A method for fracture type recognition and reduction path planning

Abstract

The present invention discloses a method for fracture type recognition and reduction path planning, which relates to the technical field of medical image recognition. The present invention includes the following steps: Step S1: Obtain the medical image data of the patient's fracture area; Step S2: Construct a three-dimensional model of the patient's fracture part based on the medical image data of the fracture area; Step S3: The fracture type recognition model recognizes the fracture contour on the three-dimensional model of the patient's fracture part; Step S4: Stitch the fracture cracks of the fracture part; Step S5: Calculate the offset distance between the bones on both sides of the fracture crack; Step S6: Visualize the reduction path on the three-dimensional model of the patient's fracture part. By obtaining the medical image data of the patient's fracture area to construct a three-dimensional model of the patient's fracture part, the present invention uses the fracture type recognition model to identify the patient's fracture type and fracture contour, calculates the offset distance of the bones, generates a reduction path for visualization display, and assists doctors in performing fracture surgery to improve the accuracy of the three-dimensional model.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image recognition, and particularly relates to a method for fracture type recognition and reduction path planning. Background Art

[0002] Intraoperative navigation technology is an important technology applied in modern medical surgeries. By combining preoperatively obtained images (such as CT, MRI, etc.) with the intraoperative real-time patient anatomical position information, it provides precise three-dimensional spatial guidance for surgeons, thereby improving the accuracy, safety, and efficiency of surgeries. This technology has demonstrated great value especially in the fields of neurosurgery, orthopedics, thoracic surgery, etc. For example, traditional fracture surgeries rely on surgeons' experience and two-dimensional image guidance (such as X-ray films), which have limitations in locating the fracture site, determining the position of implants, and avoiding damage to surrounding important structures. With the progress of computer technology, imaging technology, and robotic technology, intraoperative navigation systems have emerged, providing more precise three-dimensional visualization and positioning means for orthopedic surgeries. In fracture surgeries, intraoperative navigation can help doctors monitor the position of surgical instruments in real time, accurately reduce the fracture ends, optimize the implantation of internal fixation, reduce surgical complications, and shorten the rehabilitation period. Preoperative planning is one of the key steps in the intraoperative navigation process. In this stage, the imaging data of the patient is usually used for three-dimensional reconstruction to create a high-precision virtual model of the patient's anatomical structure. This model can not only help surgeons have an in-depth understanding of the surgical area before the surgery, but also be used to simulate the surgical path, evaluate the surgical risk, design personalized surgical plans, etc., thereby optimizing the surgical strategy and improving the surgical success rate. However, the fracture three-dimensional reconstruction technology still faces many challenges in its implementation process. For example, accurately registering the preoperative three-dimensional reconstruction model with the patient's actual anatomical structure is a major challenge. Factors such as body position changes and tissue deformation during the surgical process may cause deviations between the preoperative plan and the actual situation. Soft tissue influence: The soft tissues (such as muscles, blood vessels, nerves) around the fracture are difficult to clearly display in two-dimensional or three-dimensional images, affecting the comprehensiveness of surgical planning, especially when the protection of these structures needs to be considered. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for fracture type recognition and reduction path planning. By obtaining the medical image data of the patient's fracture area to construct a three-dimensional model of the patient's fracture part, using the fracture type recognition model to identify the patient's fracture type and fracture contour, and calculating the offset distance of the bone to generate a reduction path, the problems that the existing three-dimensional reconstruction model is difficult to accurately register with the actual situation of the patient and the fracture crack area is difficult to clearly display are solved.

[0004] To solve the above technical problems, the present invention is realized through the following technical solutions:

[0005] The present invention is a method for fracture type identification and reduction path planning, including the following steps:

[0006] Step S1: Obtain the medical image data of the patient's fracture area, where the medical image data includes CT images, X-ray images, and MRI images;

[0007] Step S2: Construct a three-dimensional model of the patient's fracture part based on the medical image data of the fracture area;

[0008] Step S3: The fracture type identification model identifies the fracture contour on the three-dimensional model of the patient's fracture part;

[0009] Step S4: Splice the fracture cracks of the fracture part;

[0010] Step S5: Calculate the offset distance between the bones on both sides of the fracture crack;

[0011] Step S6: Visualize the reduction path on the three-dimensional model of the patient's fracture part.

