index.

7. The method for classifying trochanteric fracture types according to claim 1, characterized in that: Classifying the fused features includes the following steps: Process the fused feature vector through the fully connected layer to obtain the score value z i and z j ; By equation Map the eigenvectors, where P(y|F final ) represents the probability of fracture type, z i represents the score value of category i, z j represents the score value of category j, K represents the number of fracture types; The probability values ​​of different fracture types are compared to determine the fracture type in the X-ray image.

8. An electronic device comprising: at least one processor; as well as a memory communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor. The instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to make a computer execute the method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images