[0012] As a preferred technical solution, in step S1, the obtained medical image data needs to be preprocessed, and the specific preprocessing process is as follows:

[0013] Step S11: Denoise the medical image data through Gaussian filtering; Gaussian filtering can smooth the medical image, reduce the noise of the medical image through blurring, and create a clearer contour basis for edge detection;

[0014] Step S12: Use the Sobel algorithm to calculate the gradient value of each pixel point on the medical image, obtain the edge intensity and direction of the pixel point in the horizontal and vertical directions, and realize the monitoring of the medical image edge. The specific formula for the gradient amplitude is:

[0015] ;

[0016] In the formula, respectively represent the gradient values of the pixel point in the horizontal and vertical directions, is the gradient amplitude;

[0017] The specific formula for the gradient direction is:

[0018] ;

[0019] In the formula, is the arctangent function, is the gradient direction;

[0020] The Sobel algorithm can effectively detect the edge structure of the graphics in each medical image, calculate the gradient values of the medical image in the horizontal and vertical directions. Through the gradient calculations in these two directions, the gradient magnitude and its direction of each pixel can be obtained, thus achieving accurate detection of the edges of the medical image;

[0021] Step S13: Use the non-maximum suppression algorithm to enhance the contour clarity of the medical image; only keep the pixels with the local maximum gradient value in the gradient image and suppress other non-local maximum pixels, making the edge part of the obtained medical image more delicate and sharp, and enhancing the clarity of the edge contour of the medical image;

[0022] Step S14: Use the double-threshold detection algorithm to classify the pixels in the medical image; the classification is divided into strong edges and weak edges according to the size of the gradient threshold; the strong edges and weak edges are judged by a preset threshold. The gradient magnitude exceeding the threshold is recognized as a strong edge, and the gradient magnitude lower than the threshold is recognized as a weak edge, providing an important basis for the accurate extraction of the edge contour;

[0023] Step S15: Connect the strong-edge pixels to form the edge contour of the bone, and connect the weak-edge pixels to form the cracks of the bone fracture;

[0024] Step S16: Classify and label the medical images, compare through the similarity of various medical image features, use the artificial intelligence algorithm for deep learning, establish a fracture type recognition model, and form a fracture database.

[0025] As a preferred technical solution, in the step S16, the specific process of establishing the fracture type recognition model is as follows:

[0026] Step S161: Combine the fracture region category of the medical image with its corresponding sub-block image to form a label and a training sample set;

[0027] Step S162: Introduce the label and the training sample set into the YOLOv5 network for network training;

[0028] Step S163: The YOLOv5 network uses different-sized convolutional kernels to achieve feature fusion at different scales;

[0029] Step S164: Extract features through convolution and pooling, and use three consecutive Inception modules to perform convolution and pooling in parallel respectively;

[0030] Step S165: Integrate the features after multiple convolutions and poolings, and connect them to the output layer in a fully connected manner;

[0031] Step S166: Output the fracture classification result according to the task requirements.

[0032] As a preferred technical solution, in step S162, the trained original data is used to obtain the mean value and the covariance matrix of Gaussian density estimation , and the following Gaussian probability density function is obtained for each category:

[0033] ;

[0034] In the formula, n is the number of training samples, D is the number of dimensions, is the mean vector, is the covariance matrix of the distribution, is the Gaussian probability density function.

[0035] As a preferred technical solution, in step S165, during the training process of the YOLOv5 network, the loss function corresponding to the classifier is as follows:

[0036] ;

[0037] In the formula, , is the activation function, , are the internal weights, inputs and biases of its nodes.

[0038] As a preferred technical solution, in step S2, the specific process of constructing the three-dimensional model of the patient's fracture part is as follows:

[0039] Step S21: The patient needs to undergo thin-slice spiral CT scanning to obtain a high-resolution two-dimensional tomographic image sequence;

[0040] Step S22: The medical images are processed using the preprocessing method in step S1;

[0041] Step S23: The two-dimensional sequence is converted into a three-dimensional data set using three-dimensional reconstruction technology;

[0042] Step S24: The mask editing tool is used for complex regions to manually remove adhesions or irrelevant tissues;

[0043] Step S25: A smoothing algorithm is applied to reduce the three-dimensional surface jaggedness;

[0044] Step S26: Adjust the color rendering and transparency, and export the standard format for 3D printing or surgical planning systems.

[0045] As a preferred technical solution, in step S4, the fracture cracks of the same fracture part are connected. Let the two endpoints of the fracture crack be and , and the slopes of the two endpoints are calculated respectively Use the point - slope form to fit a straight line between two end - points. The specific formula is as follows:

[0046] ;

[0047] Calculate the slope in the formula , and use the point - slope form and loop statements in MATLAB to find the points on the straight line and connect them with a smooth curve.

[0048] As a preferred technical solution, in step S5, the calculation process of the offset distance of the bones on both sides of the fracture crack is as follows:

[0049] Step S51: Select 3 consecutive rectangular areas with the same length and width with the left - most vertex of the bone above the fracture crack as the origin, and name them A, B, and C respectively, and the corresponding area distributions are ;

[0050] Step S52: Select 3 consecutive rectangular areas with the same length and width at the left - most of the bone below the fracture crack, and name them D, E, and F respectively, and the corresponding area distributions are ;

[0051] Step S53: Use the cv2.contour function in OpenCV to select the bone crack and the selected 6 contour areas for calculation, and respectively obtain the areas composed of the three rectangular areas above and the fracture crack , and the areas composed of the three rectangular areas below and the fracture crack ;

[0052] Step S54: According to the length and area data of the fracture crack, judge the offset distance of the bones on both sides of the fracture crack.

[0053] As a preferred technical solution, in step S54, the Spearman correlation coefficient method is used to analyze different medical images after fracture, and relevant image feature data is obtained to determine the offset distance of the bones on both sides of the fracture crack. The specific calculation formula of the Spearman correlation coefficient method is as follows:

[0054] ;

[0055] In the formula, are the ranks of the observed values after sorting at different angles respectively, are the average ranks of the variables respectively, and N is the total amount of data.

[0056] The present invention has the following beneficial effects:

[0057] (1) The present invention constructs a three-dimensional model of the patient's fracture site by acquiring medical image data of the fracture area of the patient, uses a fracture type recognition model to identify the patient's fracture type and fracture contour, and calculates the offset distance of the bone to generate a reduction path to assist the doctor in performing a fracture operation and improve the accuracy of the three-dimensional model.

[0058] (2) The present invention identifies two endpoints of the fracture crack and calculates the slopes of the two endpoints respectively uses the point-slope form to fit a straight line between the two endpoints, and calculates the slope in the formula , uses the point-slope form and loop statements in MATLAB to find the points on the straight line, and connects them with a smooth curve to complete the fracture crack, thereby improving the accuracy of the operation.

[0059] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. Description of the Drawings

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for describing the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0061] Figure 1 is a flowchart of a method for fracture type recognition and reduction path planning according to the present invention;

[0062] Figure 2 is a schematic diagram of the offset distance of the bones on both sides of the fracture crack. Detailed Embodiments

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0064] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0065] To make the purpose, technical solutions and advantages of the present application clearer, the following will further elaborate on the present application Figure 1-2 with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0066] Please refer to Figure 1 As shown, the present invention is a method for fracture type identification and reduction path planning, including the following steps:

[0067] Step S1: Obtain the medical image data of the patient's fracture area, where the medical image data includes CT images, X-ray images, and MRI images;

[0068] Step S2: Construct a three-dimensional model of the patient's fracture part based on the medical image data of the fracture area;

[0069] Step S3: The fracture type recognition model recognizes the fracture contour on the three-dimensional model of the patient's fracture part;

[0070] Step S4: Stitch the fracture cracks of the fracture part;

[0071] Step S5: Calculate the offset distance between the bones on both sides of the fracture crack;

[0072] Step S6: Visualize the reduction path on the three-dimensional model of the patient's fracture part.

[0073] In step S1, the obtained medical image data needs to be preprocessed, and the specific preprocessing process is as follows:

[0074] Step S11: Denoise the medical image data through Gaussian filtering; Gaussian filtering can smooth the medical image, reduce the noise of the medical image through blurring, and create a clearer contour basis for edge detection;

[0075] Step S12: Use the Sobel algorithm to calculate the gradient value of each pixel point on the medical image, obtain the edge intensity and direction of the pixel point in the horizontal and vertical directions, and realize the monitoring of the medical image edge. The specific formula for the gradient amplitude is:

[0076] ;

[0077] In the formula, respectively represent the gradient values of the pixel point in the horizontal and vertical directions, is the gradient amplitude;

[0078] The specific formula for the gradient direction is:

[0079] ;

[0080] In the formula, is the arctangent function, is the gradient direction;

[0081] The Sobel algorithm can effectively detect the edge structure of the graphics in each medical image, calculate the gradient values of the medical image in the horizontal and vertical directions. Through the gradient calculation in these two directions, the gradient magnitude and its direction of each pixel point can be obtained, thereby realizing the accurate detection of the medical image edge;

[0082] Step S13: Use the non-maximum suppression algorithm to enhance the contour clarity of the medical image; only retain the pixel points with the local maximum gradient value in the gradient image, and suppress other non-local maximum pixel points, making the edge part of the obtained medical image finer and sharper, and enhancing the clarity of the edge contour of the medical image;

[0083] Step S14: Use the double-threshold detection algorithm to classify the pixels in the medical image; the classification is divided into strong edges and weak edges according to the size of the gradient threshold; the strong edges and weak edges are judged by a preset threshold. The gradient magnitude exceeding the threshold is recognized as a strong edge, and the gradient magnitude lower than the threshold is recognized as a weak edge, providing an important basis for the accurate extraction of the edge contour; because the gaps between bones have clear anatomical structures and obvious density differences, they usually show relatively high and uniform gradient magnitudes, so the gradient magnitudes higher than the threshold are recognized as the gaps between two bones; while for fracture cracks, due to the irregular fracture surface or accompanied by surrounding tissue edema, the gradient magnitude may show local sudden changes or lower regions, especially at the blurred fracture edges, so the gradient magnitudes lower than the threshold are recognized as the crack gaps formed after bone fractures. The specific judgment of the gradient magnitude is as follows in the table:

[0084]

[0085] In addition to the above judgment steps, the edge contour of the bone or the crack of the bone fracture can also be judged according to the edge morphology;

[0086] The edges of bone gaps are mostly continuous and smooth curves, which conform to the anatomical structure characteristics, and their coherence can be enhanced by improved methods such as shear wave transform. Because fracture cracks often show discontinuous, serrated or bifurcated morphologies, with low gradient direction consistency, it is necessary to analyze the fracture points in combination with the non-maximum suppression technique.

[0087] Step S15: Connect the strong edge pixels to form the edge contour of the bone, and connect the weak edge pixels to form the crack of the bone fracture;

[0088] Step S16: Classify and label the medical images, compare through the similarity of various medical image features, use the artificial intelligence algorithm for deep learning, establish a fracture type recognition model, and form a fracture database.

[0089] The model training in this embodiment is carried out under the Windows 11 Professional Workstation Edition operating system. The processor model is Intel Core i7-10700 @ 2.9 GHz, the graphics card model is Nvidia RTX 4090 24GB, the memory is DDR4 2400 MHz 48GB, the framework is Pytorch 2.1, Python 3.10, CUDA 11.7, Anaconda 23.1.0, MMDetection 3.10, and the IED tool is VS Code. To ensure the controllability of variables, all experiments are run and tested in this environment.

[0090] In step S16, the specific process of establishing the fracture type recognition model is as follows:

[0091] Step S161: The medical image fracture region category and its corresponding sub-block image constitute a label and training sample set;

[0092] Step S162: introducing the labels and training sample set into the YOLOv5 network for network training;

[0093] The YOLOv5 network mainly consists of data input layer, convolution layer, pooling layer, fully connected layer and output layer; the specific functions of each layer are as follows:

[0094] Data input layer: Receives image and other data, usually a multidimensional array, such as a color image is a three-dimensional array (height, width, color channels).

[0095] Convolutional layer: This layer performs a convolution operation on the input data using a convolution kernel to extract local features of the image. The convolution kernel slides across the image, performing a weighted summation on each local region to obtain a feature map.

[0096] Pooling layer: Downsamples the feature maps obtained by convolution, reducing the data dimension and computational complexity while preserving the main features as much as possible. Common pooling methods include maximum pooling and average pooling.

[0097] Fully connected layer: Integrates the features after multiple convolutions and pooling, and connects them to the output layer through a fully connected method to achieve tasks such as image classification or regression.

[0098] Output layer: The output layer outputs classification results or predicted values according to task requirements.

[0099] Step S163: The YOLOv5 network uses convolution kernels of different sizes to achieve fusion of features of different scales;

[0100] Step S164: extract features through convolution and pooling, using three consecutive Inception modules to perform convolution and pooling in parallel;

[0101] Step S165: Integrate the features after multiple convolutions and poolings, and connect them to the output layer through a fully connected manner;

[0102] Step S166: Output the fracture classification result according to the task requirements.

[0103] In step S162, the trained original data is used to obtain the mean and the covariance matrix of Gaussian density estimation , and obtain the following Gaussian probability density function for each category:

[0104] ;

[0105] In the formula, n is the number of training samples, D is the number of dimensions, is the mean vector, is the covariance matrix of the distribution, is the Gaussian probability density function

[0106] The loss value is used to measure the difference between the model prediction value and the true value. The calculation of the loss value is carried out using the cross-entropy loss function. Its purpose is to optimize the parameters of the model by minimizing the loss function so that the model can better fit the training data. The training dataset is divided into two parts. One part is the original dataset, and the other part is the augmented dataset. The original dataset is obtained as the true value of the training set. The original and augmented datasets are combined into the training set and input into the YOLOv5 network model for training. The loss function corresponding to the classifier is as follows:

[0107] ;

[0108] In the formula, , is the activation function, , are the internal weights, inputs and biases of its nodes.

[0109] In step S2, the specific process of constructing the three-dimensional model of the patient's fracture part is as follows:

[0110] Step S21: The patient needs to undergo thin-slice spiral CT scanning (slice thickness ≤ 1 mm) to obtain a high-resolution two-dimensional tomographic image sequence. The bone CT needs to use bone window parameters (window width 2000 HU, window level 500 HU) to enhance the contrast between bone and soft tissue;

[0111] Step S22: Perform image enhancement, noise reduction and contrast adjustment on the original CT data to optimize the clarity of the bone edges. For example, reduce the artifact interference through the algorithm and improve the distinguishability of the boundary between the cortical bone and the cancellous bone;

[0112] Step S23: Convert the 2D sequence into a 3D dataset using 3D reconstruction techniques (e.g., surface rendering, voxel modeling), such as automatically segmenting continuous bone regions using a seed point expansion algorithm;

[0113] Step S24: Use mask editing tools to manually remove adhesions or irrelevant tissues in complex areas. You can also use the Magic Wand tool to quickly remove non-target bones such as the scapula and sternum.

[0114] Step S25: Apply a smoothing algorithm (such as smooth3D+) to reduce 3D surface jaggedness and improve the visual quality of the model.

[0115] Step S26: Adjust color rendering and transparency to enhance anatomical structure recognition, rotate the 3D model from multiple angles to observe, confirm that the bone morphology is consistent with clinical needs, and export it to a standard format for 3D printing or surgical planning systems.

[0116] In step S4, due to the angle of medical image acquisition, it is sometimes inconvenient to perform 360-degree spiral photography of the fracture site. Therefore, when generating the three-dimensional bone model, the fracture cracks are intermittent. Therefore, it is necessary to connect the fracture cracks of the same fracture part after generating the three-dimensional bone model. The two end points of the fracture crack are respectively and , calculate the slope of the two endpoints respectively Use the point-slope method to fit a straight line between two endpoints. The specific formula is as follows:

[0117] ;

[0118] Calculate the slope In MATLAB, the point-slope formula and loop statements are used to find the points on the straight line and connect them with smooth curves to achieve artificial interactive fracture crack connection.

[0119] See also Figure 2 As shown, in step S5, the calculation process of the offset distance of the bones on both sides of the fracture crack is as follows:

[0120] Step S51: Select three continuous rectangular areas with the same length and width at the leftmost vertex of the bone above the fracture as the origin, and name them A, B, and C respectively. The corresponding area distribution is ;

[0121] Step S52: Select three continuous rectangular areas with the same length and width on the left side of the bone below the fracture crack, and name them D, E, and F respectively. The corresponding area distribution is ;

[0122] Step S53: Use the cv2.contour function in OpenCV to select the bone fracture and calculate the areas of the selected 6 contours, and obtain the areas composed of the three upper rectangles and the fracture crack respectively , as well as the areas composed of the three lower rectangles and the fracture crack ;

[0123] Step S54: Determine the offset distance of the bones on both sides of the fracture crack according to the length and area data of the fracture crack.

[0124] In Step S54, the Spearman correlation coefficient method is used to analyze the medical images after different fractures, and relevant image feature data is obtained to determine the offset distance of the bones on both sides of the fracture crack. The specific calculation formula of the Spearman correlation coefficient method is as follows:

[0125] ;

[0126] In the formula, are the ranks of the observed values after sorting at different angles respectively , are the average ranks of the variables respectively, and N is the total amount of data.

[0127] Let the difference between the ranks of the observed values in the two variables be , then:

[0128] .

[0129] It should be noted that in the above system embodiments, the included units are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0130] In addition, those of ordinary skill in the art can understand that all or part of the steps in the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0131] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to the specific implementation manners. Obviously, according to the content of this specification, many modifications and variations can be made. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the present invention, so that those skilled in the art in the relevant technical field can understand and utilize the present invention well. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A method for fracture type recognition and reduction path planning, characterized in that It includes the following steps: Step S1: Obtain the medical image data of the patient's fracture area, where the medical image data includes CT images, X-ray images, and MRI images; Step S2: Construct a three-dimensional model of the patient's fracture part based on the medical image data of the fracture area; Step S3: The fracture type recognition model recognizes the fracture contour on the three-dimensional model of the patient's fracture part; Step S4: Stitch the fracture cracks in the fracture part; Step S5: Calculate the offset distance between the bones on both sides of the fracture crack; Step S6: Visualize the reduction path on the three-dimensional model of the patient's fracture part; In step S5, the calculation process of the offset distance between the bones on both sides of the fracture crack is as follows: Step S51: Select three consecutive rectangular regions with the same length and width, with the leftmost vertex of the bone above the fracture crack as the origin, and name them A, B, and C respectively, and the corresponding area distributions are ; Step S52: Select three consecutive rectangular areas with the same length and width at the leftmost side of the bone below the fracture crack, name them D, E, and F respectively, and the corresponding area distributions are ; Step S53: Use the cv2.contour function in OpenCV to select the bone fracture and calculate the areas of the selected 6 contours, and respectively obtain the areas formed by the three rectangles above and the fracture, as well as the areas formed by the three rectangles below and the fracture. ; Step S54: Determine the offset distance between the bones on both sides of the fracture crack according to the length and area data of the fracture crack; In step S54, the Spearman correlation coefficient method is used to analyze different medical images after fracture, and relevant image feature data is obtained to determine the offset distance between the bones on both sides of the fracture crack. The specific calculation formula of the Spearman correlation coefficient method is as follows: ； wherein, are the observed values sorted at different angles, are the ranks, are the average ranks of the variables, and N is the total amount of data.

2. The method for fracture type recognition and reduction path planning according to claim 1, characterized in that In step S1, the obtained medical image data needs to be preprocessed. The specific preprocessing process is as follows: Step S11: Denoise the medical image data through Gaussian filtering; Step S12: Use the Sobel algorithm to calculate the gradient value of each pixel point on the medical image, and obtain the edge intensity and direction of the pixel point in the horizontal and vertical directions to realize the monitoring of the medical image edge. The specific formula for the gradient amplitude is: ； In the formula, respectively represent the gradient values of the pixel point in the horizontal and vertical directions, is the gradient amplitude; The specific formula for the gradient direction is: ； In the formula, is the arctangent function, is the gradient direction; Step S13: Use the non-maximum suppression algorithm to enhance the contour clarity of the medical image; Step S14: Use the double-threshold detection algorithm to classify the pixels in the medical image; the classification is divided into strong edges and weak edges according to the size of the gradient threshold; Step S15: Connect the strong edge pixels to form the edge contour of the bone, and connect the weak edge pixels to form the fracture crack of the bone; Step S16: Classify and label the medical images, compare them through the similarity of various medical image features, and use the artificial intelligence algorithm for deep learning to establish a fracture type recognition model and form a fracture database.

3. The method for fracture type recognition and reduction path planning according to claim 2, characterized in that, In step S16, the specific process of establishing the fracture type recognition model is as follows: Step S161: Combine the medical image fracture area category with its corresponding sub-block image to form a label and a training sample set; Step S162: Introduce the label and the training sample set into the YOLOv5 network for network training; Step S163: The YOLOv5 network uses different-sized convolutional kernels to achieve feature fusion at different scales; Step S164: Extract features through convolution and pooling, and use three consecutive Inception modules to perform convolution and pooling in parallel respectively; Step S165: Integrate the features after multiple convolutions and poolings, and connect them to the output layer through a fully connected manner; Step S166: Output the fracture classification result according to the task requirements.

4. A method for fracture type recognition and reduction path planning according to claim 3, characterized in that, In the step S162, the trained original data is used to obtain the mean value and the covariance matrix of the Gaussian density estimation , and the following Gaussian probability density function is obtained for each category: ； where n is the number of training samples, D is the number of dimensions, is the mean vector, is the covariance matrix of the distribution, is the Gaussian probability density function.

5. The method for fracture type recognition and reduction path planning according to claim 3, characterized in that In step S165, during the training process of the YOLOv5 network, the loss function corresponding to the classifier is as follows: ； wherein, , is the activation function, , are the internal weights, inputs and biases of its nodes.

6. A method for fracture type recognition and reduction path planning according to claim 1, characterized in that, In step S2, the specific process of constructing the three-dimensional model of the patient's fracture part is as follows: Step S21: The patient needs to undergo thin-slice spiral CT scanning to obtain a high-resolution two-dimensional tomographic image sequence; Step S22: Use the preprocessing method in Step S1 to process the medical images; Step S23: Use 3D reconstruction technology to convert the two-dimensional sequence into a 3D data set; Step S24: Use a mask editing tool for complex regions to manually remove adhesions or irrelevant tissues; Step S25: Apply a smoothing algorithm to reduce 3D surface jagging; Step S26: Adjust color rendering and transparency and export in a standard format for 3D printing or surgical planning systems.

7. A method for fracture type recognition and reduction path planning according to claim 1, characterized in that In Step S3, the fracture type recognition model identifies each bone fragment and other tissues in the medical image, extracts the edge feature points of each bone fragment and compares them with the feature points in the fracture database to obtain the spatial position, boundary, and shape of the fractured bone fragments, and assigns corresponding colors to each bone fragment respectively.

8. A method for fracture type recognition and reduction path planning according to claim 1, characterized in that In the step S4, the fracture cracks of the same fracture part are connected. Let the two end points of the fracture crack be and , respectively calculate the slopes of the two end points Use the point-slope form to fit the straight line between the two end points. The specific formula is as follows: ； Calculate the slope in the formula , use the point-slope form and loop statements in MATLAB to find the points on the line and connect them with a smooth curve.

Images